Ninja radar.

On Bank of America’s third-quarter 2026 earnings call (October 14, 2026), CEO Brian Moynihan said “AI has helped us” and that the bank has “about 130 different projects implemented,” each with “a payback,” whose combined benefit he expects to grow from “$250, $300 [million]” this year to “$700 million next year.” Later in the same answer he gave a headcount figure of “down by 3,000, 1.5% year-over-year.” CFO Alastair Borthwick’s prepared remarks gave a different number for the same quarter, saying headcount “was down nearly 3,500 from a year ago as we automate processes and simplify work across the company,” a roughly 500-person gap between the two company speakers that the call did not reconcile. The transcript does not label all 130 projects as AI specifically, and neither executive draws an explicit causal line from the headcount decline to AI; Borthwick’s own word for the driver is “automate.” Both figures are company statements made on the earnings call, not independently measured or audited.

Archive note: web.archive.org is unreachable from this headless session (save request failed); the quotes above were confirmed against the live transcript page this run.

CEO Matt Comyn told Commonwealth Bank’s 2026 AGM (14 October 2026) that the bank’s AI scam-disruption bots, which intervene by call or message when a customer looks like they are being scammed, have held more than 350,000 conversations since the program started in August 2025. The same address says CBA was ranked the number 1 bank in Asia Pacific and number 4 globally for AI maturity in the 2026 Evident AI Index. The figure is company-reported, from the CEO’s own AGM remarks on CBA’s own newsroom page; no independent auditor’s figure for scams actually disrupted, dollars saved, or false-positive rate accompanies it. A separate, older CBA claim (a 20 per cent cut in customer fraud losses in 1H26, from a third-party financial-commentary blog, not from this address) is not reused here because it cannot be tied to this bot program specifically.

Archive note: web.archive.org is unreachable from this headless session (save requests failed); the quote above was confirmed against the live page this run.

Kogan.com, the Australian online retailer, rebuilt its customer-service chatbot on Salesforce Agentforce after a one-day internal hackathon, going from prototype to production in six weeks (itnews.com.au, October 14, 2026, reporting on CTO Goran Stefkovski’s remarks at Dreamforce). The company and Salesforce’s own marketing both quote a headline figure of “67% of conversations” resolved by the AI agent, up from the original deterministic bot’s 6.2%. But the body of Salesforce’s own case study on the deployment defines a stricter “true resolution rate” and gives a different, lower number for the same deployment: “approximately 20%” today, against the same 6.2% starting point. Neither source reconciles the two figures or says what the 67% metric excludes that the “true resolution” metric counts. Kogan also reports customer satisfaction scores seven percentage points higher among users of the bot, and the deployment now covers 10 customer-service use cases, with order-tracking (“where is my order”) queries making up roughly 60% of inbound volume. Both the 67% and the “true resolution” figures are company or vendor-reported, not independently audited.

Archive note: web.archive.org is unreachable from this headless session (save requests timed out); the quotes above were confirmed against the live pages this run.

On Citigroup’s Q3 2026 earnings call (13 October 2026), CEO Jane Fraser said 14,000 of the bank’s engineers now use its agentic coding tool Devin, which she tied directly to a 30% year-over-year cut in technology vendor spend from in-sourcing work the bank used to pay outside vendors for. In the same answer she said AI has sped up Citi’s product launches by “typically… 30% to 40%,” and that AI-powered client-service agents operating in 82 countries have cut response times for clients by 21%. All three figures are Citigroup’s own, from the bank’s own earnings call transcript; none is independently measured, and Fraser did not give a baseline dollar figure for the vendor-spend cut. This is a separate claim from Citigroup’s earlier Stylus adoption figure (87% of employees, 65 million interactions, flashed 2026-09-29 in radar/2026-10-01/02-citigroup-stylus-ai-adoption.md): no overlap in program or figures.

Archive note: web.archive.org is unreachable from this headless session (save requests failed); the quote above was confirmed against the live PDF this run (extracted from the raw PDF stream, since the hosted file renders as unreadable binary to automated fetch tools).

On UnitedHealth Group’s Q3 2026 earnings call (13 October 2026), CEO Steve Hemsley said the company has expanded AI-enabled chart summarization to more than 40% of its employed clinicians, which PYMNTS reports has cut chart review time by 33%. The same report says AI-enabled information gathering and research have reduced appeal cycle times by 26%, which Hemsley said he expects to accelerate further next year. UnitedHealth’s own Q3 earnings-release slides (filed with the SEC the same day) mention AI only in general terms and do not carry these three figures; PYMNTS attributes them to Hemsley’s remarks on the call itself, not to the written release. No independent body has measured these figures, and the underlying denominators (how many clinicians, how many appeals) are not given.

Archive note: web.archive.org is unreachable from this headless session (save requests failed); the quote above was confirmed against the live page this run.

Kroger CEO Greg Foran told staff in an internal memo, dated 12 October 2026 and first reported by Bloomberg, that the grocer will cut about 3,000 corporate roles over the next three years as it expands its use of AI. That figure represents roughly 37% of Kroger’s 8,200 corporate employees; store and distribution center jobs are not affected. The Associated Press (via its wire, 13 October 2026) quotes Foran’s memo saying “Technology is reshaping what’s possible” and that the company needs to “adopt capabilities, reduce duplicative work and look closely at where work is best fit to be done.” Kroger has not identified which roles or departments will go, has not disclosed expected savings, and reporting is explicit that the company has not said AI is directly responsible for the cuts: the memo pairs technology adoption with a broader complaint about excess management layers and overhead. This follows an earlier cut of about 1,000 corporate positions in June 2025.

Archive note: web.archive.org is unreachable from this headless session (save requests failed); both quotes were confirmed against the live pages this run.

Aston Martin Aramco’s Formula One team built an AI system with cloud vendor CoreWeave that transcribes, diarizes and categorizes all 40 race radio channels (22 rival-team channels plus its own) within five seconds of capture. At the 2026 Monaco Grand Prix the system processed over 50 hours of audio and more than 21,000 individual messages across the race weekend. Before the system, a question such as a rival’s tire compound took engineers “several minutes” of manual monitoring to answer; engineers can now get the same answer “in seconds, not minutes.” The model went through 75 iterations trained on more than 3,000 labeled samples of race audio, and ran as a live test at the final two 2025 races before full use in 2026. The processing-speed and scale figures are CoreWeave’s own reporting (PYMNTS notes “CoreWeave’s figures are self-reported”); no independent party has measured the system’s in-race accuracy.

CarMax, the US used-car retailer (255+ stores, roughly 28,000 associates), now routes 100% of inbound store and customer-experience-center calls through AI voice agents built with Sierra, the AI customer-service vendor led by Bret Taylor. The figure is CEO Keith Barr’s own statement on CarMax’s Sept. 29, 2026 earnings call (per a Motley Fool transcript), reported independently by PYMNTS on Oct. 6, 2026. The voice agents went live in May 2026 and currently handle common questions such as store hours and vehicle availability; neither CarMax nor the article gives a resolution or transfer-to-human rate, so “routes” should not be read as “resolves.” Sierra is a commercial vendor on this deployment (it raised $950M in May 2026 at a $15.8B valuation) but the 100% figure itself is CarMax’s own disclosure, not Sierra’s marketing claim.

Hilton’s AI Planner, a generative-AI trip-planning tool launched in beta on hilton.com in March 2026, had held more than 18,000 conversations with guests as of August, Hilton told PYMNTS. The tool is meant to turn a vague travel idea into a destination, hotel, and trip plan while keeping the booking direct on Hilton’s own site rather than an OTA. Hilton frames the rollout as a “test-and-learn” beta still limited to a small share of site traffic, so the 18,000 figure is a usage count, not a conversion, booking, or revenue outcome; no percentage or dollar figure was found for the tool. Single source: only PYMNTS (October 7, 2026) carries the figure, attributed to Hilton’s own statement; no Hilton newsroom release or second outlet was found repeating it.

L’Oréal’s consumer-care advisers across 25 North American brands cut average case-handling time 64% after adopting Salesforce’s Agentforce tools, which summarize cases, draft emails, validate intent, and generate promo codes inside the service console; a typical case now closes in about 2.5 minutes, down from about seven. Separately, L’Oréal is testing AI agents on brand sites such as YSL Beauty and Kiehl’s to answer “where is my order” questions, with a stated goal of resolving 60% of those inquiries without a human; that 60% figure is explicitly a target, not a measured result, and is excluded from the headline figure for that reason. All three figures originate from a Salesforce customer case study, reported by PYMNTS (October 5, 2026); L’Oréal has not stated these figures in its own voice, and no independent (non-vendor) measurement was found. L’Oréal’s own shareholder letter confirms a company-wide AI push (“One L’Oréal, powered by AI,” 56,000+ employees with access to its internal L’OréalGPT model) but does not itself cite the 64% figure.

Truist said its AI Call Summaries tool, which turns client calls into structured notes within seconds, generated more than 4.5 million summaries in Q2 2026 and supported more than 70% of care center inbound client interactions. The bank estimates the tool saved teammates more than 36,000 hours in that quarter, based on an assumed average time saving of approximately 30 seconds per call against an average call length of more than eight minutes. The 70% coverage and the 4.5 million summary count are the bank’s own reported scale figures; the 36,000-hour figure is Truist’s own calculated estimate, not a direct time-and-motion measurement, and no independent outlet has yet corroborated either number. Single source: only Truist’s own September 9, 2026 newsroom release and its verbatim reprints (Pluang, BriefGlance, finviz) were found; no independent outlet has covered the release with added reporting.

Norway’s largest bank, DNB, announced around 6 to 7 October 2026 that it will cut about 400 full-time roles, nearly 4% of its workforce of more than 10,500 staff, primarily in its technology and services division. CEO Kjerstin Braathen said “AI is changing the way we work and how we deliver services to our customers,” and that the bank is “already seeing considerable gains, and are therefore adapting our organisation to a new reality.” The bank’s own statement cites efficiency gains from AI in customer data management, know-your-customer checks, and software coding, specifically describing agentic AI taking on manual work in those areas; reporting does not quantify those gains separately from the headcount figure. The workforce reduction is expected to finish in Q4 2026; DNB’s shares rose about 1% after the announcement.

Archive note: web.archive.org is unreachable from this headless session (save requests failed); all three quotes were confirmed against the live page this run (fetched via the r.jina.ai proxy after a direct fetch returned HTTP 403).

The Insurance Emporium, a York-based UK personal lines insurer, says claims submitted through its digital claims journey are processed about 40% faster than claims submitted through its traditional notification route, after it added AI-powered claims intake. The new tool, built on agentic AI provider Paperbox paired with Claim Technology’s electronic notification of loss (eNOL) system, reads incoming claim emails and attachments and enters the extracted data directly into the claims system, work staff previously did by hand. Insurance Edge reported the figure on 7 October 2026; the 40% carries an unresolved footnote marker in the source article, originates with the insurer itself, and has no independent measurement behind it.

Archive note: web.archive.org is unreachable from this headless session (both WebFetch and a save request failed); the quote above was confirmed live but no Wayback snapshot could be taken this run.

Branch, a US home and auto insurer, replaced its outsourced call center for first-notice-of-loss claim reporting with AI agents and digital intake built by Liberate. Reporting a claim now takes about 7 minutes, down from more than 12 minutes, a 42% cut, and 43% of Branch’s claims now arrive through these AI channels. The 42% figure is Liberate’s own characterization of its client’s result, quoted by PYMNTS (2026-10-02), not an independently audited number or a figure Branch stated in its own voice; the further claim that these claims will cost “roughly 70% less to handle” is explicitly Branch’s expectation, not a measured result, and is excluded from the headline figure for that reason. Single source: no second outlet or Branch-originated statement was found confirming the 7-to-12-minute change.

Fair Isaac Corporation (FICO) disclosed in an 8-K filed 6 October 2026 that it will eliminate approximately 15% of positions company-wide, applying its own reported headcount of 3,811 that is roughly 570 roles, explicitly naming “integrating AI-driven product development” alongside reducing organizational layers and simplifying its operating structure as the reasons. The company expects about $27.0 million in pre-tax severance and related charges in fiscal Q4 2026, with the plan substantially completed by the end of fiscal Q3 2027. FICO has not disclosed which departments or locations are affected, or broken out how much of the cut is attributable to AI specifically versus the layer reduction and restructuring also named in the same filing. Reuters coverage (via The Star Malaysia) independently confirms the 15% figure and restructuring framing the same week.

Archive note: web.archive.org is unreachable from this headless session (save requests failed); both quotes were confirmed against the live SEC filing and the live news page this run.

Deutsche Telekom is replacing its phone-menu (IVR) systems with AI voice agents across its 10 European operations. Chief AI officer Kartik Sheth told Business Insider’s Hugh Langley at the HumanX conference in Amsterdam (2026-09-24) that a newly launched AI use case handles no more than 20% of calls, climbing to 24-25% within a week, then 28%, reaching 40% within about two months; beyond 40%, progress slows sharply (his words: 60-80% is “really hard,” above 80% “really, really hard”). He also said the share of interactions that never reach a human is growing 1-2 percentage points per month company-wide. These are the CAIO’s own self-reported figures from a conference interview, not an independently audited metric, and no vendor or specific use case is named in this reporting. TIN’s corpus already carries T-Mobile US (a Deutsche Telekom subsidiary) on voice-AI call handling, so this is the same corporate family from a different angle (parent-level, European operations) rather than a wholly new deployer.

DoubleVerify announced on October 5, 2026 that its attribution unit DV Rockerbox can now measure ChatGPT Ads as a touchpoint, and featured WeightWatchers as its first named example: the brand’s attributed cost per acquisition on ChatGPT Ads ran 15.3% below its own established benchmark (OpenAI’s framing of the same figure calls it a “blended paid search benchmark,” a discrepancy neither source resolves). The figure is a single-advertiser result supplied by the vendor that sells the measurement tool, OpenAI picked it to feature, and it arrives with no disclosed spend, flight dates or conversion counts, so it cannot be independently checked. WeightWatchers’ own VP of growth is quoted welcoming the measurement capability, but the company has not published the figure itself.

Southwest Airlines has resolved 45% of eligible customer-service inquiries without human intervention since rolling out Salesforce Agentforce, across more than 2 million customer interactions, per a Salesforce customer-story page published 2026-09-08. Salesforce also credits the deployment with a 900% increase in customer satisfaction and projects $6 million in annual savings, a seven-fold return on the investment. Southwest began the phased rollout in November 2025, so these are cumulative figures through the case study’s publication date, not a single-period result. This is a vendor-published case study (Salesforce, tier 3), not independently measured; no second source corroborates the figures, and “customer satisfaction up 900%” is not defined (no baseline metric or scale given). Aviation carries only one other case file in TIN’s corpus (Cathay Pacific, contrail avoidance), so this is a thin beat for TIN, not an over-covered one.

Affirm rolled out a transformer-based underwriting model on 2026-09-17, replacing a tree-based system built over roughly 14 years of lending history. In a live test against a control group, the new model produced 3.4% more completed purchases “at comparable levels of risk,” according to Affirm, reported by PYMNTS. The risk-comparability claim and the control-group design are Affirm’s own characterization, not independently audited; no second source corroborates the figure, and Affirm’s CEO has separately said the change will not put human underwriters “out of business.”

Barclays is expanding its use of Anthropic’s Claude across the bank, per a 2026-10-01 vendor case study: an internal knowledge assistant has over 16,000 users and has handled more than one million searches, and in Global Markets, Claude routes about 120,000 emails a day, which Barclays says reduces manual handling. Barclays also expects Claude Code to reach half its developers by year-end 2026. This is a vendor-published case study (Anthropic writing about its own customer), not independent press, and carries no baseline or error-rate figure to size the “reduced manual handling” claim. Banking already carries a cluster from this week’s Fortune bank-AI-ROI feature (2026-10-01 radar run); this is a different bank and a different mechanism (internal ops tooling, not customer-facing lending or fraud).

A six-month live facial-recognition trial by British Transport Police at busy London rail stations (February to July 2026) cost more than £320,000, consumed almost 100 hours of police officers’ time, and produced one watchlist alert that turned out to be a false match, with zero arrests, according to a Liberty Investigates freedom-of-information request reported by The Guardian (2026-09-29). The trial scanned over 500,000 faces across 18 deployments. It was later extended, with three confirmed alerts reported in the following four months and a planned expansion to the London Underground; this flash covers the initial six-month phase only. A walk-back/failure item: the deployer’s own stated enforcement goal was not met at the measured cost.

On FactSet’s fiscal Q4 2026 earnings call (2026-09-30), CEO Sanoke Viswanathan said the company’s client assist AI agent, rolled out to its full client base that quarter, now resolves more than 30% of all user-initiated requests without a human. This is a self-reported figure from the CEO on an earnings call, relayed through independent financial press (Investing.com’s transcript), not an independently audited measurement; FactSet did not disclose the volume of requests behind the 30% or how it is calculated. No second source corroborates the figure.

At Skift Global Forum (New York, reported 2026-09-28), Wyndham disclosed that a Salesforce Agentforce-built voice agent now answers every call at participating hotels, where roughly 25% of the roughly 4,000 annual calls to a typical property used to go unanswered entirely. With the agent live, only 15% of guests ask to be transferred to a human, and the article reports the AI-handled calls convert to bookings at a higher rate than the prior baseline, though no exact conversion-rate figure was given for that claim. Both the dropped-call baseline and the human-escalation rate are company-disclosed, relayed through independent press (Skift) rather than independently audited; no second source corroborates the figures. Hospitality is one of TIN’s thinnest beats to date.

Bank of America told Fortune (published September 29, 2026) that EricaAssist, a generative AI agent giving customer service reps contextual guidance during calls, cuts average call times by nearly one minute per interaction. The bank separately said an internal version of its Erica assistant, Erica for Employees, is now used by 90% of its workforce. Both figures are company-reported, relayed through Fortune’s interviews with bank executives rather than independently measured or audited; no comparison baseline for “average call time” before EricaAssist is given in the source.

BNY told Fortune (published September 29, 2026) that it ended last year with 160 AI initiatives in production, a count that has grown to nearly 400 across the company today, and that 70% of its employees regularly use AI in their workflows, with more than 50% deemed “power users.” These are company-reported figures relayed through Fortune’s interviews with bank executives; no independent audit of the initiative count or usage rate is cited.

Capital One told Fortune (published September 29, 2026) that its approach to AI-drafted case summaries has led to customer service agents editing just one-fifth of the AI-generated copy, which the company frames as more thorough and consistent summarization. The figure is company-reported and relayed through Fortune’s interviews; no baseline editing rate before the AI tool is given, so the size of the improvement cannot be independently assessed from the source.

Citigroup told Fortune (published September 29, 2026) that 87% of its employees have adopted Stylus, its internal generative AI tool, producing 65 million enterprise-focused AI interactions. The figure is a company-reported adoption and usage-volume statistic relayed through Fortune’s reporting, not an independently measured productivity or cost outcome; no time window for the 65 million interaction count is given in the source.

Wells Fargo told Fortune (published September 29, 2026) that licensed branch bankers see a 31% increase in product sales, meaning new deposit accounts or credit cards opened, when aided by AI tools. The bank separately said referrals to its AI-enabled wealth advisors have doubled, per an executive named Van Beurden quoted in the piece. Both are company-reported figures relayed through Fortune’s interviews, with no stated comparison period or independent audit; it is not stated whether the 31% figure is a controlled comparison or a before/after rollout measure.

CIO.com’s October 1, 2026 survey of IT leaders quotes West Monroe Partners CIO Kevin Rooney saying the consulting firm’s agentic AI deployment for tier-1 IT support (built on Moveworks, now owned by ServiceNow) has driven a 40% reduction in its yearly managed-service-provider costs and an estimated 2,700 hours per year of operational time savings. Moveworks’ own customer story (undated on the page) puts the same deployment’s first-year total at 8,000 tickets resolved and 4,000 more accelerated, deployed in under 90 days, for $1.4 million in cost savings. The 40% and 2,700-hour figures are the CIO’s own reported numbers to an independent outlet; the ticket counts and dollar total are the vendor’s case-study framing and have not been independently verified against West Monroe’s own disclosures.

Concentrix, the customer-experience and business-services outsourcer, told investors on its Q3 fiscal 2026 earnings call (September 29, 2026) that half its revenue now comes from business “generated in the last three years,” the period since it introduced AI into its service delivery at the start of fiscal 2023. CEO Chris Caldwell said this newer revenue is expected to grow about 30% year over year in fiscal 2026 and carries a revenue-retention rate four times higher than legacy business, and that net-new-logo sales involving AI grew 63% quarter over quarter. These are company-reported, self-disclosed figures from the earnings call, not an independent audit; Concentrix is both the deployer of AI in its own operations and the vendor selling AI-enabled services to its clients, so the revenue-mix figure describes Concentrix’s own business transformation rather than a single client’s measured outcome. Overall Q3 revenue of $2.45 billion missed analyst expectations even as the AI-linked segment grew.

Singapore General Hospital has deployed iScreen, an AI tool built with Synapxe and Iota MedTech, as the first head-CT model on Singapore’s national AimSG medical imaging platform. It automatically analyzes CT scans, flags suspected intracranial hemorrhage for urgent review, and produces results in about six minutes, with radiologists verifying every finding before a report is issued. Honest caveat: the source gives no pre-AI baseline turnaround time, so the six-minute figure cannot be read as a measured improvement, only as the system’s own cycle time; it is also a single source, not independently corroborated. Kept because it is a named hospital, named tool, and a disclosed scale figure in a thin, non-US beat.

On IDT’s Q4 FY2026 earnings call (2026-09-28), CEO Shmuel Jonas said roughly 70-plus percent of communications in IDT’s own customer service operations are now handled by AI in some way, citing it as proof of what its net2phone subsidiary’s AI products can do commercially. This is a self-reported, unaudited figure, and the deployer and the vendor are the same corporate family (IDT owns net2phone), which is a material conflict of interest the company did not address on the call; no breakdown of “communications” (chat vs. voice vs. partial-AI-assisted) was given. Single source (the earnings call transcript); no independent press has covered the figure separately. Included as a data point on production-scale internal AI-agent use, not as a client case.

Adarryl Roberts, the Defense Logistics Agency’s chief information officer, said at DLA’s Industry Collider Day on 2026-09-09 that the combat-support agency now runs approximately 185 to 190 automation bots, 90 to 95 percent of them unattended, which he called “digital employees.” DLA’s own figures put the 2025 payoff at an estimated 300,000 hours of work saved, moving things like food, fuel, medical supplies, and clothing. Government Executive, an independent trade outlet, reported the CIO’s remarks and the agency’s hours-saved figure on 2026-09-14; neither figure carries independent verification beyond the agency’s own disclosure. Government and public-sector AI outcomes are one of the thinnest beats in the corpus, and this is a named deployer with a concrete, dated figure.

The FAA’s SMART tool (Strategic Management of Airspace Routing Trajectories), meant to advise air traffic controllers on rerouting around congestion and weather, will debut in a narrower form than early concept discussions suggested: initially limited to the three airports serving Washington, DC, with a 90-day testing period before any wider rollout. Politico reported (via Yahoo News, 2026-09-18, and picked up by AllSides on 2026-09-21) that airline industry officials had spent weeks worried about how extensively the AI would be woven into day-to-day air traffic management, fearing it could prompt more flight cancellations and reschedulings in the name of avoiding bottlenecks; the FAA’s decision to start narrow followed that pushback. The tool creates no new procedures for controllers or airlines; it generates alternative-route suggestions passed through the FAA’s existing systems. This is a walk-back on the tool’s scope, not a deployment failure. No figure on delay reduction or cost was claimed by either side; the concrete fact here is the scope itself (3 airports, 90-day test) shrinking from what had been signaled.

Rich Argentieri, president of NextEra Analytics, said at a Google Cloud virtual media roundtable on 2026-09-02 that NextEra’s Grid Composer platform, built on Google’s Gemini Enterprise Agent Platform and run across Florida Power & Light’s operations on roughly half a trillion data points a day, had produced “over $20 million of savings to our customers” so far this year through optimized dispatch and outage scheduling. NextEra built the first version of the tool in under 12 weeks. This is a company-reported figure disclosed at a vendor-hosted event with no independent audit named in either source; PowerMag and Grid Innovation Daily Brief, two independent trade-press outlets, both reported the same figure the same week. A second company at the same event, Santee Cooper, said a custom model on Google’s WeatherNext should help it avoid $100,000-an-hour spot purchases, but its own reporting notes that tool “is not yet in production,” so it is a projection rather than a deployed outcome and is not included here.

Dutch grocer Albert Heijn (part of Ahold Delhaize) disclosed at the Groceryshop 2026 conference in Las Vegas that Steijn, its in-app AI meal-planning assistant built on Azure AI Foundry with access to more than 20,000 Albert Heijn recipes, now has 1.1 million active users who spend 56% more than customers who do not use it. Honest caveats: this figure comes from one independent trade-press recap of a conference presentation (Grocery Dive), not from an Albert Heijn press release or filing, so it is single-sourced; a 56% spend difference between AI-chef users and non-users is a correlation disclosed by the retailer, not an independently measured causal lift, and heavier grocery shoppers may simply be more likely to try an in-app assistant in the first place. An earlier (June 2025) Microsoft customer story on Steijn cited a different, non-comparable metric (“over one million conversations” and 80,000 store colleagues assisted daily) that is not reused here to avoid conflating conversation counts with the active-user and spend figures above.

CFO Gary Millerchip told analysts on Costco’s Q4 fiscal 2026 earnings call (2026-09-24) that traffic to Costco.com from AI search engines (he named Gemini, Anthropic, and OpenAI) grew triple digits for the second consecutive quarter, and that this AI-search traffic carries the highest conversion rate of any traffic source on the site. Separately, Millerchip said 10% of all Costco.com orders now include a personalized item, with triple-digit growth in sales from personalized initiatives in Q4, and that Costco membership itself is among the top items originating from AI searches. All figures are first-party, from the CFO on the earnings call; “triple digits” is not a precise percentage, and no independent measurement of AI-search traffic or attribution was found. Costco does no paid digital advertising, so management frames this as an organic, curated-assortment effect rather than a paid placement.

DoorDash co-founder Andy Fang told the Groceryshop 2026 conference in Las Vegas (2026-09-23) that the company’s “Ask DoorDash” conversational AI shopping assistant, which turns recipes, photos, or natural-language chat into a grocery cart, now drives 50% higher basket sizes versus non-AI orders and checkouts five times faster. Fang also said customers using AI for grocery orders purchase 60% more unique SKUs than those who do not. These are improvements over the figures DoorDash reported at the assistant’s June 2026 launch (35% larger baskets, three times faster checkout). All figures are first-party, from a DoorDash co-founder at a single industry conference; no independent measurement was found, and DoorDash has not disclosed the underlying order volume or how “AI orders” versus “non-AI orders” are defined.

A joint Cognizant/Cognition press release (2026-09-23) says logistics provider Odyssey Logistics recorded a 37% net cost saving, against the conventional development approach, rebuilding legacy order-intake and shipment-lifecycle forms (originally built in Microsoft Access and Visual Basic for Applications) as cloud-native applications on its OdysseyONE platform, using Cognition’s Devin AI coding agent. The release also reports delivery throughput rose by roughly one-third and a separate legacy application project closed 21 days after kickoff. Odyssey’s CIO and CDO Maneet Singh is quoted by name endorsing the outcome (“this was about moving faster without compromising discipline”), which is stronger corroboration than a vendor-only case study, but the release is jointly issued by the two vendors (Cognizant, an IT services firm, and Cognition, the AI agent maker) rather than by Odyssey itself or independent press, and no independent figure was found.

Beacon Health System (South Bend, Indiana) has saved more than $95 million since implementing utilization-management vendor Xsolis’ AI-driven medical necessity platform in 2019, and the tool has helped its team support 140% more patients a day, per Becker’s Hospital Review (2026-09-11). The article explicitly attributes the savings figure to the vendor, Xsolis, not to Beacon itself or an independent auditor; this is a six-year cumulative figure, not a recent-period result, and no independent measurement was found.

Boston Children’s Hospital says it has reclaimed about 60,000 hours of staff time, worth more than $7 million in redeployed labor, across more than 50 AI-enabled automations since its enterprise-level partnership with OpenAI. Becker’s Hospital Review (2026-09-11) reports more than a third of employees now use AI daily, spanning supply chain invoices, OR scheduling, and clinical decision support; the hospital also reports more than 40 rare-disease diagnoses previously thought impossible. Both the hospital-stated figure and OpenAI’s own case-study page (a vendor page, not independent) agree on the same numbers; no independently audited figure was found, and the dollar value is the hospital’s own conversion of hours to redeployed-labor cost.

CommonSpirit Health CIO Daniel Barchi told Becker’s Hospital Review (2026-09-11) the Chicago-based system is “probably generating north of $100 million in annual savings using AI and RPA tools” in fiscal 2025, across 242 live applications, up from about 160 two years earlier. The article attributes measurable reductions in sepsis mortality, imaging scan times cut by as much as 50%, and $10 million generated by its Insightli AI assistant alone. Barchi also said governance came before scale: CommonSpirit has rejected 15 proposed AI use cases through its review process. The figure is CIO-stated, not independently audited, and Becker’s is the only outlet carrying this specific quote; an earlier Becker’s piece (2026-02-05) reported a consistent but not identical $100 million figure for 2025 tied to 242 deployments, same publication.

Community Health Network (Indianapolis, 200 sites) is using AI from vendor Notable to schedule wellness visits, scrub charts, deflect inbound calls, and re-engage lost patients, generating $6 million in added revenue, according to Becker’s Hospital Review (2026-09-11). The article explicitly attributes the $6 million figure to the vendor, Notable, not to Community Health Network or an independent source. Separately, the system has set, not yet met, a $10 million AI cost-reduction target for 2025.

Mount Sinai Health System chief digital transformation officer Robbie Freeman told Becker’s Hospital Review (2026-09-11) the system is “projecting a $50 million bottom-line impact from its AI portfolio this year, with more than a 3-to-1 return on investment.” An internally developed malnutrition-detection tool alone has generated roughly $20 million in revenue impact through earlier intervention and documentation. This is an explicit projection, not a closed-year audited result, and the same roughly $50 million figure was already circulating in a May 2026 secondary write-up, so the underlying estimate is not brand new even though this is a fresher restatement with the 3-to-1 ROI detail attached.

OhioHealth (Columbus) says it lowered costs by nearly $550,000 in the first month of using an AI-powered discharge-planning tool from vendor Qventus, reducing excess inpatient days by nearly 1,400 in that same month, per Becker’s Hospital Review (2026-09-11). The $550,000 figure is a single month’s realized result; the $6.6 million annualized figure the article also cites is explicitly Qventus’ own forecast, not a result OhioHealth has achieved yet, and the two should not be conflated.

Providence (Renton, Washington) freed up $33.6 million in observation length-of-stay costs in the first eight months after deploying Xsolis’ AI and predictive-analytics platform, cutting observation length of stay by 32%, per Becker’s Hospital Review (2026-09-11). Providence has since expanded the tool to 28 additional hospitals across six states. The article does not specify the eight-month window’s exact dates or state whether the figure is Providence’s own accounting or Xsolis’, unlike the Beacon Health System item in this same article, which explicitly credits the vendor.

Sysco announced a three-year technology and AI program on 2026-09-09 targeting $500 million in cumulative cost savings, with $100 million projected for fiscal 2027 alone, per its investor-relations release and CIO Dive’s coverage. CEO Kevin Hourican said the company can “improve service to our customers, do our work more efficiently and reduce our structural operating expenses by leveraging best-in-class technology.” The program spans supply-chain productivity, back-office simplification, routing-software modernization, inventory forecasting, and automation in merchandising and procurement; it is not AI-exclusive, and every figure here is a forward-looking target from the company itself, not yet a realized outcome.

Tampa General Hospital has accumulated $40 million in systemwide reduced inefficiencies since launching its CareComm AI command center, built on GE HealthCare’s software, in August 2019, according to Becker’s Hospital Review (2026-09-11). The hospital credits the platform with helping it operate at maximum capacity, decreasing average length of stay, and cutting emergency-room diversion by 25% for its Level 1 trauma center. This is a cumulative figure spanning more than six years, not a recent-period number, and the article does not state whether the $40 million estimate came from the hospital or from GE HealthCare.

UNC Health told Becker’s Hospital Review (2026-09-11) that ambient AI scribing has saved providers an estimated $6 million by cutting charting time on nights and weekends, one of four AI applications it says are running in parallel: AI-driven infusion scheduling (about $5 million in added revenue), automated prior authorization ($3 million to $4 million in savings and revenue), and AI-enhanced nursing leadership communications (an estimated $5.4 million in improved staff retention). CIO Brent Lamm said the board cares about “patient outcomes, access, provider engagement, cost savings and revenue growth.” All figures are health-system estimates carried by a single trade-press source; no independent audit was found.

FedEx Chief Digital Information Officer Vishal Sayal told Citi’s 2026 Global TMT Conference on September 8, 2026 that applying AI to aircraft maintenance research cut the research time from 30 minutes to 3 minutes, which he characterized as “about a 90% efficiency” gain. In the same session he said AI-driven network precision narrowed FedEx’s predictable delivery-time window from 4 hours to 2 hours, a change he called “50% better than where we were a couple of months ago.” Both figures are company-stated, from a single investor-conference transcript (Investing.com’s republished record of the session); no independently measured or audited figure was found, and no baseline methodology for “predictable delivery time” is given.

Autism Queensland, an Australian disability-services nonprofit, says an AI employee-service agent named Paige, built on Salesforce Agentforce, resolves 70 percent of staff administrative requests across Slack, internal portals, and other workplace tools. The figure comes from Salesforce’s September 11, 2026 announcement of its expanded Agentforce agent portfolio; no independent measurement or Autism Queensland-side statement was found, and Salesforce does not define “administrative requests.” This is a non-US, non-corporate deployment, and no earlier public figure for it was found.

The Defense Logistics Agency, which supplies the US military’s roughly 25,000-person supply chain workforce, says unattended software bots saved an estimated 300,000 hours of work in 2025. DLA chief information officer Adarryl Roberts told Government Executive the agency now runs approximately 185 to 190 bots, 90 to 95 percent of them fully unattended, and is moving from scripted bots toward agentic systems that work “across the department.” The 300,000-hour figure is DLA’s own estimate, not independently audited; Government Executive reported it September 14, 2026, citing DLA’s own news article as the source of the hours figure.

Sporting goods retailer Hibbett says an AI shopper agent, built on Salesforce Agentforce, now handles 90 percent of core shopper journeys, and that it went live six weeks after the company started building it. The figure comes from Salesforce’s own September 11, 2026 announcement of its expanded Agentforce agent portfolio, naming Hibbett as one of several customer proof points; no independent measurement or Hibbett-side statement was found. “Core shopper journeys” is not defined in the source, and no dedicated Hibbett case study existed before this announcement, so this appears to be the first public figure for this deployment.

Perk, the AI-native travel and spend-management platform formerly branded TravelPerk, says an autonomous outbound sales agent tripled its sales reps’ call volume and now builds 60 percent of its North American sales pipeline. Both figures are from Salesforce’s own customer story and its September 11, 2026 Agentforce portfolio announcement; the two sources disagree on the agent’s name and status: the customer story calls it “Prospecting Agent” and describes it as already deployed with results, while Salesforce’s own announcement names the equivalent product “Hunter” and describes it as in pilot ahead of a November 2026 general-availability date. Perk’s Chief of Staff for North America, Kaitlyn Hrynewich, is quoted by name in the customer story; no independent measurement was found.

Mahindra Finance, the Mahindra Group’s non-banking financial company, announced at Global Fintech Fest 2026 that voice AI agents built with Sarvam on its platform have completed over one crore (10 million) calls across 12 Indian languages. The partnership now spans three areas of the business: sales, collections, and frontline employee engagement, automating pre-due payment reminders, debt collection workflows, and cross-selling for vehicle, SME, and housing loans in rural and semi-urban markets. Honest caveat: both sources trace to the same company/vendor announcement (Sarvam’s own case study and a reprinted press release); no independent productivity, cost, or conversion figures were disclosed alongside the call-volume figure.

A 2026-09-17 blog post from LEGALFLY, an AI contract-review vendor, claims that customer MSIG Insurance Europe (a Cologne-based insurer) cut its contract review time from a full day to about 20 minutes using the vendor’s platform. The figure is vendor-originated marketing copy in a software-comparison listicle, not a dedicated case study: there is no named MSIG executive quote, no date for when MSIG adopted the tool, no volume (how many contracts, what type), and no independent press or MSIG statement corroborating the number. Flagging as a lead only; the claim itself is not independently verifiable from what is published.

Related case files

None. No MSIG or LEGALFLY case file exists in the corpus.

Siemens and Procter & Gamble announced on 2026-09-16 (dateline Erlangen) that they are scaling their jointly built Visual Inspection Cockpit, an edge-based AI system that inspects consumer products in real time on the production line, across P&G’s global manufacturing operations. The companies say the system reduces scrap rates by 10 to 20 percent depending on the product line, and that new line installations are commissioned five to ten times faster than traditional bespoke optical vision systems. Both figures come from the companies’ own joint press release, syndicated verbatim across roughly ten trade outlets (industryemea, webwire, automation-mag and others) with no independent outlet reporting an original figure; no baseline scrap-rate or per-line volume is given.

Related case files

None yet; no Siemens or Procter & Gamble case file exists in the corpus.

The Adecco Group, the global staffing and recruitment firm, announced on September 15, 2026 (the opening day of Salesforce’s Dreamforce conference) that it is expanding its Salesforce Agentforce deployment from a 10-country recruitment-workflow rollout, already covering business lines worth 50% of Adecco’s revenue, to an “Agentforce Coworker” rollout reaching 27,000 employees across more than 40 countries. The release states the agents draw on context from more than 2.5 million agent-candidate interactions logged since April 2025. CEO Denis Machuel is quoted saying the tool “frees up our sales teams’ and recruiters’” time, but the release gives no productivity percentage or dollar figure for the new Coworker rollout itself (an earlier, separate March 2026 Adecco release had cited “15 percent time savings” from a UK recruitment pilot, a different, narrower deployment).

Caveat: first-party company press release (tier 3), issued jointly with Salesforce as vendor; no independent measurement of the claimed revenue share or interaction count found.

A study in the September 14, 2026 issue of the Journal of Hospital Medicine examined Abridge’s ambient AI scribe across 20 hospitals at Sutter Health, a large northern and central California health system that was the first in the US to fully integrate Abridge within Epic. Only 59 physicians participated, half of them hospitalists, and usage was partial: hospital physicians used the scribe for less than half their patient notes, and just 53% reported a “high degree of success.” A meaningful share of inpatient physicians did not use it to create even a single note, and physician belief that AI would free up capacity for an extra patient per shift fell from 55% to 45% after the trial. The rollout also could not initially be used by inpatient providers for emergency department consultations. Honest caveat: this is a walk-back-shaped result, not a failure narrative; the technology worked, adoption and perceived value did not clearly follow.

The US Transportation Security Administration deployed a Salesforce Agentforce-built agent named Ace, which since going live during the summer 2026 travel surge has handled roughly 100,000 routine traveler conversations a month, resolving 96% of them (things like liquid-packing rules or traveling with a medical device) without escalation to a human. The release also states TSA projects 70,000 automated transactions a year going forward, freeing an estimated 11,660-plus staff hours, and estimates $11 million in savings over three years; Ace is also said to have cut the cost of a consumer interaction by more than 90%. TSA had reportedly needed the equivalent of 12 full-time staff to cover the excess workload the agent now absorbs.

Independent trade press (MeriTalk, CDO Magazine, Unite.AI) restated the same 100,000/96% figures the same day, all crediting the joint TSA/Salesforce announcement; none appear to have independently verified the underlying transaction logs.

The VA first deployed Knowtex’s ambient AI scribe in October 2025 at 10 medical centers. Nextgov/FCW reports the tool has since reached 79 VA medical centers across 10 Veterans Integrated Service Networks, more than 7,000 clinician users, and over 986,000 primary care appointments, saving a reported 450,000 hours of EHR documentation time, with an 88% user adoption rate and a 4.48-of-5 average satisfaction rating across 5,100+ ratings. A three-month Kansas City pilot separately reported 95.8% veteran satisfaction. Caveat: these are the vendor’s and VA’s own reported figures relayed by one trade-press outlet; single-sourced, no independent audit of the hours-saved methodology found.

Related

None yet.

On outsourcer IBEX’s Q4 fiscal 2026 earnings call (2026-09-10), CEO Robert Dechant said BJ’s Wholesale Club launched an AI agent (built with Sierra AI) in June 2026 and is “attaining resolution rates above 40% and CSAT scores above 4.7 out of 5.0,” results he said exceeded what BJ’s previous BPO vendor’s human agents delivered, leading BJ’s to plan adding more human agents in H1 fiscal 2027 alongside the AI. This figure is vendor-stated: IBEX, the BPO running the deployment for its client, disclosed it on its own earnings call, not BJ’s Wholesale itself, and no independent measurement of the resolution rate or CSAT score was cited. Single-sourced to the one earnings-call transcript found this run.

BMO CEO Darryl White said last month that predictive AI modelling in the bank’s insurance business, the SmartDecision platform, can deliver underwriting decisions in as little as 10 seconds against an industry standard of at least 28 days, as reported in a CBC News piece published 2026-09-10. BMO’s own July 9 launch announcement for SmartDecision confirms the underlying product (real-time, point-of-sale decisions on eligible term, universal and whole life applications up to $5 million, cycle times cut from weeks to seconds) but does not itself state an exact seconds figure or the 28-day baseline; those two specific numbers are sourced only to White’s restated claim as quoted by CBC. Honest caveat: single-sourced figures from the deploying company’s own CEO and press materials, no independent measurement found, and the precise “10 seconds” number could not be corroborated against BMO’s own primary announcement.

Related case files

None yet; no BMO case file exists in the corpus.

On the same IBEX Q4 fiscal 2026 earnings call (2026-09-10), CEO Robert Dechant said Philippine Airlines launched an AI customer-service agent, with Sierra AI, handling English, Tagalog and Taglish, and that it “achieved resolution rates above 20% and delivered CSAT above 4.7 out of 5.0,” a proof-of-concept-stage result Dechant said matched traditional human-agent performance. As with the BJ’s Wholesale figure from the same call, this is vendor-stated (IBEX, the BPO delivering the deployment, not Philippine Airlines itself) and not independently measured; the 20%+ resolution figure is also a wider, less precise band than BJ’s 40%+ figure, consistent with an earlier-stage deployment. Single-sourced to the one earnings-call transcript found this run.

Scotiabank CEO Scott Thomson told the bank’s own Financials Summit in Toronto this week that AI has saved the bank roughly 24,000 days’ worth of work over about four and a half months, part of what he called “a transformational opportunity both from an effectiveness, and an efficiency perspective.” Honest caveat: the figure is a CEO estimate against an unstated counterfactual, given in a company-hosted venue with no breakdown of which AI tools or functions produced it, and no independent measurement was found; CBC reports the bank has not seen matching headcount reductions so far.

Related case files

None yet; no Scotiabank case file exists in the corpus. A TD Bank mortgage pre-adjudication story (15 hours to 3 minutes) is already in the research pipeline; this is a distinct Canadian bank and a distinct claim.

Duke Energy says self-healing grid technology, which automatically detects outages and reroutes electricity, helped New Hanover County, North Carolina customers avoid about 19,000 outages and saved an estimated 31,000 hours of total outage time so far in 2026, with about a quarter of the summer benefit coming during major storms. Honest caveat: every figure comes from the company itself (no independent verification found), the article names no Duke Energy official, and the “about 19,000” and “31,000 hours” figures are company estimates against a counterfactual (what would have happened without the technology) rather than a directly observed count.

Related case files

None yet; no Duke Energy case file exists in the corpus.

Uniper, the German-Finnish energy company, put an HR recruiting agent built on Celonis’ Context Model inside Microsoft Copilot Studio into full production, after piloting it as a natural-language tool for hiring managers to query open requisitions and shortlisted candidates. Diginomica (2026-09-09) reports the deployment “reduced” average time-to-sign for candidates “by 27 days” and cut the average screening phase “by 13 days”; cfotech’s coverage the same day carries the same two figures. The agent also flags overdue recruiting tasks and can draft outreach emails to candidates and colleagues. Both figures are presented as results from Uniper’s own live production use, but neither article attributes them to an independent audit; they read as company- and vendor-stated (Uniper, Celonis, Microsoft), not third-party measured. The underlying partnership (Celonis, Microsoft, Uniper) was first announced in May 2025; this is fresh reporting on outcomes from having since gone live.

AOK PLUS, a German statutory health insurer serving more than 3.4 million members in Saxony and Thuringia, went live on NiCE’s unified CXone and Cognigy CX AI platform on September 8, 2026, running in NiCE’s EU Sovereign Cloud. The deployment covers more than 5 million annual member interactions handled by 2,400 employees across 120 specialist skills, with a reported call acceptance rate above 95%. AOK PLUS says it is the first insurer in Saxony and Thuringia to deploy AI-based voice automation in a sovereign cloud environment. Honest caveat: this is a “went live” and capacity announcement, not a before/after outcome measurement; the 95% acceptance rate and interaction volume are the vendor and insurer’s own disclosed figures, with no independent audit found.

AOK PLUS, a statutory German health insurer covering Saxony and Thuringia, went live on NiCE’s unified CXone CX AI platform, connecting AI-powered self-service with 2,400 employees and 120 skill-based routing groups. The deployment supports more than 5 million annual member interactions, has achieved a call acceptance rate above 95%, and migrated more than 1,400 telephone numbers with zero downtime. AOK PLUS says it is among the first public health insurers in Germany to move member service operations to the cloud, and the first insurer in Saxony and Thuringia to run AI-powered voice automation in a sovereign cloud environment.

Caveat: this is a first-party vendor press release (NiCE, tier 3), co-quoted with an AOK PLUS project lead. No independent outlet has re-measured the 95% acceptance-rate figure or the interaction volume; treated as single-source.

CMS’s WISeR pilot (Wasteful and Inappropriate Service Reduction) uses AI vendors to review Medicare prior authorization requests in six states (New Jersey, Ohio, Oklahoma, Texas, Arizona, Washington), launched January 2026. Records released to the Electronic Frontier Foundation via a FOIA lawsuit show one vendor, Virtix, denied more requests than it approved, and two vendors combined denied 5,944 requests in the program’s first three months. One status report cites a request left unanswered for 83 days. A separate STAT News report (paywalled beyond its lede) states a WISeR vendor itself warned CMS before launch that a working product by the target date was unrealistic. Caveat: the released figures come from CMS’s own vendor status reports via FOIA, not an independent audit of denial correctness; EFF and Medicare Rights Center are advocacy-affiliated reporters of the documents, not neutral auditors, though the underlying documents are primary CMS/vendor records.

Related

None yet; this is a fresh walk-back-shaped candidate for a research/ case file (vendor claims vs. delivered reliability), not yet graduated.

FedEx Chief Digital Information Officer Vishal Sayal told Citi’s 2026 Global TMT Conference on 2026-09-08 that an AI tool cut the research time technicians spend on aircraft maintenance from 30 minutes to 3 minutes per case, which he called “about a 90% efficiency” gain. He separately said FedEx has narrowed its quoted delivery window from 4 hours to 2 hours, citing this as a differentiator, though he did not say this second figure is AI-driven specifically or give a volume/scope for either change. Both figures are self-reported by the executive on a single investor-conference transcript (investing.com), with no second outlet or FedEx press release independently corroborating either number; no baseline volume (how many maintenance cases, what fraction of the fleet or delivery network) is given.

Related case files

None yet. An existing case file, fedex-mobius-predictive-maintenance-prevents-17000-downtime-hours-18000-assets, covers a different FedEx AI system (Mobius predictive maintenance, downtime hours prevented across 18,000 assets); this flash is a distinct claim (aircraft maintenance research time, delivery-window precision) from the same company.

XPeng commissioned its own humanoid robot manufacturing facility in Guangzhou on 2026-09-08, with its IRON humanoid walking off the line unaided; the company says the line runs with “more than 80% of core processes automated,” built on the quality systems XPeng already uses for its electric vehicles, with over 85% of IRON’s supply chain shared with its EV business. Honest caveat: the 80% automation figure and the supply-chain overlap figure are both company claims with no independent audit found; mass production is targeted for year-end 2026 with external deliveries in 2027, so this is a commissioning milestone, not yet a sustained production record.

Related case files

None yet; no XPeng case file exists in the corpus.

Google and Cathay Pacific announced on 2026-09-07 an expanded Asia-Pacific trial in which Google’s AI analyzes satellite imagery to route ultra-long-haul flights around contrail-forming airspace with small altitude adjustments. Of more than 100 targeted flights, more than 80 followed the AI-recommended contrail-avoidance routes, and Google estimates those flights cut the warming impact of their contrails by roughly 40%; one source adds that the Hong Kong-Singapore corridor alone delivered more than half of the trial’s total climate benefit. The 40% figure is Google’s own satellite-based estimate of warming impact, not an independently audited measurement, and the announcement is partly vendor-authored (Google’s own blog); this is a flight-operations deployment rather than a marketing/finance one, included here because it is a named airline’s operational AI with a disclosed outcome figure. Published 25 days ago; kept because the scale (80+ flights, two independent outlets both figures confirmed live) is more significant than the typical pilot announcement.

Cashfree Payments moved its Relay “AI super agent” from merchant beta to general availability for small and medium businesses, automating payment reconciliation, failed-payment retries, cart-abandonment follow-up, and dispute filing. Cashfree’s own blog states merchants have used Relay in beta since May 2026, and that it cuts the roughly 60 hours a week an SMB spends on payment operations down to 45 minutes. This is a company-disclosed, vendor-originated figure about its own product’s effect on customers, not an independently audited measurement, and no merchant count or transaction volume was disclosed alongside it.

Incore Bank, a Swiss B2B transaction bank, completed a proof of concept with Kyndryl and Google Cloud applying agentic AI (Kyndryl’s Agentic AI Framework on Google’s Gemini models) to know-your-customer document checks, announced September 2, 2026. The system hit up to 99% accuracy extracting data from onboarding documents and, per fintechnews.ch, “showed potential to cut onboarding times from months to days.” The honest caveat, on the record from PYMNTS: the 99% figure measures how well the AI read the paperwork, not whether it made the right call on a customer’s risk; final judgment still sits with a compliance officer, and this is a completed proof of concept, not yet a production rollout with a measured before/after onboarding time.

Walmart CEO John Furner told analysts on the company’s Q2 FY27 earnings call (reported August 20, 2026) that customers using its AI shopping assistant Sparky, launched June 2025, spend 40% more per order than those who don’t, with the number of users up 70% year over year. This is an earnings-call disclosure, not an independently audited study: Walmart did not disclose a holdout experiment isolating Sparky’s causal effect, so the 40% figure shows correlation between Sparky use and order size, not proof Sparky caused the extra spending. Multiple outlets (PYMNTS, Benzinga, and others) reported the same on-the-record figures the same week; all trace to the single earnings-call statement.

Smarsh, a data-archiving and compliance software vendor, published results from its own two customer-facing Salesforce Agentforce deployments: a customer support agent named Archie and an internal agent named Emmy. Archie achieved a 72% deflection rate and a 2.6-out-of-3 resolution confidence score, resolving more than 7 in 10 of 405 customer sessions in Q2 2026 without human support. Separately, Emmy reached 65% adoption among support representatives by the end of June (120 of 185 reps), resolving 31% of cases with AI since launch and saving reps an average of 7.5 hours per case compared with the same reps’ pre-AI handling time.

Caveat: this is a first-party company press release (tier 3), co-quoted with Salesforce’s own EVP of Agentforce, announcing Smarsh’s decision to scale its Agentforce usage. No independent outlet has re-measured these figures; other outlets that covered it (Financial IT, Insurance Edge, Insurance-Canada) reproduce the same press release rather than independently reporting it, so this is treated as single-source.

Zurich UK says its underwriting workbench for commercial insurance has cut processing times 78% in under two years, attributed on the record to Asha Kumar, the unit’s chief technology officer, who frames it as evidence that insurers are being pushed to show measurable business outcomes rather than just operational activity. Insurance Post’s reporting names the specific business line (commercial insurance) and the exact figure but does not name the underlying software vendor or give a precise start date beyond “less than two years.” This is trade-press reporting with a named executive on the record, not a Zurich press release; no second outlet has independently corroborated the 78% figure yet, so it is flagged single-source.

Related case files

Manulife’s MAUDE: AI underwriting that auto-approved 58% of eligible life-insurance cases is the corpus’s other AI-underwriting case file; both concern insurers automating underwriting decisions, though MAUDE is life insurance and Zurich’s workbench is commercial P&C.

Morrisons, the UK’s fifth-largest supermarket chain, is trialling Instacart’s AI-powered Caper Cart smart trolleys (built-in cameras and scales that recognise items as they’re loaded, first deployed at its Preston store in July 2026). A study by Bayes Business School at City St George’s, University of London, found shoppers using the AI trolleys spent 32% more on average than customers using standard carts, bought 25% more items, and remained in store 23% longer than standard-cart shoppers.

Caveat: the study’s methodology (sample size, control for self-selection into smart-cart use) is not detailed in this coverage; Instacart is the vendor, not the study’s author, and no figure here is vendor-originated. Independent academic research (tier 2), not a company press release.

Sammons Financial Group, a life insurance and annuities holding company, was named a winner in Salesforce’s 2026 Customer Success Awards for its Agentforce Voice deployment, which the release says handled more than 16,000 policy calls in its first six months and absorbed seasonal call-volume spikes without adding temporary staff. Salesforce’s separate customer-story page for Sammons (same publish date) gives overlapping but distinct figures for what appears to be the same program: “60% of calls resolved without a rep” and “more than 92,000 minutes of rep time” saved in the first six months. The two Salesforce-published pages do not reconcile which figure (16,000 calls vs. 92,000 minutes saved) is the primary metric, so both are reported here rather than merged.

Caveat: both sources are Salesforce’s own award/customer-story content (tier 3, vendor-published); no independent outlet or Sammons-issued release was found corroborating either figure.

The Grout Guy, a small tile-and-grout restoration business, was named a winner in Salesforce’s 2026 Customer Success Awards for a photo-analyzing Agentforce agent that turns customer-submitted photos into job quotes. The release says quote turnaround fell from 3 to 5 days to within 20 minutes, work orders now process in under two minutes 24/7, and administrator capacity rose from 4 technicians per administrator to more than 25.

Caveat: first-party Salesforce award content (tier 3, vendor-published), the smallest and least independently checkable company in this run; no independent coverage or company-issued statement found. Included because the beat definition does not set a company-size floor and the figures are specific and dated, not because the sourcing is any stronger than the other Salesforce-award items above.

Tottenham Hotspur Football Club, the English Premier League club, was named a winner in Salesforce’s 2026 Customer Success Awards for an Agentforce-built fan-service agent that consolidated a “golden record” spanning 4.6 million fans and 30 previously disparate systems. The release credits the agent with saving 80,000 minutes of handling time in its first month and says it is on pace to handle over 200,000 fan calls a year, with agent-assisted inquiries resolved in 10 seconds or less.

Caveat: first-party Salesforce award content (tier 3, vendor-published); no independent outlet or club-issued statement found corroborating the figures, and no baseline (pre-AI handling time or headcount) is given to judge the 80,000-minute claim against.

The US Department of War added OpenAI’s ChatGPT Mil and Starshield AI’s Grok for Government to GenAI.mil on August 31, 2026, joining Google’s Gemini for Government, which the platform launched with in December 2025. The platform is now available to more than 3 million military members, civilian employees, and contractors. Honest caveat: this is an access and scale figure, not a usage outcome; no hours-saved, error-rate, or cost figure was disclosed for what the 3 million personnel are actually doing with the tools, only that the tools are now available to them.

Microsoft disclosed that it now runs about 25 AI agents across its own supply chain operations, including a demand-planning agent for data center components, a computer-vision system that predicts spare-parts storage needs, and CargoPilot, which recommends shipping methods by weighing cost, speed, and carbon impact against transit time. Microsoft says the effort is saving its logistics teams “hundreds of hours each month,” an unquantified range rather than an exact figure. The same reporting notes Microsoft’s blog separately claims similar agents “save Dow millions of dollars,” a vendor-adjacent figure for a different company that this flash does not adopt. Honest caveat: single source, first-party claim about Microsoft’s own operations, no percentage or dollar figure for Microsoft itself, and the “hundreds of hours” range is imprecise by design.

Postal Savings Bank of China’s H1/Q2 2026 earnings-call highlights, relayed by Yahoo Finance (Aug 28, 2026), report 370 large-model-powered AI scenarios now live across the bank, a 40% increase in approval efficiency, and a 72% reduction in large-model deployment costs. All figures are company-reported from the bank’s own earnings disclosure, not independently audited, and no baseline period or absolute volumes are given for the approval-efficiency or deployment-cost figures, only the percentage change. Only one live source could be confirmed for this item; no independent press corroboration was found.

Cisco is rolling out MyAgent, a personalized AI agent built on its internal Circuit platform, to its entire 90,000-person global workforce, moving from pilot to company-wide deployment of supervised autonomous execution across Outlook, Webex, Jira, SharePoint and other internal tools. Cisco’s own blog post says agentic interactions grew nearly 350% quarter over quarter, though it gives no baseline volume for that growth. Both figures are company-disclosed (Cisco’s own blog, relayed independently by PYMNTS); no third party has audited the interaction-growth figure.

Related case files

None yet; no Cisco case file exists in the corpus.

Discovery Bank CEO Hylton Kallner told the CNBC Africa AI Summit in Sandton, South Africa (2026-08-27) that more than half of all client interactions now run through the bank’s in-app, natural-language AI channel, a shift completed in three to four months, with client satisfaction scores on the channel the best the bank has recorded. Two independent outlets covering the same summit, CNBC Africa and the Sandton Times, carry the identical figure. A separate PYMNTS report five days later adds that the same system “fully resolves about 55% of client queries on the first interaction” and references a headline 500% ROI figure for the deployment. All figures are company-disclosed by the CEO on the record at a public event; no independent audit of the interaction share or the ROI number was found.

Related case files

None yet; no Discovery Bank case file exists in the corpus.

At the same CNBC Africa AI Summit in Sandton (2026-08-27), a PwC Africa speaker described an internal sustainability-reporting tool that “has taken work that used to take two weeks and reduced it into a couple of hours,” part of a broader claim that nearly two-thirds of African PwC workers are already using AI day to day (a figure the Sandton Times’ separate report of the same summit also carries). The two-weeks-to-hours figure itself was found in only one outlet; “a couple of hours” is PwC’s own approximation, not a precise duration, and no baseline task volume or independent verification was found.

Related case files

None yet; no PwC Africa case file exists in the corpus.

On TD Bank Group’s Q3 fiscal 2026 earnings call (2026-08-27), Group President and CEO Raymond Chun told analysts the bank had “essentially hit” its fiscal 2026 target of $200 million in value extracted from AI, three quarters into the year, spanning predictive, generative and agentic AI use cases across areas including credit-application document review and contact-center interactions. This is a first-party company figure, stated by the CEO on the earnings call itself rather than measured independently; no breakdown by use case or methodology for the $200 million figure was given in the transcript. Only one transcript source was checked this run (Motley Fool’s reproduction of the call); other outlets carrying the same transcript would repeat, not independently corroborate, the same CEO statement, so this is marked single-source.

TD Bank Group told investors it generated 195 million Canadian dollars (about $141 million) in AI value across the first three fiscal quarters of 2026, putting it ahead of pace on its full-year target of 200 million Canadian dollars (about $144 million), according to CEO Raymond Chun on the bank’s earnings call. The bank names three concentration areas: retail end-to-end credit (application submission, document review, decision acceleration), the software development cycle, and contact centers, plus TD Insurance’s AI vehicle-damage estimation for auto claims and TD Auto Finance Canada, where the bank says it “automated approximately one-third of the manual processes in funding and launching digital income verifications.” The figure is a company-disclosed, CEO-stated internal value estimate relayed by PYMNTS from the earnings call; no independent third party has audited the $141 million calculation methodology.

Related case files

None yet; no TD Bank Group case file exists in the corpus.

At the CNBC Africa AI Summit in Sandton (2026-08-27), Naeem Seedat, Telkom’s group executive for digital strategy and venturing, said the operator posted “a 99.9% network uptime year, built on machine learning that predicts faults before they occur.” He framed the strategy as reaching an “AI last mile,” connecting “every person, every business, every entrepreneur” to AI affordably. Honest caveat: this is a single self-reported snapshot figure with no stated prior-year baseline or independent confirmation, only one outlet reported it, and Telkom’s own public network-performance disclosures elsewhere cite different per-network-layer uptime numbers (99.94%/99.88%/99.99% across access, transport and core), so it is not clear which measurement “99.9%” rolls up or over what period.

Related case files

None yet; no Telkom case file exists in the corpus.

Cigna’s chief AI and analytics officer told Fortune (Aug 26, 2026) that AI-driven identification of patients with emerging chronic conditions, paired with proactive outreach from more than 1,250 Cigna Healthcare clinicians, is projected to save an estimated $200 million over three years by connecting patients with care earlier. A separate Cigna release (July 23, 2026) ties that same $200 million figure to a $2,000-per-year average medical-cost reduction for engaged customers and a 42% cut in avoidable inpatient stays among early-engaging customers. Cigna also told Fortune that AI-enabled summarization at its MDLIVE telehealth service has cut clinician note-taking time by up to 90%. All figures are company-reported, projected or self-measured; none carry independent audit, and the $200 million is a forward estimate over a three-year window rather than a figure already realized.

Meta drew up “Project OT” (Organization Transformation) at a January leadership retreat, planning two waves of cuts built around AI taking over work currently done by thousands of employees. Scenario-planning for the plan explored reducing the headcount of some individual teams by as much as 60%, though Meta says it never intended to cut 60% of its overall workforce. Internally tracked engineering metrics cited in the reporting cut against the AI-native case: code changes to Meta’s AI platforms and infrastructure rose sharply year over year, but new or improved features that actually reached users rose just 36%, while technical and security incidents rose 40% and the time employees spent resolving them grew further. Hours before the first wave of layoffs on 20 May, CEO Mark Zuckerberg called off the planned second wave for November; Meta went ahead with a 10% workforce reduction the next day. Both sources trace back to the same Reuters investigation and Meta’s confirmation of Project OT’s existence, so this is one underlying account relayed by two independent outlets, not two independent measurements.

Related case files

None yet; no Meta case file exists in the corpus.

Singapore’s Land Transport Authority has been using AI-powered wall-panel mounting robots, built with Shenzhen-based startup ZoeRobot, on the Cross Island Line rail project since 2025. The outlet reports the robots have cut manpower requirements for compaction work during concrete casting by around 30%, in an item framed as part of Singapore-Guangdong digital-economy cooperation; other coverage of the same figure attributes it to EnterpriseSG, the Singapore government trade agency, though that attribution could not be independently confirmed on a second live-fetched page this run. Single-sourced: only one outlet carrying a live-quotable figure was found.

Related case files

None yet; no Singapore LTA or transport-construction case file exists in the corpus.

Shipping company Eastern Pacific Shipping (EPS) is expanding its use of Orca AI’s collision-avoidance and bridge decision-support platform to dozens more vessels after a pilot on five ships covering more than 130,000 nautical miles. Across the pilot fleet, close-encounter events fell 48% per 1,000 nautical miles and average minimum passing distance rose 10%; one vessel, the Pacific Sentinel, saw an 84% reduction in open-water close-encounter events and a 50% increase in minimum passing distance. The originating account is a joint Orca AI/EPS release (vendor- and-client co-authored, tier 3); independent maritime trade press (Digital Ship, Aug 26 2026) relayed the same 48% figure without adding independent measurement. No third party appears to have audited the underlying safety data.

Bradesco CEO Marcelo Noronha told the Febraban Tech 2026 conference in Sao Paulo on 2026-08-24 that the bank’s generative-AI assistant, BIA, processed 74 million transactions through the app and WhatsApp. Three independent Brazilian trade outlets covered the same remarks; two (Consumidor Moderno, Paraiba Business) quote the period as “o primeiro semestre” (the first half of 2026), while a third (Mobile Time) quotes Noronha, in language repeated across two separate fetches of the same article, as saying “o primeiro trimestre” (the first quarter). The two windows conflict and neither this session nor the cited outlets resolve it; both are reported here rather than merged. Paraiba Business additionally cites an 87% resolution rate for BIA without human handoff, a figure not corroborated elsewhere. In the same reporting window, a Sao Paulo bank workers’ union survey found Bradesco closed 324 branches and cut about 2,400 jobs between June 2025 and June 2026; Bradesco’s own stated justification, per Bnews, is that roughly 98% of transactions now happen through digital channels, not an explicit AI attribution, so the branch and headcount figures are reported here as context rather than as a claimed consequence of BIA specifically.

Related case files

None yet; no Bradesco or Brazilian-banking case file exists in the corpus.

Livia Chanes, Nubank’s CEO in Brazil, told a panel at Febraban Tech 2026 on 2026-08-24 that AI tooling has already lifted the bank’s engineering code-shipping capacity by close to 90%, describing it as an observed result (“a gente já viu”, “we’ve already seen”) rather than a target. Two independent Brazilian trade outlets covering the same panel quote nearly identical language. Neither report states the baseline period being compared, what share of engineering teams the figure covers, or whether it is a speed measure (commits or deploys per unit time) versus a headcount-adjusted output measure; Chanes is also quoted saying similar gains show up in marketing and operations without a figure for either. This is a company executive’s own stated internal metric, not independently audited. A separate, unrelated Nubank figure (a 37-point AI-assistant NPS gain, sourced to a 2026-07-07 report) was not repeated in this panel and is not part of this flash.

Related case files

None yet; no Nubank case file exists in the corpus.

Albertsons Companies reported that average order value rises 10% when shoppers use its conversational search feature and 26% when they use its more comprehensive AI shopping assistants, per PYMNTS’s Q2 2026 retail-earnings roundup. No absolute order-value figures or total revenue impact were disclosed, and the comparison is self-reported by the company rather than independently measured.

Amazon disclosed that U.S. customers who use Alexa for Shopping spend 40% more per order than customers who do not, with active users of the feature nearly doubling year over year and more than 350 million shoppers having used the tool in the past year, per PYMNTS’s roundup of Q2 2026 retail earnings commentary. As with Walmart’s Sparky figure, this is a self-reported correlation between AI-tool users and non-users rather than an independently isolated causal effect, and Amazon did not break out revenue attributable to the assistant.

On Deere’s fiscal Q3 2026 earnings call (August 20, 2026), Deanna Kovar, president of Deere’s Worldwide Agriculture and Turf division, said customers running Deere’s computer-vision See & Spray system, which identifies weeds versus crops and sprays only the weeds, are achieving more than 50% herbicide savings on significantly more acres year over year. Factory adoption is expected to nearly double, with the system on order for roughly one-third of North American sprayers for model year 2027. The savings figure is company-stated on an earnings call, relayed via an independent transcript service; no independent agronomic audit of the herbicide-reduction figure is cited.

Related case files

None yet; no agriculture case file exists in the corpus.

Lowe’s reported second-quarter fiscal 2026 results (quarter ended around 31 July, released 19 August) showing online sales up 15.7% year over year, which the company’s own release and earnings call tied in part to Mylow, its AI-powered digital shopping assistant. On the earnings call, management said online customers who use Mylow convert at three times the rate of those who do not, and that Mylow, which also powers an associate-facing AI companion app, has answered more than 25 million questions from customers and associates since launch. Total company sales were $26 billion, up 8.3% year over year, while comparable sales grew only 0.2%, so the AI-attributed lift is concentrated in the online channel, not the business overall. The conversion and question-count figures are self-reported by Lowe’s management on the earnings call; no independent measurement of the AI-attributable share of the 15.7% online growth exists.

Rakuten Group, the Tokyo-based e-commerce and fintech conglomerate, reported Q2 FY2026 results on 10 August showing its advertising business grew to 65.6 billion yen in revenue, up 15.6% year over year, which the company attributed to AI-powered search and targeted banner advertising on its Rakuten Ichiba marketplace and Rakuten Travel unit. The company said it uses AI both to generate ad creative (static banners and video) and to handle screening and bidding for high volumes of campaigns. The figure originates from Rakuten’s own first-party investor disclosure and was independently reported the same day by MediaPost’s Performance Marketing Insider. No third party has separated the AI-attributable share of the advertising growth from other factors such as overall marketplace volume growth; the 15.6% figure covers the whole advertising segment, not an isolated AI effect.

Suffolk County Council is rolling out an AI transcription and write-up tool (Liquidlogic’s FormFlow AI Assistant) to every adult social care practitioner after a pilot found the time to write up an assessment fell from around three hours to roughly 45 minutes, reported by BBC Radio Suffolk political reporter Vikki Irwin on 20 August 2026. Vendor case-study material from System C, published in March 2026, gives a more precise figure from the same pilot: a 67.9% reduction in write-up time across all completed form types. Social workers still review the AI-gathered information and apply their own professional judgement before recording a decision, and the council’s own numbers are provisional, drawn from a small control group with no measured effect yet on waiting lists or caseload size. The British Association of Social Workers has warned the resulting time savings could be used to cut social work posts rather than reinvested in reducing caseloads, and a University of Suffolk social work professor cautioned the tool must not be used to do the analysis itself.

Target told investors that digital traffic arriving from external AI platforms is growing more than three and a half times faster than the industry average, and that AI-powered teacher and college wish lists saw creation volume jump more than 50% ahead of the back-to-school season, according to PYMNTS’s Q2 2026 retail-earnings roundup. Both figures are growth-rate comparisons rather than absolute traffic or revenue numbers, and neither was independently verified outside the company’s own earnings commentary.

Walmart told investors during its Q2 fiscal 2026 earnings round that shoppers who use Sparky, its AI shopping assistant, spend 40% more per order than those who do not, and that total Sparky users are up 70% year over year. PYMNTS reported the figures as part of a roundup of AI-attributed spend increases across several retailers this earnings season. The comparison is a self-reported behavioral correlation between AI-assistant users and non-users, not an independently measured causal lift, and no dollar total or absolute user count was disclosed alongside the percentages.

Absa Group’s H1 2026 results statement (2026-08-19) says straight-through processing has cut SME account-opening turnaround from two days to under 30 minutes, and that financial spreading for credit decisioning, previously two to five days, can now be done in around four hours. IT News Africa’s independent report two days later confirms both figures and adds contact-centre numbers: Amazon Connect-based AI “helped improve contact-centre response times by up to 21%, reduce average handling time by 14% and cut call-holding time by 44%.” These are new figures for a broader H1 2026 technology update, distinct from Absa’s already-reported 41% document-indexing gain in its debt review team (2026-08-05/06, in the corpus at radar/2026-08-24/ and radar/2026-08-29/). All numbers are company-disclosed in an earnings statement; no independent audit was found, and Absa gives no baseline volume for the SME onboarding or financial spreading figures.

Related case files

None yet; no Absa Group case file exists in the corpus.

Crash Champions, one of the largest founder-led multi-shop collision repair operators in the US, expanded its use of CCC Intelligent Solutions’ Mobile Jumpstart 2.0, an AI-powered estimating tool, according to a joint announcement on 13 August 2026. The companies say Mobile Jumpstart now captures, on average, 84% of the final total repair bill in about 70 seconds, versus a manual estimating process that typically takes 15 to 30 minutes. Crash Champions was the first multi-shop operator to adopt Mobile Jumpstart in 2025 and has now renewed and expanded the relationship. The figure is vendor-and-client co-reported (CCC Intelligent Solutions and Crash Champions jointly issued the release); no independent third party has measured the capture rate or timing, and no other outlet has yet confirmed these specific numbers independently.

DBS Bank has rolled out an agentic AI credit-memo tool to approximately 1,500 relationship managers and credit risk managers globally, up from a pilot of about 150 staff, according to the bank’s own August 19, 2026 newsroom post, independently relayed two days later by fintech.garden. The system runs more than 70 specialised-agent tasks, drawing on annual reports, industry research and internal bank records to produce a review-ready first draft of a corporate credit memo. DBS states preparing credit memos and related credit activities currently “can account for up to 40%” of a relationship manager’s time, and the bank is targeting “at least 30%” reduction in that time, a stated goal rather than a measured, achieved result; the 1,500-employee global rollout scale itself is the concrete, achieved figure. The tool is explicitly positioned as a drafting aid, with bankers remaining responsible for reviewing, revising and finalising the memo. No independent measurement of the actual time saved once at this scale has been published yet.

Related case files

None yet; no DBS case file exists in the corpus.

Hyundai Motor Group disclosed (originally via a company announcement around Aug 12, 2026, reported by PYMNTS on Aug 19, 2026) that its H Chat Pro generative AI platform, which gives employees access to models including ChatGPT, Gemini and Claude, has cut the time engineers spend searching for and reviewing historical crash-test cases by 90%, and that active use surpassed 30,000 employees in July 2026, about 80% of general staff at Hyundai Motor and Kia. Separate, distinct AI systems on the production line are credited with cutting maintenance response time by 42% and reducing unnecessary production downtime by 86% via a reinforcement-learning cart-routing system. All figures are company- reported; PYMNTS’s report is the fullest independent relay found, and no outlet has independently audited the underlying data.

Singapore Airlines has deployed more than 160 AI applications across its business, drawn from a pipeline of more than 550 identified generative-AI use cases with over 140 implemented, according to reporting on 18 August 2026 that cites the airline’s FY2025/26 annual report and a 2025 AGM presentation. The carrier’s no-code GenAI assistant platform, JARVIS, reached 5,079 unique users out of 5,600 ground staff, around 90% penetration. The figures trace to Singapore Airlines’ own annual report and shareholder-meeting disclosures (first-party, though from regulated corporate reporting) and were reported by two independent Asia-based tech outlets. No independent party has verified the underlying usage logs; the adoption figures are the airline’s self-reported numbers.

Standard Bank Group, the Johannesburg-based lender and Africa’s largest bank by assets, disclosed in its 2026 interim results (six months to 30 June) that 72% of employees were active users of generative AI tools, up from a much smaller base, with 87 use cases approved across the group. The bank also said its AI-enabled recommendation capabilities supported more than 10 million personalised client interactions in the period. The figures originate from Standard Bank’s own interim results booklet, a regulated financial disclosure, and were reported independently by two South African tech publications (ITWeb and TechBooky) on 13 August 2026. No third party has independently measured the underlying usage or the client-interaction count; both are the bank’s self-reported figures from a first-party financial filing.

Sauder Woodworking, the Ohio furniture manufacturer, says its plant floor deployment of QAD Redzone’s ChampionAI agent cut average changeover time from 19 minutes to 1 minute 24 seconds, a 93% drop, and raised productivity 40% in its first year on the platform. In the pilot area, overall equipment effectiveness climbed from about 50% to above 70%; an outside ROI analysis credited the deployment with a 665% return on investment and a seven-month payback. The figures are Redzone-published, vendor-attributed to Sauder’s own VP of Operational Excellence, and not independently audited.

Related case files

None yet; no manufacturing case file exists in the corpus for this deployer.

Delivery Hero launched an agentic AI assistant in August 2026 that helps local restaurant and shop partners on its platform brands, including Glovo, run growth tasks such as promotions and menu changes without a dedicated marketing team. Coverage says the assistant has supported more than 40,000 restaurants and produced a 15% increase in orders for Glovo users, alongside a qualitative merchant testimonial in an independent write-up. Neither source names the total merchant base the 40,000 draws from beyond Delivery Hero’s roughly 1.5 million platform partners, nor gives a measurement window for the 15% figure; both write-ups draw on the same company talking points and executive quotes rather than independent measurement, so this is treated as single-source despite two outlets carrying it.

Related case files

Delivery Hero and n8n: account-lockout recovery automated in a single workflow is the corpus’s existing Delivery Hero case file, covering a different function (IT ops automation, not merchant-facing marketing).

On DocGo’s Q2 2026 earnings call (August 17, 2026), CEO Lee Bienstock said the mobile-health company’s internally built AI communications tool now handles 60% of inbound patient calls without a live agent, and 100% of outbound appointment-scheduling calls, with the agent fielding over 1,000 calls a day without human intervention. Bienstock separately said AI data-entry integrations now account for 65% of orders received, rising to an anticipated 90% by year-end, and that the new system is bringing the cost to process each order “from $2 down to mere pennies.” All figures are company self-reported from the earnings call; no independent audit or baseline methodology was disclosed, and no independent press coverage of the specific tool was found at time of writing.

Oracle’s headcount fell from about 162,000 to roughly 141,000 over the fiscal year ended May 31, 2026, a 21,000-person, 13 percent decline. The company has said in its own filings that AI adoption is part of the cause, not the whole cause: it also cites its 2026 Restructuring Plan and the ordinary churn of a company redirecting spend toward AI data-center buildout ($55.7 billion in capex in fiscal 2026, up from $21.2 billion a year earlier). Reporting on August 12, 2026 says Oracle is planning a fresh round of cuts, targeting the September 1 start of its next fiscal quarter, with some teams facing double-digit percentage reductions. The AI-attribution language comes from Oracle itself, not an independent headcount audit, so treat it as a company claim about causation layered on an independently reportable number.

Unilever’s Hefei, China factory-to-consumer operation, which manages roughly 31,000 orders a day across more than 400 products, reported an 8% increase in overall equipment effectiveness, a 20% waste reduction, and a 25% drop in logistics costs after adopting AI-driven digital twins and demand planning, per two independent trade outlets covering Unilever’s supply-chain briefing this week. Separately, Unilever said AI-driven weather-data analysis improved ice-cream demand forecast accuracy by 10% in Sweden, though that figure appears in only one of the two outlets checked. Both outlets frame the figures as Unilever’s own disclosures rather than independently audited measurements, and Unilever’s own newsroom page carrying the original claims could not be fetched directly (403) for a byte-level check.

On India’s 80th Independence Day (15 August 2026), the Prime Minister’s Red Fort address was live-translated into 22 Indian languages by Bhashini, the government’s open-source AI speech-to-text/translation/text-to-speech pipeline built under the Ministry of Electronics and Information Technology. Coverage frames this as the first Independence Day address processed through an AI translation pipeline at national scale, though that framing traces to a single outlet (Tech Times) whose page could not be fetched (HTTP 403) and so is not included as a quoted claim here; the 22-language, national-broadcast scale figure itself is corroborated across two independent outlets reporting the same day. No error rate, latency, or usage figure for the translation pipeline was disclosed by either source. Public-sector deployment, a persistently thin beat in this corpus.

Speaking on Carvana’s Q2 2026 earnings call (July 29, 2026), CEO Ernie Garcia credited the company’s AI customer experience agent, Sebastian, with cutting customer care costs year over year for four straight years: 40% three years ago, another 30% two years ago, another 20% one year ago, and another 10% this year. Each figure is a year-over-year reduction on a shrinking base, not a cumulative 100% cut, and the article does not state the dollar or per-contact baseline these percentages are measured against, nor whether the reductions are net of any growth in Carvana’s underlying contact volume. This is a company-reported figure from an earnings call, not independently audited.

Reporting its full-year 2026 results on August 12, 2026, Commonwealth Bank of Australia said AI-generated use cases delivered around $200 million (AUD) in gross benefits during the fiscal year, including reinvested capacity, with about $100 million of that generated incrementally during FY26 itself. The same results period covered a more than twofold increase in auto-decisioned small business loans through its BizExpress tool and an AI banker workbench expanded to more than 800 frontline staff. This is the bank’s own reported figure from its results release, not an independently audited number, and the article does not break out how “gross benefits” was calculated or which use cases contributed most.

Covering Lemonade’s Q2 2026 results (reported July 29, 2026, revenue up 79% year over year), Insurance Business magazine reported that the insurer says roughly 40% of its claims are now handled instantly by its AI claims bot, known as AI Jim, without human involvement. The 40% figure is attributed to the company itself rather than an independent audit, and the article does not specify what counts as “instant” (seconds versus same-day) or how the figure is measured across claim types. Lemonade has said elsewhere it aims for the underlying AI Jim tool to process straightforward payouts in seconds, though that per-claim speed claim is not independently quoted here.

On Porch Group’s Q2 2026 earnings call (August 7, 2026), COO Matt Neagle said broad adoption of AI tooling across the engineering organization drove a 2.4x increase in lines of code changed and a 73% increase in merge requests created, without giving a baseline period or headcount context for the comparison. Lines of code and merge-request counts are activity metrics, not direct measures of software quality or output value, and the company did not disclose whether engineering headcount, defect rates, or release cadence changed alongside the increase. This is a company self-reported figure from its own earnings call, not independently measured.

On its Q2 2026 earnings call (July 28, 2026), UPS management said 68.5% of US package volume now flows through automated facilities, up from 64% a year earlier, and that automated buildings run at a 28% lower cost per package than non-automated ones. This mixes robotics, sorting automation, and AI-powered routing rather than a single AI tool, and the figure is a company-reported milestone rather than an independent audit; the same call also noted RFID and AI-powered digital twins are now fully deployed across UPS domestic operations, feeding the routing gains. Freshness note: this flash draws on the July 28 call, 17 days before this run, kept because of the scale (a national parcel network) and the specificity of the figure rather than a vague trend claim.

On Airbnb’s Q2 2026 earnings call (August 13, 2026), CEO Brian Chesky said AI-driven engineering has cut the time from concept to launch on some key initiatives by as much as 60 percent, and lifted the number of features and improvements shipped this year by nearly 80 percent versus the same six months last year. This is a distinct claim from Airbnb’s AI-assistant customer-support figures already covered in radar/2026-08-09 (45 percent issue resolution, 16 percent lower cost per booking); this one concerns internal engineering velocity, not customer-facing support. Both figures are self-reported by the CEO on an earnings call with no independent audit, and “key initiatives” and “features and improvements” are not defined or itemized in the transcript.

On Klaviyo’s Q2 2026 earnings call (August 12, 2026), management said retailer Boston Proper launched Klaviyo’s Customer Agent and within two weeks generated 3x ROI in incremental revenue from it being embedded on their website, while the agent also autonomously resolved more than half of their support tickets. No dollar baseline or absolute revenue figure is given beyond the 3x multiple, and the number is vendor-reported by Klaviyo describing a customer’s early results, not independently audited.

On Amplitude’s Q2 2026 earnings call (August 12, 2026), management described Coca-Cola FEMSA, which sells to hundreds of thousands of small shops across Latin America, using AI to write a distinct weekly recommendation message for each retailer instead of one broad campaign. Using Amplitude to measure the results, the AI-personalized campaigns got an 11% click-through rate, 4 times higher than the company’s previous approach, with the lift reported to persist in the following weeks. The figure is vendor-reported by Amplitude describing a customer’s campaign results, not independently audited, and no absolute click volume or revenue figure is disclosed.

On NICE’s Q2 2026 earnings call (August 12, 2026), management said GXBank, described as Malaysia’s first digital bank, has AI agents that autonomously resolve 70% of customer chat interactions while holding a 95% customer satisfaction rate. The figure is vendor-reported (NICE describing a customer’s deployment of its platform), not independently audited, and no baseline pre-AI resolution rate was disclosed.

On HubSpot’s Q2 2026 earnings call (August 12, 2026), CEO Yamini Rangan said HubSpot’s own customer agent now resolves 72% of support tickets without human escalation, offered as evidence its AI agents deliver real outcomes rather than just usage volume. The figure is company self-reported from an earnings call with no independent audit and no stated baseline or time window for the 72% resolution rate.

On Freshworks’ Q2 2026 earnings call (August 11, 2026), management described iQor, a business process outsourcing firm with more than 40,000 employees, replacing a legacy on-premise system with Freshservice and the Freddy AI agent. The company’s agentic AI solution fully automated 35% of iQor’s IT service delivery and cut its monthly ticket volume by 39%. The figures are vendor-reported by Freshworks describing a customer’s deployment, not independently audited, and no pre-deployment baseline or measurement window is given beyond “monthly.”

Microsoft is discontinuing four consumer Copilot features (Group Chats, AI-generated podcasts, Copilot Labs experimental features, and Deep Research) by August 18, 2026, and dropping the Mico animated-character mascot, while merging its separate consumer and business Copilot apps into one. TechCrunch reports Microsoft EVP Jacob Andreou told staff in an internal memo (via The Information) that Copilot needed to earn “the right to exist” in customers’ lives, which meant “moving on from features that didn’t work.” This is a walk-back rather than an outcome figure: no usage or adoption numbers are disclosed for the discontinued features, only the list of what is being cut and the date. The web.archive.org save request for the TechCrunch source returned rate-limited (429) at the time of this run and could not be confirmed; the quote above was confirmed byte-for-byte against the live page.

On HubSpot’s Q2 2026 earnings call (August 12, 2026), management said Sesame HR, described as a 400-person HR software platform, could only respond to 70% of incoming support tickets before deploying HubSpot’s customer agent. After deployment, Sesame HR covers all tickets received, with 60% fully resolved without human escalation, and has since bought more than 1 million additional credits and expanded into other HubSpot agents. The figures are vendor-reported by HubSpot describing a customer’s deployment, not independently audited.

On NICE’s Q2 2026 earnings call (August 12, 2026), management said TripAdvisor’s AI agent, live on automated voice calls within 2.5 months of deployment, delivers a 90 customer sentiment score versus 71% achieved by human agents. The two scores are not stated to be on strictly the same scale in the transcript beyond “90” versus “71%,” and the figure is vendor-reported by NICE describing a customer’s results, not independently audited.

On its Q2 2026 earnings call (August 12, 2026), Choice Hotels said its property-management-system AI teammate Charlie reduced requests for operational support by about 40 percent in an early pilot, and that its AI-enabled EasyBid platform improved group RFP conversion by 360 basis points, contributing to 16 percent year-over-year growth in group revenue. Both figures are self-reported by management on the earnings call, not independently measured. “Early pilot” is not scoped to a location count or a time window, and the 360-basis-point RFP figure is attributed to EasyBid alongside other unspecified factors in the same sentence about group revenue growth.

On Cisco’s Q4 FY2026 earnings call (August 12, 2026), Chair and CEO Chuck Robbins said 145,000 customer support cases were resolved entirely by AI with zero human intervention over fiscal 2026. Robbins also said Circuit, Cisco’s internal on-prem AI assistant, is “fully embedded in how Cisco operates,” supporting over 75 million prompts in the fourth quarter alone. This is Cisco describing its own internal support and operations use, not a third-party client deployment; the figures are company self-reported from an earnings call with no independent audit or baseline disclosed.

On CVS Health’s Q2 2026 earnings call, management said the company moved millions of calls out of its retail pharmacy business into conversational AI, which “allowed us to refocus 1 million hours of pharmacist hours” toward clinical care instead of administrative calls. Separately, executives said Aetna One advocates “used to spend 90 minutes preparing and reviewing a case to help a member and now it takes them only 2 minutes,” attributing the drop to internal AI tools. Both figures are company self-reports on an earnings call, with no independent audit and no breakdown of how the 1-million-hour figure was calculated or over what period.

Credit Acceptance Corporation’s Q2 2026 earnings call materials (transcript published August 11, 2026, covering results reported August 4) show its AI agent handling 67 percent of inbound customer service and account solutions calls in June 2026, up from 27 percent in March, a jump of 40 percentage points in three months. The company frames this as driving “improved efficiency, enabling faster 24/7 customer self-service, and reducing cost-to-serve at scale.” No independent measurement of call outcomes, resolution quality, or customer satisfaction accompanies the routing-share figure; it is a first-party operational metric, not an audited result.

Duke Energy said its automated self-healing grid technology, which reroutes power around faults in seconds, prevented more than 12,600 customer outages in Wayne County, North Carolina during 2025. The company said more than 75% of its Wayne County customers are now served by the technology, part of a broader North Carolina rollout that has added roughly 2 million customers to the system since 2022. This is a company self-reported figure with no independent verification; the article does not use the terms “AI” or “artificial intelligence,” describing the system only as automated sensing and switching hardware, so its qualification here rests on the automation half of the beat rather than an AI label.

On its Q2 2026 earnings call (August 11, 2026), FIS CEO Stephanie Ferris said the company has launched five agentic AI programs across its own internal servicing operations, cutting manual tickets by 70% and triage time by nearly 75%. This is a company self-reported figure describing FIS’s own service-desk operations, not a client’s; Ferris also cited more than 40,000 active internal AI Copilot users generating over 16 million assisted actions and 1.5 to 2x engineering throughput with 30% fewer defects, none independently audited.

On its Q2 2026 earnings call (August 8, 2026), Global Payments CEO Cameron Bready said the company’s Revenue Boost product, already credited with $2 billion in annual approval uplift for merchant clients, is now producing an additional 50 basis point increase in approval rates through AI-powered intelligent decisioning. This is a company self-reported figure from an earnings call, not an independently audited result, and Global Payments did not disclose the underlying transaction volume or methodology behind the uplift calculation.

On Kimco Realty’s Q2 2026 earnings call, EVP Will Teichman said the shopping-center REIT has yielded approximately 5x the return on the money it has spent and invested in AI initiatives year to date, meaning expense savings are running five times the cost of the effort. He separately said weekly AI utilization is “above 80% among all of our associates,” with AI-powered workflows now in place across asset management, leasing, underwriting, legal and other key functions. Both figures are company self-reports on an earnings call, not independently audited, and the transcript gives no breakdown of which functions drove the savings or how the 80% utilization figure is measured.

On its Q2 2026 earnings call (August 11, 2026), Snap CEO Evan Spiegel said the company’s internal AI code reviewer now reviews 90% of pull requests across the company and has saved an estimated 30,000 hours of code review time, while its AI-powered support agent answers about 3.9 million Snapchatter questions a month and has cut support ticket volume by roughly 62% since the start of the year. Both figures are company self-reported from an earnings call with no independent audit or underlying baseline disclosed; Snap separately said AI also lifted first-pass image review automation in advertising from 40% to nearly 90% year over year.

Verizon has removed roughly 16,600 positions from its payroll since October 2025 across three rounds of store-franchise transfers, corporate cuts, and retail-worker transitions, according to a tally built from WARN filings and company statements. On the July 16 round specifically, Verizon spokesman Rich Young told Fierce Network the cuts “have nothing to do with AI.” That denial sits against CEO Dan Schulman’s public statements this year that Verizon is racing to complete an internal AI tech stack, and CFO Tony Skiadas’s claim on the Q1 2026 earnings call that AI in network operations already resolves 85% of network issues without manual intervention. The two figures come from different rounds of the story (an April earnings-call claim about network automation, a July denial about the layoffs) and should not be merged; whether the July cuts are AI-driven or not is genuinely disputed between the company’s own executives and its spokesperson.

On its Q2 2026 earnings call (August 11, 2026), Wayfair co-founder and CEO Niraj Shah said the company’s proprietary, in-house AI pipeline for its Perigold luxury brand replaced roughly $2 million in traditional seasonal imagery production costs (travel, crews, studio time) with less than $10,000, a reduction of over 99%. The pipeline composes and renders photorealistic room scenes with automated quality checks and stylist guidance; the figure is company self-reported from an earnings call with no third-party audit of the underlying production budget.

Wyndham CEO Geoff Ballotti told analysts the company’s Wyndham Connect AI guest-messaging platform, now live at more than 5,000 hotels, has handled over 40 million guest messages and averages about 260,000 interactions a day. He said reservations completed autonomously through the system carry average daily rates 15% higher than phone-booked reservations, and that participating hotels have seen direct-channel contribution rise by as much as 500 basis points. A separate premium tier, Wyndham AI Concierge, is live at about 1,500 properties, and highly engaged franchisees reported $100,000 or more in additional revenue from selling upgrades and extended stays through the platform. These are company-reported figures relayed through executive remarks, not independently audited results.

Cumbria County Council, UK, deployed a tool that automatically processes some parking permit applications and school invoices without staff involvement, saving more than a tonne of paper over three months, according to a Conversation piece on council AI adoption (Aug 10, 2026). The only figure given is the paper-volume saving; no cost, headcount, or processing-time figure is disclosed, and the source is a research-relay article rather than a council statement or dedicated news report, so this is a thin, single-sourced item flagged mainly because it is a named, verifiable case in a beat (UK local government) the corpus is short on.

Kinney Drugs, a pharmacy chain serving upstate New York and Vermont, rolled back its AI phone assistant “Burt” after hundreds of customer complaints described incoherent calls, wrong dosages, and missed prescription notifications. Burt launched in May 2026 to handle incoming patient calls. President John Marraffa said: “One thing I want to point out is getting privacy and security right does not mean we got the experience right. We did not, and we own that.” The company reverted incoming patient calls to its “classic touch-tone phone system” and will keep Burt only for outbound refill texts requiring customer opt-in. Single source found (WWSB/mysuncoast.com, a Gray Media ABC affiliate); the underlying detail (dosage and notification errors, the president’s quote) reads as a direct company statement to the outlet, not independently measured.

On its Q2 2026 earnings call (August 7, 2026), Carter’s Chief Retail and Digital Officer Allison Peterson said the company’s upgraded AI consumer chat tool now manages one-third of customer contacts, freeing staff for what she called premium high-touch care for the company’s best consumers. This is a company self-reported figure from an earnings call; no independent measurement or baseline contact volume was disclosed.

Wealth manager Edward Jones has 300 advisors and associates running the AI tax-planning tool Worthy in an ongoing pilot, InvestmentNews reports. The vendor says Worthy has now processed more than $1 billion in taxable client income uploads across its customer base, and describes the time savings on a tax-planning task as taking “something that’s traditionally a three or four-hour job down to 15, 20 minutes.” The $1 billion figure and time-savings estimate are vendor-originated (Worthy’s own claims relayed by the vendor to the reporter), not an Edward Jones-published or independently audited number, and the article does not say how many of Edward Jones’ 300 pilot associates are actively using the tool versus onboarded.

Kern Family Health Care, the largest Medi-Cal plan in Kern County, California, deployed an AI voice outreach agent named Angelica (built by startup Careforce) to call members and help them complete Medi-Cal renewal paperwork ahead of new federal work-requirement rules. Since launch late last year the agent has placed more than 800,000 calls to 387,000 members in 30-plus languages, at a software cost of about $370,000 against an estimated $2.4 million the plan says it would have spent to staff the same volume by hand. KFF Health News reports Kern Family’s April Medi-Cal renewal rate reached 94.9%, though the plan’s own figures on cost avoidance are estimates, not an independently audited comparison, and the article does not isolate how much of the renewal rate is attributable to the AI calls versus other outreach.

On Lloyds Banking Group’s Q2 2026 earnings call, CEO Charlie Nunn said the bank is on course to deliver more than 100 million pounds of value from generative and agentic AI within 2026, tied to its “Accelerate 2030” efficiency programme. This is forward-looking guidance stated on a public earnings call, not a completed or independently audited outcome, and the transcript does not break out how much of the 100 million has already been realized versus is still projected for the rest of the year. web.archive.org’s Save Page Now returned no successful capture for this Motley Fool transcript page (503 from the origin on repeated attempts); the CDX index shows no prior snapshot either, so this citation is currently unarchived, an open gap rather than an omission.

On OneSpaWorld’s Q2 2026 earnings call, management (Stephen Lazarus) said AVA, the company’s AI virtual assistant supporting ship operations since its August 2025 launch, has autonomously resolved 96 percent of support tickets without human intervention. The figure is company self-reported on an earnings call, covers roughly one year of operation, and the transcript does not define “support ticket” or state total ticket volume, so the base the percentage is computed against is unclear.

Reporting on Rocket Companies’ Q2 2026 results (August 7, 2026), coverage said the company’s AI Voice platform handled more than 1 million inbound mortgage-servicing calls within three months of launch, that more than half of those calls would otherwise have needed a human servicing-team member, that task resolution ran nearly 25 percent faster than traditional IVR, and that client satisfaction on the system reached 4.5 out of 5. These are company-reported figures relayed through earnings coverage rather than independently audited, and the article does not state how “client satisfaction” was surveyed or over what sample.

On its Q2 2026 earnings call, specialty insurance broker Ryan Specialty’s CEO Tim Turner said the company’s reinsurance FAC workbench now turns a submission into a priced, decision-ready file in minutes rather than days, and that the firm is extending agentic AI into its property inspection process, which he said is “reducing cycle times by removing the need for thousands of manual touch points each month.” These are executive statements on an earnings call, not independently measured or audited figures, and no baseline cycle time or specific percentage reduction was disclosed. A Wayback Machine save request for the source transcript returned an error on this run; the quote is confirmed live but not yet independently archived.

The Salamanca City Central School District in western New York approved a nearly $60,000 purchase of a Realbotix humanoid AI teaching assistant, nicknamed Sally and running Realbotix’s Optio AI software, for a single classroom. The rollout is now on hold: New York’s state education commissioner raised student-data-privacy concerns, and parents and teachers objected after learning Realbotix is a sister company to RealDoll, maker of AI-powered sex dolls. Education Week reports the district is working through “enhanced student data privacy agreements” with state officials before any deployment resumes. This is a walk-back, not a deployment outcome: no classroom use of the robot has yet occurred, and the figure here is a procurement cost, not a performance result.

Shopify’s Q2 2026 results, as covered by this report on the company’s investor slides, show its AI merchant assistant Sidekick recorded 34 million conversations in the quarter, with daily active merchant usage of the assistant up 3.6 times year over year. Shopify also reported AI-driven storefront traffic and orders each roughly tripling year over year in the same period. These are company-disclosed usage figures from investor materials, not independently audited, and the article does not break out how much of the AI-attributed traffic or order growth is incremental versus AI simply being credited as a new referral category.

Travel-search company trivago disclosed on its Q2 2026 earnings call that 93% of its staff, which the company calls “talents,” now use AI daily, saving an average of 55 minutes per day each, according to this coverage of the call. This is a company-reported internal productivity metric, not independently measured, and the article does not say how the 55-minute figure was calculated (self-reported survey versus tool telemetry) or what tasks the time savings apply to. A Wayback Machine save request for the source page returned an error on this run and no prior capture exists in the CDX index; the quote is confirmed live but not yet independently archived.

On its Q2 2026 earnings call, Wayfair said it is using a proprietary AI pipeline for product imagery that reduced traditional production costs by more than 99%, from $2 million to under $10,000, per this reporting of the call’s highlights. The figure is a company-disclosed comparison on an earnings call, not an independently audited cost study, and the article does not specify the production volume (number of images or projects) the $2 million versus under-$10,000 comparison covers.

On its Q2 2026 earnings call, staffing giant Adecco Group said it hit its year-end target early: 50% of group revenue is now “enabled by agents,” with the bar raised to 70% by the end of 2026. Management tied the rollout to a 10% improvement in overall fill rate and a 40% cut in time-to-submit, plus a 25-35% productivity benefit from recruiter time saved, on agents that have completed 2.2 million conversations to date. The figures are company self-reported on an earnings call and summarized here via an independent financial-press recap of that call rather than a direct company transcript; no third party has audited the fill-rate or time-to-submit numbers, and Adecco has a fixed-cost contract with its AI provider that keeps its own AI spend from scaling with volume.

Absa says deploying an AI and optical character recognition (OCR) tool in its South African debt review team produced a 41% gain in document indexing efficiency and now gives a one-day turnaround on newly received documents and emails, according to an iafrica.com report dated 2026-08-06. Kendrick Chauke, Absa’s National Manager for Debt Review, said: “The introduction of an AI/OCR solution in the debt review team has enabled us to automate manual processes and improve operational efficiency,” adding that employee concerns about role impact were addressed through change management. Robert Benvenuti, Absa’s CIO for Data and Applied AI, called it a step forward in using AI for customer and operational outcomes. This is a single company-sourced report with no independent press corroboration found; the 41% figure and its baseline are Absa’s own.

Related case files

None yet; no Absa Group case file exists in the corpus.

On Airbnb’s Q2 2026 earnings call, CFO Ellie Mertz said the company’s AI assistant now resolves roughly 45 percent of support issues without a human agent, and that customer support cost per booking fell 16 percent year over year, crediting AI investment as a driver of the decline. Both figures are company self-reported on an earnings call, not independently measured, and no breakdown separates the AI assistant’s contribution to the cost decline from other efficiency work in the same period.

The same FreightWaves awards writeup names BASF as an enterprise customer of Fixefy, an autonomous freight-audit AI agent that checks every shipment charge against contracts and rates and automatically files carrier disputes. Fixefy is credited with recovering 5% to 12% of total outsourced supply-chain and freight expenditure with 100% audit coverage for enterprises including BASF. The figure is vendor-supplied (Fixefy) and relayed by one trade outlet; BASF itself is not quoted directly confirming the range.

Booking Holdings has deployed voice AI across the majority of eligible inbound traveler calls, per PYMNTS’ reporting on the company’s Q2 2026 earnings. Despite that investment, traffic and bookings arriving via AI chatbots and LLM referrals remain below 1% of room nights. This is an honest, mixed result on the beat: heavy AI investment in service operations, but AI-agent booking traffic has not become a meaningful channel yet. Sourced to one independent outlet’s earnings coverage; PYMNTS does not specify exactly how the sub-1% figure was measured or over what window.

In the same FreightWaves awards coverage, Fullbay’s fleet-maintenance AI subsidiary Pitstop, which claims 95.5% accuracy predicting failures from live sensor data, is credited with helping the City of Long Beach’s municipal fleet operation save more than $800,000 a year. The figure is vendor (Fullbay/Pitstop)-supplied and reported by one trade outlet, not independently verified by the city, and no breakdown of how the savings figure was calculated is given.

DBS Bank said its generative-AI virtual assistants, DBS Joy for corporate and SME customers and DBS digibot for individuals, now serve more than 10 million customers across Singapore, Hong Kong and Taiwan and have gone agentic in Singapore. In H1 2026, active users of Joy rose 61% year on year, which the bank says contributed to a 7% cut in calls or emails to customer service. This is a company newsroom release (tier 3), not independently audited, and DBS does not break out how much of the call reduction is attributable to the assistant versus other factors.

On Etsy’s Q2 2026 earnings call (Aug 6), reported by PYMNTS, the company said traffic arriving through AI shopping agent platforms remains under 1 percent of its total site traffic, even as that sliver of traffic converts at a higher rate and larger average order size than the rest of the site. This is a useful honest counterweight to agentic-commerce hype: a named marketplace disclosing that AI agent referrals are still a rounding error on overall volume, with the conversion-rate figures given only in relative terms, not exact percentages.

Singapore’s National University Health System unveiled NCAIP, a four-tool pharmacy AI platform (Admission/Discharge MedRecon, MedTriage, MedVerify) built over nine months on top of five years of underlying data-plumbing work. GovInsider reports the platform is expected to cut in-person dispensing time by up to 20% and save as much as S$2 million a year. These are the health system’s own projections, not yet a realized, audited result, since full four-tool rollout still needs funding and is not complete. Elsewhere, coverage of the same launch (Healthcare IT News) cites an 850-pharmacist-hours-per-week estimate, but that page returned a bot-challenge wall on live re-check, so only the GovInsider-confirmed figures are used here.

FreightWaves’ 2026 AI Excellence in Supply Chain Awards coverage names Penske Logistics as a customer of Augment’s AI teammate “Augie,” which automates load-status validation across phone, email and shipper portals. Penske is using Augie to validate the status of an estimated 600,000 loads and anticipates productivity gains of 30% to 40%. The figure originates with the vendor (Augment) and is relayed by one trade outlet in an awards writeup, not independently audited by Penske, and the productivity gain is described as anticipated rather than already realized.

On its Q2 2026 earnings call, SPS Commerce CEO Chad Collins said the company completed its first fully agentic customer onboarding, automating pre-sale contacts and account provisioning for one simpler case. The Motley Fool’s own writeup of the call describes this as reducing account provisioning and pre-sale contacts “from days to minutes.” Collins said the company is next working toward more complex onboarding involving ERP integrations. This is a single, company-disclosed example on an earnings call, not a portfolio-wide average, and the day-to-minute figure is qualitative rather than a precise metric.

The same FreightWaves awards writeup credits Summit Materials, a construction-materials manufacturer, with saving $2,000 per vehicle per year on its fleet using Pitstop’s AI predictive-maintenance agents. This is a vendor-supplied case-study figure relayed by one trade outlet, not independently audited by Summit Materials, and the same article bundles this figure with two other companies’ results in a single sentence.

An eight-bank Taiwanese anti-fraud alliance using federated-learning AI, which shares model parameters across banks without exchanging customer data, reported first-half 2026 results: Taiwan Cooperative Bank’s warning account detection precision rose 67 percent, KGI Bank’s precision surged 5.6 times, and Taipei Fubon Bank intercepted 1,614 fraud cases worth over NT$272 million (about $8.4 million). BigGo Finance’s coverage synthesizes disclosures from multiple banks and Taiwan’s Financial Supervisory Commission rather than a single press release; the underlying bank-level disclosures were not independently re-verified here, and precision gains are reported without a shared baseline across the three banks.

Absa Group’s South African debt review team deployed AI and OCR to automate extraction and capture of information from standard documents, replacing manual data entry. Since implementation in late June 2026, Absa reports a 41% increase in document indexing efficiency and a one-day turnaround on all newly received documents and emails, down from what press coverage describes as a multi-day manual process. The figures come from Absa itself, quoted from national debt-review manager Kendrick Chauke and CIO for data and applied AI Robert Benvenuti, and were reported near-identically across several South African tech outlets (ITWeb, IT-Online, IT News Africa) on the same day, all tracing to the same company announcement rather than independent measurement. No third-party audit of the 41% figure was found. Absa frames the deployment as removing repetitive tasks rather than cutting headcount, with staff redirected to judgement-based work.

Akido Health, which describes ScopeAI as an “AI-native” primary-care platform, says the tool deployed via iPads at Los Angeles street-medicine encampment clinics achieved a 55% reduction in Emergency Department utilization compared with national rates, alongside 53% same-day care access. The release also cites 83% three-month and 63% six-month patient retention, and says the platform has been tested against 203,000 clinical encounters across a population of 1 million patients in California, New York City and Rhode Island. This is a single company press release with no independent measurement found; the comparison baseline (“national rates”) and the specific patient population behind the 55% figure are not detailed in the release.

Global freight brokerage C.H. Robinson has scaled more than 30 AI agents that handle quoting, order-setting, appointment scheduling and shipment tracking, according to Fortune’s reporting on CEO Dave Bozeman. Fortune reports a 45% uplift in employee productivity since 2022 attributed to the AI push, and cites agents cutting quote delivery from a 20-minute human task to 31 seconds. Fortune also quotes the company as getting “hundreds of millions of dollars of benefit with a token cost of less than $2 million,” a company-originated figure relayed by the reporter, and notes the company has not needed to backfill roles despite 11% to 14% annual attrition. Independent press coverage (tier 2), but the productivity and cost-benefit figures trace back to company statements rather than an outside audit.

ConnectOne Bank, a New Jersey commercial bank, put nCino’s agentic operating system into production for commercial lending, including a “Banking Advisor” knowledge tool and two custom AI agents covering 16 banker-facing skills. The joint release states document search, previously a 20-minute task, now takes as little as 30 seconds, a 97.5% reduction, and that active users of the tools grew 41% in 10 weeks. This is a joint vendor/deployer press release, not independent measurement, and no third-party outlet has yet corroborated the figures.

CVS Health’s second-quarter 2026 earnings exhibit, filed with the SEC on August 5, 2026, states that Aetna’s second-generation Claims Assist Manager (CAM), an AI-powered claims platform, cuts processing time by more than 20% on complex claims that require manual review. The figure comes from a primary regulatory filing, not a press release or trade write-up, but it is self-reported by the company with no independent audit of the underlying claims data, and the filing gives no baseline processing time or sample size behind the 20% figure. This clears the radar’s 30-day freshness window by exactly one day.

Related case files

Omega Healthcare and UiPath: document automation across billing, coding, and payer correspondence is the closest existing case file, covering AI in claims-adjacent healthcare document processing, though for a different company and vendor.

Fox ESS, a solar inverter and battery-storage manufacturer, says its new 46,000-square-metre smart factory in Wenzhou, China, which opened in June 2026, runs at over 60 percent automation and produces a battery pack every seven seconds. The company states the facility “has increased production efficiency by 66.7%,” but the release does not specify the baseline being compared against. This is a company press release distributed through PR Newswire, not an independently measured figure. Save Page Now could not archive this PR Newswire page on repeated attempts and no prior capture exists, an open archiving gap.

Kenco, a North American third-party logistics provider, put its first six AI agents from startup DeepFabric into live production in three months, automating bids, audits and exception handling. COO David Caines said the rollout did not disrupt service to any customer, and the company is on a path to 20 agents across the enterprise over the next year. This quote comes from a company and vendor joint announcement syndicated through trade press; no independent measurement of the deployment was found, and the 20-agent figure is a forward plan, not a completed outcome.

MakeMyTrip co-founder and group CEO Rajesh Magow told investors on the company’s Q1 FY2027 earnings call (dated August 5, 2026) that “our AI-powered customer support bot independently resolves over 50% of customer calls,” framing it as a result of the company’s AI-first transformation push. This is a first-party figure from a company executive on an earnings call, with no independent measurement of call volume, resolution quality, or customer satisfaction cited alongside it; no baseline (pre-AI resolution rate) was given for comparison.

PG&E CEO Patricia Poppe told analysts on the Q2 2026 earnings call (July 23) that the utility’s Continuous Monitoring Center, which uses machine learning on smart-meter and sensor signals to detect faults on residential service lines before they cause outages or ignitions, has avoided nearly 20 million outage minutes, 28 high-fire-risk-area ignitions and over 5,000 emergency response hours while saving more than $11 million in lower-cost repairs, all since January 2025. An earlier May 1, 2026 company press release put the same program’s cumulative total at 12 million outage minutes avoided at that point, so the July figure is a later, larger cumulative snapshot of the same rolling metric, not a new period. Both sources are PG&E’s own disclosures (earnings call and press release); no independent measurement of these figures was found.

WellSky, the home-health software vendor, published an analysis comparing clinicians using its AI-powered ambient-listening Scribe solution against peers at the same agencies who do not use it: users cared for 8.6 more unique patients per clinician per month and spent 22% less time on after-hours documentation. The comparison is a WellSky-commissioned analysis of its own customer base, not an independent or peer-reviewed study, and no agency names or sample sizes are disclosed in the release.

Related case files

None yet; no home-health or WellSky case file exists in the corpus.

ADNOC and oilfield-services firm SLB rolled out an AI-enabled Real-Time Operations Center, built on SLB’s DrillOps platform, across more than 120 of ADNOC’s onshore and offshore drilling rigs. The companies say the platform cuts engineering effort by 30 to 40 percent and lets each engineer oversee two to three times more rigs, with reporting cycles that used to take days now completed in hours and incident response times cut by 4 to 12 hours. All figures originate from the joint ADNOC and SLB announcement; there is no independent audit of the numbers, and The National’s independent report restates the same figures rather than measuring them separately. Save Page Now could not archive ADNOC’s own press release (no capture registered on repeated attempts); the independent-press citation to The National is archived.

Baldwin Insurance Group CEO Trevor Baldwin told investors on the Q2 2026 earnings call that AI deployed into the company’s direct-bill processing raised monthly reconciliation from roughly 90% to a sustained 98%, cut run-rate costs from $3 million to $1 million a year, and reduced internal labor cost on the process from $1.2 million to about $400,000. Over 47,000 tasks were completed by the system in the prior 17 weeks at a 98%+ quality rate, rising above 99% in recent weeks. Baldwin described the deployment as running at production scale rather than pilot scale; the figures are company-disclosed on a public earnings call, not independently audited.

On its Q2 2026 earnings call, Booking Holdings said voice AI is now scaled across the majority of eligible inbound traveler calls and is contributing to a double-digit year-over-year decline in customer service cost per booking. Separately, management raised its expected annual run-rate savings from its broader transformation program to approximately $650 million, up from a prior $550 million target, with CFO Ewout Steenbergen attributing most of the incremental $100 million to procurement efficiencies rather than AI specifically. Both figures are company self-reported on an earnings call; the article does not break out how much of the customer-service cost decline is attributable to voice AI versus other factors. This is a separate claim from Booking Holdings’ already-covered sub-1%-of-room-nights LLM-referral-traffic figure from the same call (radar/2026-08-06).

Finnish telecom operator Elisa said agentic AI, layered on a network digital twin and running alongside human engineers, has reduced customer-impacting network incidents by more than 80% and increased preventive actions to 99%. The figures trace to an RCR Wireless News report, republished here by DROAM; the RCR Wireless original was paywalled from this session and could not be independently re-fetched, so this is treated as single-source pending a second newsroom carrying the same figures, and tiered down accordingly.

Lemonade president and co-founder Shai Wininger told investors the insurer delivered its “best ever LAE ratio result of 5%” in Q2 2026, and accompanying coverage of the same results ties that record to AI now automating more than 50% of claims. The 5% figure is a direct company disclosure on the earnings call; the “more than 50%” automation share comes from a secondary summary of the same results rather than the call transcript itself, so it is flagged as slightly less precise than the LAE figure.

South Africa’s Nedbank says its Nedbank Intelligent Hyper Automation (NIHA) programme, which combines AI, generative AI, analytics, machine learning and robotic process automation, has “already unlocked more than R375 million in annualised benefits” (about $21 million), reported by ITWeb from the bank’s H1 2026 results on 2026-08-04. On the same day’s earnings call, Chief Operating Officer Mfundo Nkuhlu said AI-driven next-best-offer capabilities in Personal and Private Banking (PPB) increased “from 5% to 13% of sales, contributing to the improved cross-sell ratio in PPB.” Both figures are company-disclosed and self-measured; no independent audit of the R375 million total or the underlying attribution methodology was found.

Related case files

None yet; no Nedbank case file exists in the corpus.

RWJBarnabas Health and Rutgers Robert Wood Johnson Medical School researchers found that an AI-enabled early warning system, built on the Epic Deterioration Index, was associated with an 18% reduction in the risk-adjusted odds of in-hospital death among high-risk patients, with observed mortality falling from 23.1% to 18.6%. Rapid response team activations rose from 25.3% to 37.5% of high-risk hospital stays after rollout, consistent with earlier intervention. The finding comes from a peer-reviewed study published in NEJM AI and a joint press release; the figures were also independently repeated by ROI-NJ, NJBIZ, Medical Xpress and other outlets, but this flash cites the joint release as the primary source.

Teleperformance CEO Jorge Amar told investors on the Q2 2026 earnings call that the company is raising its internal savings target from over EUR 100 million to a range of EUR 150 to 170 million, attributing the gains to SG&A control and an internal AI efficiencies program applied to functions including recruiting, quality assurance and workforce management. This is a company-set forward target based on savings already realized this year, not an independently measured historical outcome, and no separate breakdown of the AI-attributable share of the total was given.

XPO CEO Mario Harik told investors on the company’s Q2 2026 earnings call that an AI trailer-loading tool, in which dock workers photograph each tier of a trailer as they load it and get real-time feedback from the AI, improved load quality by more than 40% and cut damages by 50% at pilot sites. The figures are company-disclosed on a public earnings call rather than independently audited, and the scope (how many docks or lanes) was not specified in the quoted remarks.

Commonwealth Bank of Australia’s customer-service chat platform was resolving nearly nine in ten conversations without any human agent involvement as of May 2026, according to bank executive Rachel Round, cited by both HCA Magazine and Information Age. The rollout has coincided with job losses: Australia’s Finance Sector Union documented 276 redundancies in July 2026 across technology, operations and HR, and close to 800 roles cut across the year. The nine-in-ten figure is a company executive’s own disclosure, not an independently audited measurement, though it is consistently reported across two independent trade outlets.

Delaware County commissioners in Indiana approved an $85,000-a-year contract with AI vendor Aurelian to answer non-emergency calls at the county’s 911 dispatch center, funded from a statewide 911 fund rather than the county budget. The center fields more than 100,000 non-emergency calls a year, and a report cited in the story says the county’s 911 center needs 11 more employees to keep up with that volume without AI help. The 911 director cited a neighboring county’s account that it cut 80 percent of nuisance fireworks calls before they reached a live dispatcher, a secondhand figure attributed to a conversation, not an independently published measurement. The contract is pending legal review before installation, so this is a signed commitment, not yet a measured deployment outcome.

On Gallagher’s Q2 2026 earnings call, Chairman and CEO J. Patrick Gallagher Jr. said the company’s AI fraud detection capabilities, used within its Gallagher Bassett claims unit, saved one client $100 million, describing the figure as “auditable numbers.” No independent audit or methodology has been published; the figure is a company-reported claim on an earnings call, and the transcript does not name the client, specify a time window, or describe what “auditable” means in this context. The quote itself carries a mid-sentence restart in the original transcript, reproduced verbatim here.

Somerset NHS Foundation Trust reports that CompliMind, a Cambridge-founded compliance-AI platform, cut the staff time spent searching for building and estates compliance information by 35%, freeing more than 400 staff hours a month at a single site. The figure is CompliMind’s own claim about its client, relayed by two independent trade outlets (dtawards.ie and resultsense.com) using near-identical wording, rather than an NHS trust statement or independently audited measurement; no baseline hours or absolute search-volume figures are given, only the percentage and monthly-hours change. A parallel, independent business-press piece (businesscloud.co.uk, 29 Jul 2026) confirms Somerset NHS Foundation Trust as a named CompliMind client among six named health bodies but does not itself carry the quantified figures, so the outcome numbers rest on the vendor-relay coverage only. This is an operations/estates-compliance deployment, not a clinical one.

Related coverage

CompliMind’s expansion to 15 NHS trusts (announced ~29 Jul-3 Aug 2026) also names University Hospital Southampton Estates Limited, flashed separately today (02-university-hospital-southampton-complimind-contract-savings.md), and Swansea Bay University Health Board, Belfast Health and Social Care Trust, Cambridge University Hospitals and Liverpool University Hospitals as clients, without further quantified figures found this run.

CompliMind’s compliance-AI platform surfaced a contract problem at University Hospital Southampton’s estates arm (University Hospital Southampton Estates Limited) that had reportedly gone unnoticed for eight years and was costing the trust £45,000 annually, according to two independent trade write-ups of CompliMind’s expansion to 15 NHS trusts. As with the Somerset flash filed alongside this one, the figure is the vendor’s own claim about its client’s result, not a trust statement or an independently audited finding, and no detail is given on what the contract issue was or how the saving was confirmed. An independent business-press piece (businesscloud.co.uk, 29 Jul 2026) separately confirms University Hospital Southampton Estates Limited as a named CompliMind client without itself carrying this figure.

Related coverage

See 01-somerset-nhs-complimind-compliance-hours.md for the Somerset NHS Foundation Trust figure from the same CompliMind expansion announcement.

On its Q2 2026 earnings call, insurance broker WTW disclosed a $625 million AI and automation investment plan called Propel, targeting $400 million in gross run-rate savings by the end of 2028. Among the cited examples: generating a schedule of insurance that once took four hours is now done in five minutes. This is a company-disclosed figure from an earnings call, not independently measured, and the $400 million is a forward target rather than a realized result.

Google rolled back its newly launched Nano Banana-based AI image generator inside Google Earth worldwide less than 24 hours after it went live, after an investigative journalist used it to fabricate convincing satellite imagery, including a nonexistent nuclear plant in Iran, that defeated the built-in SynthID watermark check simply by screenshotting the result. Google’s statement, quoted by both outlets, said it was rolling the feature back “while we work on implementing stronger guardrails.” This is a self-initiated walk-back within a day of launch, not a regulator action; Google says the AI-generated images never appeared in the main Google Earth experience for other users and were watermarked at creation.

Google switched on an AI image-generation feature inside Google Earth’s web version on July 30, 2026, letting anyone type a prompt and get a photorealistic image anchored to real satellite, aerial, and 3D imagery of any coordinate on Earth. It pulled the feature worldwide less than 24 hours later after journalists and researchers, including Henk van Ess and outlets like NPR and Futurism, showed it could generate a fabricated nuclear power plant in Iran and misleading refugee imagery near the US-Mexico border. Google’s own explanation: “We’ve seen people sharing screenshots of generated imagery that appear to violate our policies. So we’re rolling back this feature in Google Earth while we work on implementing stronger guardrails.” No figure on volume of generated images is available; this is a walk-back, not an outcome metric, included under the beat’s walk-backs-qualify rule.

Grab’s engineering blog reports that AI agents cut the share of mechanical tickets (data preparation, alerting, reporting) its analysts closed from 44% in February 2026 to 30% by June 2026, with cycle times down about 33%. Self-serve resolution without human involvement also rose over the same window: 53% to 67% for metric questions, 63% to 90% for data pulls, and 50% to 81% for SQL requests. The figures are first-party (Grab’s own engineering blog); InfoQ independently republished and confirmed the same 44%-to-30% figure on 2026-08-17, so this is flagged single_source: false, but both trace to the same underlying Grab data, not a separately audited measurement.

Allianz Partners CEO Tomas Kunzmann confirmed the travel-insurance arm will cut up to 1,800 roles over 12 to 18 months as AI takes over call-center and claims-processing work, per a Reuters wire story picked up by Yahoo Finance and corroborated by multiple other outlets. This is a headcount-reduction disclosure from the company itself, not an independently audited efficiency study, and no separate error-rate or throughput figure has been published to explain how the AI performs the displaced work.

Atlas Energy Solutions runs a 28-truck fleet of Kodiak AI-enabled driverless trucks hauling frac sand in the Permian Basin, work that won FreightWaves’ 2026 AI Excellence in Supply Chain Award. Energy Tech reports the fleet logged more than 23,500 hours in Q1 2026, a 110 percent jump from 10,700 hours in Q4 2025, delivering roughly 450,000 tons of sand on 15,000-plus loads, and cut diesel consumption by 7 million gallons over 18 months. Both the throughput and fuel-savings figures are reported by the trade press citing Atlas and Kodiak, so they are company-sourced, though corroborated across multiple outlets and consistent with Atlas’s own investor materials.

Chime is cutting about 150 jobs, roughly 10% of its workforce, which CEO Chris Britt tied to AI efficiencies and a shift to smaller, flatter teams. Britt told staff “AI is changing what’s possible but requires new skills” and that “smaller teams with fewer layers are moving faster than ever and getting more done.” The figure and the AI rationale both come from the company’s own internal memo as reported by Banking Dive; no independent measure of AI’s specific contribution to the headcount decision was found.

Deutsche Telekom, with over 200,000 employees and 300 million-plus customers, rolled out ChatGPT Enterprise across customer care, network operations, and internal workflows. Telecoms Tech News reports more than 50,000 monthly active users of ChatGPT and API tooling, and a 546 percent jump in internal AI tool usage since the start of 2026. The usage figures come from Deutsche Telekom’s own reporting on the rollout; no independent measurement of downstream cost or productivity impact has been published yet.

LinkedIn deployed AI-based detection against automated low-quality content and, according to an executive quoted by Fortune, now detects and blocks hundreds of thousands of automated slop-comment attempts every day, and has prevented billions of other automation attempts (mass posting, slop) in the last couple of months alone. LinkedIn also rolled out a new “seems like AI slop” button letting members flag posts. Separately, an outside study by AI-detection firm Pangram found 40 percent of long-form posts and 30 percent of short-form posts on LinkedIn were flagged as fully AI-generated; that estimate is Pangram’s own measurement, not LinkedIn’s count, and the two figures measure different things (LinkedIn’s automated-attempt blocking versus Pangram’s content-generation estimate). This is a single-source item: only Fortune’s report of the executive’s on-record figures was found, though LinkedIn is the deployer speaking on the record, not an anonymous vendor claim.

Mount Sinai Medical Center in Miami Beach became the first health system in Florida, and fourth nationally, to extend Epic’s “Chart with Art” ambient AI documentation tool to inpatient nursing staff, after a physician pilot of the same tool reduced chart time per encounter by nearly 30 percent and cut after-hours documentation. Only one outlet, HIT Consultant, has covered this so far; the figure is attributed to the hospital’s own pilot results, not an outside audit.

OCBC’s private bank, Bank of Singapore, rolled out an agentic AI platform called HELIOS for wealth-client onboarding and source-of-wealth due diligence, piloted across Singapore, Hong Kong, and Dubai. The Asian Banker reports account opening now takes 15 business days, half the roughly six-week industry median, and that relationship managers cut source-of-wealth report preparation from 10 days to one hour. Private Banker International separately reported on the same rollout with matching figures. The bank is the source of both figures; no independent audit of the onboarding-time reduction has been published.

Revolut built a proprietary AI foundation model, PRAGMA, with Nvidia, now running across fraud detection, credit risk scoring, product recommendations, and customer support for its 70 million-plus customers. Forbes reports a 64.7 percent improvement in fraud detection and a customer-service AI assistant that “currently handles 75 percent of support requests without human intervention.” Nvidia’s own case study corroborates the fraud figure but frames it slightly differently, as a 64.7 percent lift in fraud recall plus a 16.7 percent lift in precision. The underlying performance numbers come from Revolut and its model-building partner Nvidia, not an independent auditor.

Snap announced that wholly AI-generated videos will no longer be eligible for recommendation on its Spotlight feed, prioritizing human-made content while still allowing AI-enhanced or edited videos. Snap’s own newsroom post says unique Spotlight contributors globally are up more than 120 percent compared to last year, which it cites as evidence the platform is drawing more human creators; that figure is Snap’s own reported metric, not independently verified, and the post does not say how much of that growth, if any, is attributable to this policy change versus other factors. TechCrunch independently confirmed the policy change itself. This is a self-initiated content-policy tightening, not a regulator action.

Syngenta says it now runs around 50 predictive AI models to optimize roughly 15 parameters at once during crop-protection product development, drawing on some 500,000 field trials dating back to the 1970s. A company representative told AgNavigator that molecule optimization work “previously required years of laboratory and field work” and “can now be reduced to months.” The timeline claim is attributed to a Syngenta spokesperson, not an independent benchmark, and only one outlet has carried the interview so far, though a second publication (AgroSpectrum India) appears to cover the same underlying event.

WellSpan Health expanded its partnership with Hippocratic AI platform-wide, moving its generative AI voice agent “Ana” from inbound call handling into outbound and clinical triage use. HIT Consultant reports Ana now handles more than 160,000 patient calls and 7,000 conversational hours a month for appointment management and digital access queries. The rollout was also picked up via GlobeNewswire and Manila Times syndication with matching figures, but all coverage traces back to WellSpan’s and Hippocratic AI’s own announcement; no independent call-outcome data has been published.

On ADT’s Q2 2026 earnings call (July 30, 2026), CEO Jim DeVries said the company combined AI-driven call routing with virtual AI agents during the quarter, resulting in nearly 20 percent fewer customer contacts handled by human agents and a similar reduction in service tickets, alongside improved customer satisfaction. The same coverage noted more than three-quarters of code generated by ADT’s product software team was AI-generated and accepted, without a separate sourced quote for that figure. The 20 percent figure is self-reported by the CEO on the earnings call, not independently measured, and the article does not state the comparison window (quarter-over-quarter or year-over-year).

Medicare Advantage insurer Alignment Healthcare’s CEO John Kao told investors on the Q2 2026 earnings call that the company’s AVA AI stratification model now identifies the 10% of members who account for nearly 70% of hospital admissions in the following 30 days. The claim was made directly to investors on a recorded earnings call, which carries disclosure weight, but it is still a company-reported accuracy figure with no independent audit or named validation methodology given in the call.

Furniture retailer Branch used SPS Commerce’s Max AI agent to manage overdue orders with one of its key retail partners, cutting the weekly time spent on that task by 90 percent according to SPS Commerce CEO Chad Collins on the vendor’s Q2 2026 earnings call. Same call, same caveat as the Owlet item published this run: vendor-originated customer example, not independently confirmed, and the “1 of their key retail partners” is unnamed.

China Energy, a Chinese state-owned energy group, says it has deployed more than 100 AI agents across its coal, power generation, chemicals, transport and hydrogen businesses, built on its in-house “Qingyuan” large language model. In power generation alone, 41 business-specific agents have launched with 25 use cases approved for group-wide rollout. Named per-site figures cited: “a 6.7% productivity increase at a fully automated coal mine” and “a 91.8% reduction in daily manual tasks at a large coal-fired power plant.” Single source found, and it is Chinese state media (China News Service, ecns.cn) reporting on a state-owned company’s own figures, not an independent measurement; treat the percentages as company-claimed, not audited.

DBS Bank told Reuters, in an emailed statement reported 30 July 2026, that AI-assisted onboarding cut turnaround time for new high-net-worth and ultra-high-net-worth wealth clients by 50% and helped the bank sign up 20% more such clients in the first five months of 2026, versus the same period prior. The push followed a May 2026 instruction from Singapore’s Monetary Authority of Singapore asking private banks to shorten the time clients wait to open accounts. The figures are DBS’s own, given by an unnamed spokesperson rather than a named executive, and no independent audit of the 50%/20% numbers was found; multiple outlets (Private Banker International, Global Business Outlook, The Digital Banker) carried the same Reuters-sourced figures within a day of each other. DBS separately told the same outlets it can complete onboarding within a week depending on client profile, without giving a prior baseline for that claim.

Baby-tech retailer Owlet used SPS Commerce’s Max AI agent to chase down retail deduction disputes and recovered 1.4 million dollars within six months, including full recovery on one 423,000 dollar settlement. The figure comes from SPS Commerce CEO Chad Collins on the vendor’s own earnings call as a customer success story, not from Owlet directly or an independent source, so treat it as vendor-originated and unaudited.

On its Q2 2026 earnings call (July 30, 2026), Yum! Brands said corporate employees’ daily usage of internal AI productivity tools is up more than 50 percent year over year, and that teams have built more than 400 specialized internal AI agents to handle business tasks. The same call put Taco Bell’s Voice AI drive-thru footprint at more than 900 US restaurants, up from the 890-plus figure the company gave in early July 2026 (already covered in radar/2026-07-22). Both figures are company self-reported on an earnings call, not independently audited; the article does not break out what the 400 agents do individually or how “usage” is measured (logins, prompts, or something else).

Hong Kong jeweler Chow Tai Fook has deployed more than 400 AI agents, branded AI Fook, to support more than 24,000 employees, letting sales staff pull product details and inventory checks mid-conversation. The company credits the rollout with a gain of more than 70 percent in process efficiency. PYMNTS also references a sales-conversion lift figure sourced to a Microsoft retrospective, but the actual percentage is missing from the live article text, so it is omitted here rather than guessed. The efficiency figure is company-credited, not independently measured.

Digital Insurance reports Frontline Insurance’s voice AI system now autonomously resolves 70 percent of first-notice-of-loss calls, with roughly one-second response times, as part of a trade feature on which AI insurance deployments are producing measurable results. The figure is attributed to Frontline in an independent trade outlet, not a vendor release, but only one source carries it and the article does not specify Frontline’s measurement window or methodology.

On PSQ Holdings’ Q2 2026 earnings call, CEO Dusty Wunderlich said AI has significantly contributed to operational efficiency, allowing revenue per employee to grow by 300%, per Motley Fool/Yahoo Finance’s transcript coverage. The same earnings disclosed separate opex cuts of roughly 12% and a headcount reduction near 50%, but the source does not directly attribute those cuts to AI, so they are noted here as context, not merged into the AI figure.

A GPTZero investigation, verified by the Financial Times and reported here by the Irish Times, found four PwC Middle East thought-leadership reports published over the past two years with text apparently written with heavy AI help, including at least one academic paper on Riyadh air quality that appears to have been hallucinated entirely, with no trace of it in the cited journal. PwC Middle East’s response is to update a limited number of supporting citations rather than retract the reports, a narrower response than the full retractions EY and KPMG issued after similar findings. This is a walk-back/failure story: an AI-drafting shortcut in corporate research output produced fabricated sourcing that survived publication.

On Wingstop’s Q2 2026 earnings call (July 29, 2026), President and CEO Michael Skipworth said the chain’s AI-driven Smart Kitchen back-of-house platform, now operationalized system-wide, lifted digital guest satisfaction at its historically lower-performing restaurants by more than 11 percentage points, cutting the performance gap between weak and strong restaurants by more than 40%. This follows a same-store sales decline reported earlier this year; Wingstop is leaning on Smart Kitchen, loyalty, and marketing to reverse it. The figures are company self-reported from an earnings call with no independent audit; a specific ticket-time reduction figure Wingstop cited in 2025 (40% faster tickets in early Smart Kitchen stores) was not repeated on this call and is now outside the 30-day freshness window, so it was not used here.

Wyndham CEO Geoff Ballotti disclosed on the company’s July 23, 2026 earnings call that its AI guest-messaging product, Wyndham Connect, is now used at more than 5,000 hotels, and that reservations completed autonomously carry average daily rates 15 percent higher than phone-booked reservations. He also said highly engaged franchisees have generated $100,000 or more in additional revenue from upsells. These are CEO-disclosed figures on an earnings call, explicitly flagged by the trade outlet as not independently verified and variable by property.

Brinks Home’s CIO Philip Kolterman told Bloomberg that AI cut inbound call volume by roughly two-thirds, and the company halved its call center workforce from about 800 to 400 as a result. The figure comes directly from the company’s own CIO, quoted in a single Bloomberg piece syndicated via Yahoo Finance; no independent audit of the call-volume or headcount numbers was found. No corroborating second source located within the freshness window.

BuzzFeed laid off 180 employees, about a third of its remaining workforce, Futurism reported. The outlet frames this as the latest fallout from BuzzFeed’s 2023 pivot to AI-generated content (quizzes and later articles), which it says contributed to the company’s financial decline rather than reversing it. This is a walk-back story: the AI pivot did not save the jobs it was meant to protect.

The same Bloomberg-sourced Jul 28 report says Commonwealth Bank of Australia has shed hundreds of workers from its chat-support line as it integrated AI, saving “tens of millions of dollars” a year. This is a rounded, unaudited range rather than an exact headcount or dollar figure. CBA has separate AI stories already in the radar/research pipeline (its AI chatbot conversation share, its fraud-detection AI); this flash covers a distinct claim, the chat-support headcount and savings figure, not yet captured elsewhere.

Honolulu replaced its legacy permitting system with an AI plan-review platform called CivCheck, built by Clariti, starting August 2025, after residential permits had been averaging months amid a corruption scandal and staffing shortages. Q1 2026 data shows average permit decision times fell from 73 days to 32.5 days for smaller residential project types (single-family, duplex, ADUs, additions, renovations), with expansion planned for commercial and larger developments. Independently reported by HousingWire; the live article page returned an access block on the first WebFetch attempt but was confirmed via a direct fetch with a standard browser user agent, matching both the page’s meta description and body text.

Bloomberg reporting (Brody Ford, republished by Yahoo Finance on Jul 28) says Microsoft cut its customer service workforce, a mix of contractors and full-time staff, from about 50,000 to 40,000 and is saving roughly $750 million a year in customer service costs, attributed to Judson Althoff, who runs Microsoft’s sales and service operations. The savings figure is a company disclosure relayed through the reporter, not an independently audited number, and the article does not specify what share of the 10,000-person reduction is attrition versus direct cuts.

Bloomberg reporting (via Yahoo Finance) says Microsoft has cut its customer service workforce, a mix of contractors and full-time staff, from about 50,000 to 40,000 as AI tools absorbed more support work. The reduction happened “in recent years,” not as a single fresh event this week, so the number is a cumulative figure surfacing for the first time in this specific form rather than a new announcement. The source is a person familiar with the matter, not an on-record Microsoft statement, and no separate AI-only attribution study backs the figure.

The same Bloomberg-sourced report (Jul 28, via Yahoo Finance) says Uber cut 10 percent of jobs in its customer service operations as part of a broader push to “embrace artificial intelligence,” routing support requests through an in-app AI chatbot first. The 10 percent figure is itself attributed to an earlier Bloomberg report (“Bloomberg reported last week”), so this flash is one step removed from Uber’s own numbers; no independent confirmation from Uber or a labor filing is cited in the piece.

At UPS’s Q2 2026 earnings call, executives said 68.5% of US package volume now moves through automated buildings, up from 64% a year earlier, and that cost per piece in an automated building runs about 28% lower than in a non-automated one. The figures are self-reported on the earnings call and cover automation broadly (sorting, routing) rather than AI specifically; the call also disclosed $891M in after-tax transformation charges tied to workforce separations under UPS’s Driver Choice Program, which the source does not directly tie to the automation figure.

Visa is cutting about 2,600 jobs, roughly 7% of its global workforce, concentrated in technology and product teams. CEO Ryan McInerney told staff in a memo that “AI is also helping to accelerate this evolution and shape the way work gets done at Visa,” language multiple outlets (CNBC, Bloomberg, Fast Company) corroborated from the same memo. A person close to the decision told reporters AI was a significant factor but not the sole driver; the cuts also track a strategic pivot toward stablecoins and cross-border payments, so the AI figure is directional, not a standalone efficiency metric.

AT&T CTO Jeremy Legg, speaking at an AMD keynote, said the carrier now runs more than 100 generative AI models in production for network operations, consuming roughly a trillion tokens a month (about 45 billion a day). This is a trade-press writeup of a vendor-conference keynote quoting AT&T’s own CTO; no independent measurement of AT&T’s token volume exists, and the figures are AT&T’s self-reported infrastructure scale, not a customer or operational outcome metric.

HSBC’s Corporate and Institutional Banking unit says AI agents handling customer enquiries across products, markets, and regulatory requirements have cut resolution time by more than 30% since January 2025. The claim comes from HSBC’s own strategy page, not an independent measurement, and gives no baseline resolution time or sample size, so the scale of the underlying process is unclear.

Michaels built “Ask Mike,” an AI shopping assistant on Google Cloud’s Gemini Enterprise for Customer Experience, in six weeks and quietly launched it in May 2026. The retailer says it has logged 75,000 conversations since launch, that shoppers who use it convert at more than double the rate of those using traditional site search, and that 27% of Ask Mike interactions lead to a product-link click or add-to-cart. All figures are Michaels’ own, self-reported to press (Digiday, Digital Commerce 360) rather than independently audited; no baseline conversion rate or revenue figure was disclosed. Other retailers (Ulta Beauty, Macy’s, The Home Depot) have deployed the same Google Gemini Enterprise product for similar shopping-assistant features.

The State Department published a generative AI playbook for federal agencies built around its internal StateChat tool, which had grown to more than 62,000 users as of June 2026. In a survey cited in the playbook, 90 percent of regular users reported time savings, averaging 1.6 hours per week per task, and 96 percent said they would recommend the tool. These are self-reported survey figures from the deploying agency, not an independent time-and-motion study; the article does not state the survey’s sample size or response rate. Ops function here means internal federal-employee productivity tooling, not a public-facing citizen service.

Duke Energy said its automated self-healing grid technology, which automatically detects power outages and reroutes electricity around damaged sections of the grid, prevented more than 12,600 customer outages in Wayne County, North Carolina, during 2025. The company says more than 75 percent of Wayne County customers are now served by the technology, more than double 2022 levels. This is automated fault detection and rerouting rather than generative AI, the underlying performance data covers calendar 2025, and the figure comes from a Duke Energy press release relayed by a single local outlet, part of a recurring county-by-county PR push the utility runs across its service territory.

On American Express’s Q2 2026 earnings call, chairman and CEO Steve Squeri, answering a question about AI and operating leverage, said the company is seeing a “30%-40% decrease in cycle time from a coding perspective.” The figure is self-reported on an earnings call rather than independently measured, and Squeri framed it as a capacity gain for engineering rather than a headcount or cost reduction. No independent outlet has corroborated the specific percentage yet.

On Customers Bancorp’s Q2 2026 earnings call, CEO Sam Sidhu said the bank’s AI-enabled workflow automation has saved at least 46,000 hours to date, up about 65 percent quarter over quarter and equivalent to 24 full-time employees. He said employees have built more than 600 internal agents and custom GPTs, up 20 percent in 60 days, and that 100 percent of staff are now AI licensed, up from 75 percent last quarter. These are company-reported figures stated on the earnings call, not independently audited, and no dollar-cost figure was disclosed alongside the hours-saved estimate.

On its Q2 2026 earnings call (July 24, 2026), Customers Bancorp CEO Sam Sidhu said a multi-agent AI underwriting pilot closed commercial-and-industrial and commercial-real-estate loans within a week, an 85% reduction in readiness-to-close time versus the industry norm of 30 to 60 days, and that AI-enabled workflow automation has saved at least 46,000 hours company-wide, equivalent to 24 full-time employees. Both figures are company self-reported from the same earnings call; web.archive.org could not be reached to snapshot either source in this session, a known gap in this environment, so both carry an open archived field.

Ardent Health’s rollout of Ambience Healthcare’s ambient AI documentation platform has reached 87 percent of ambulatory clinical encounters network-wide, roughly double the 40 to 45 percent industry benchmark cited in the same article. The adoption figure is reported in business press coverage of the milestone but traces back to Ardent and Ambience as the source; no third-party audit of the 87 percent figure was found.

Cigna Healthcare said AI tools used to identify patients with chronic or complex conditions earlier will save customers an estimated $200 million in medical costs over the next three years, with engaged customers saving about $2,000 per year on average. More than 1,250 Cigna clinicians, including nurses and behavioral health specialists, are using the AI-generated insights, and the company says the approach helps it reach 20 percent more customers with complex or chronic needs earlier. This is Cigna’s own forward-looking projection announced in its press release, not yet an independently audited outcome, and no methodology for the $200 million estimate is disclosed.

On Valley National Bancorp’s Q2 2026 earnings call, CFO Travis Lan said the regional bank has realized about $15 million in expense run-rate savings from AI year to date, against roughly $3 to $4 million in new AI-associated costs such as headcount and vendor spend, for a net positive of about $11 to $12 million. CEO Ira Robbins confirmed the year-to-date figure on the same call. This is a company-reported, self-attributed figure from the earnings call rather than an independent audit, and the bank did not break out which functions or processes drove the savings.

On its Q2 2026 earnings call, Waste Connections chair and CEO Ron Mittelstaedt told analysts that the company’s AI-linked commercial pricing tool, fully deployed by the fourth quarter of 2025, has produced about $20 million of EBITDA improvement on a run-rate basis through 2026. Separately the company said it began pilot testing a dynamic, real-time AI-driven routing algorithm for its truck fleet in late Q2 2026, with route-related savings of $40 million to $50 million projected for 2028-2029, but that figure is a forward-looking target, not a measured result, so it is not included as a claimed outcome here. The $20 million figure is a company (CEO) claim from the earnings call, not an independently audited number.

Yiren Digital, a China-based online consumer finance platform, says that after deploying AI agents across its asset-recovery (debt collection) workflows, service tickets handled per staff member in the Month 1 workflow rose from 358 to 525, about a 47 percent improvement, while the human handling rate on eligible Day 1 tickets fell as AI agents took on 81 percent of them, up from 50 percent in 2024. This is a first-party press release from the company itself, not an independently audited measurement, and no cost or headcount figures were disclosed alongside the productivity numbers.

On its Q2 2026 earnings call (July 22, 2026), BNY CFO Dermot William McDonogh told analysts that roughly 40 percent of the software BNY’s engineers write is now written using AI, and that the capability “is coming, it is going broad, and it is also going deep into the enterprise.” Asked in the same answer about the bank’s 7 percent year-over-year headcount decline, McDonogh was explicit that headcount is “just an output” of the broader business plan and operating-leverage decisions, “not necessarily a headcount target,” and did not attribute the decline to AI specifically. So the 40 percent code-generation figure carries a clear, on-the-record AI attribution; the headcount figure does not, despite being disclosed on the same call. This is a company-reported, first-party figure from an earnings call Q&A, not an independently audited measurement, and no second source has yet corroborated the specific 40 percent number. (The transcript page’s own auto-generated summary paraphrases this same line as “approximately 40%… generated using artificial intelligence tools”; the quote above is McDonogh’s actual words from the Q&A transcript, not that paraphrase.)

On Delta’s June-quarter 2026 earnings call (July 10, 2026), Chief Operating Officer Dan Janki said the airline’s AI Concierge digital assistant drove more than a 25-point improvement in Net Promoter Score specifically during periods of irregular operations (weather delays, cancellations, rebooking). The figure is company-reported on an earnings call, not independently audited, and the transcript does not specify the comparison baseline period or the sample size behind the NPS delta. PYMNTS’ write-up is the only source found reporting this specific quote; no independent measurement of the same NPS lift was located.

FreightWaves’ 2026 AI Excellence in Supply Chain Awards (published July 16, 2026) credited vendor Gather AI’s drone-based cycle-counting deployment at warehouse operator GEODIS with cutting manual inventory counting from 4,400 hours a year to 800. GEODIS also publishes its own case study describing the same deployment (one drone operator scanning roughly 1,200 locations a day versus four manual counters covering 800). The hours figure is company/vendor reported, not independently audited, and neither source states the specific facility or time window the 4,400-to-800 comparison covers.

Skilled Nursing News reported on July 22, 2026 that Ignite Medical Resorts, a nursing home operator, has cut the time its case managers spend reviewing referral documents and sending them to health plans from 25 minutes to 10 minutes after adopting an AI case-management platform (ExaCare AI), per Justin Meara, the operator’s senior vice president of managed care. The same platform also condenses hospital transfer packets from as many as 300 pages down to about 30-page summaries in under a minute. The honest caveat: Meara said Ignite has only been using the software for a few months and that full time and financial savings are still being estimated; the 25-to-10-minute figure is the one outcome he described as already measured, not a vendor projection.

Manulife expanded its five-year AI partnership with Microsoft, adopting Microsoft 365 E7 Frontier Suite, rolling Microsoft 365 Copilot out to more than 30,000 employees and using Microsoft Agent 365 to register and govern its AI agents. In the same announcement, Manulife says its developers using GitHub Copilot “have increased their productivity by 30 per cent,” and cites a mortgage-renewal application rebuilt with Copilot in weeks as an example. The company expects the broader AI program to generate more than $1 billion in enterprise value by 2027, with $300 million already realized as of year-end 2025. The 30 percent figure and the enterprise-value target are Manulife’s own numbers, published inside a joint Manulife-Microsoft announcement (the quote above is carried on Microsoft’s own news site), so the vendor has a direct commercial interest in the reported result; no independent third party has audited the productivity claim.

Manulife is expanding Microsoft 365 Copilot to more than 30,000 employees and says its developers have seen a 30% productivity increase using GitHub Copilot, per a joint Manulife/Microsoft press release. Manulife also reiterated a target of more than $1 billion in enterprise value from the partnership by 2027, with $300 million already realized as of year-end 2025. This is a tier-3 vendor/company release, not independent reporting; no outlet has independently verified the productivity or value figures, so treat them as company-and-vendor-originated claims.

monday.com filed a Form 6-K on July 22, 2026, disclosing a restructuring plan that cuts approximately 20 percent of its global workforce (about 630 employees) to align the company around its AI Work Platform strategy. The filing frames this as a structural reorganization tied to AI product direction, not a straight cost-cutting move: the company reaffirmed its full-year 2026 revenue outlook and raised its non-GAAP operating margin guidance from about 13 percent to about 15 percent, and says it will keep hiring in key strategic areas through 2026. The filing does not state that AI systems are directly performing the work of the roles eliminated, only that the reorganization is meant to align headcount with the AI-platform strategy; that distinction is the honest caveat here.

On Neptune Insurance’s Q2 2026 earnings call, chairman and CEO Trevor Burgess said that by the end of the quarter nearly 3,700 of the flood insurer’s agents had exchanged 33,000 messages with Atlas+, a generative-AI assistant built into its agent portal that drafts sales scripts and emails, compares coverage options and answers coverage questions in real time. Burgess described the tool as intended to make agents “super agents” rather than replace them, with a second phase (task prioritization and workflow automation) due in Q3. This is a usage figure, not an independently measured outcome such as faster quote times, higher close rates or cost savings, and it comes only from the company’s own earnings call; no outside source has corroborated the message-volume figure.

FreightWaves named trucking and logistics provider NFI an “Operational AI Integration” winner in its 2026 AI Excellence in Supply Chain Awards (published July 16, 2026), crediting agentic AI that replaced manual freight-tracking workflows with automated email triage and freight-bill exception matching. The figures (100+ hours a week reclaimed, four FTE equivalent in load tracking, 20% less training time, 75% faster appointment scheduling) are attributed to NFI in the article but originate from NFI’s own award submission, not an independent audit; FreightWaves does not restate a verification methodology.

This is a walk-back: on Otis’s Q2 2026 earnings call, chair, CEO and president Judy Marks said the company is deliberately slowing its AI-driven micro-pricing rollout in its maintenance business because it has not yet produced the expected improvement in customer retention, and that this deceleration would create about a $20 million headwind versus its prior full-year outlook. Micro-pricing in the separate repair business is reportedly progressing better as part of a larger incremental price-impact goal, but that figure is a forward outlook, not a confirmed result, so it is not included here. All figures are company (CEO) statements from the earnings call, not independently verified.

Uber cut about 10 percent of its Community Operations division, the unit that handles rider and driver complaints, refunds and disputes, and ended remote work for the staff who remain. A company spokesperson framed the cuts as being “to simplify operations, strengthen in-person collaboration and continue to embrace AI,” while Megha Yethadka, VP of Global Community Operations, told the team in a memo that the organization had become “too complex and siloed” and needed to be restructured to “layer AI on.” The company did not disclose a dollar-savings figure or an AI-specific productivity number, only the headcount percentage, and it explicitly frames the cuts as a mix of AI adoption, de-layering and a return-to-office push rather than AI alone. This follows a 23 percent cut to Uber’s People (HR) division in June 2026. Bloomberg broke the story on 2026-07-22 (the original Bloomberg article returned a 403 on refetch); Engadget’s republication carries the same company and executive quotes and was independently confirmed live.

FreightWaves’ 2026 AI Excellence in Supply Chain Awards (published July 16, 2026) named Uber Freight an “Operational AI Integration” winner, reporting that its document-processing AI handles more than 20,000 proof-of-delivery documents per week at more than 99% precision. The figure is attributed to Uber Freight but, as with the other winners covered in the same article, comes from the company’s own award submission; FreightWaves does not describe an independent verification of the precision rate or define what counts as an error in this context.

FreightWaves’ 2026 AI Excellence in Supply Chain Awards (published July 16, 2026) named YMX Logistics an “Operational AI Integration” winner, reporting that AI-driven yard management let the company cut its yard truck fleet from 22 trucks to 14, a reduction of roughly 36%. As with the other award winners in the same piece, the figure comes from the company’s own award submission as relayed by FreightWaves; no independent audit of the fleet count is cited, and the article does not state the time window over which the cut occurred.

Bank of America’s generative-AI assistant for call-center staff, EricaAssist, is now used by more than 18,000 customer service representatives and reduces average call time by nearly one minute, PYMNTS reported. Both figures trace back to a Bank of America press release rather than an independent measurement, so treat them as company-originated; PYMNTS did not report a third-party audit of the call-time reduction.

HCA Healthcare’s Timpani scheduling and staffing platform is now live at more than 130 hospitals and over 1,200 nursing departments, cutting the time to build a nurse schedule down to 2 to 3 hours per cycle. HCA has previously said (in an April 2025 update on the same platform, not independently re-verified here) that the tool had saved nursing leaders “hundreds of hours” since a 2023 nine-site launch, without giving a per-cycle baseline. The July 21, 2026 figure is HCA’s own company publication describing its own internal tool, with no independent measurement of the pre-AI baseline or the new number; treat the 2-to-3-hour figure as a first-party claim.

HCA Healthcare’s internally built AI staffing platform, Timpani, is now live at more than 130 hospitals and 1,200 nursing departments, cutting scheduling cycle time to 2 to 3 hours from a prior 8 to 15 hours. Departments using it reported a 6 percent decline in turnover and reduced contract-labor use. Separately, HCA’s Nurse Handoff AI tool, which the company says nurses rated 97 percent accurate and 95 percent helpful, is being scaled toward the system’s 100,000 nurses to support the 24 million shift handoffs that occur annually across HCA hospitals. All figures are self-published on HCA’s own corporate blog, not independently audited, and no baseline methodology for the turnover or accuracy percentages is disclosed.

On JPMorgan Chase’s July 14, 2026 earnings call, CEO Jamie Dimon disclosed that AI has already eliminated 30% to 40% of headcount in some of the bank’s units, per The Motley Fool’s reporting on the call. The article paraphrases Dimon’s disclosure rather than quoting his own words verbatim, and does not specify which units or over what time window the reduction occurred. No independent measurement of the figure exists yet; it is a company-side, CEO-originated claim.

On Bank of America’s Q2 2026 earnings call (July 14, 2026), CEO Brian Moynihan said the bank’s associates are generating more than 400,000 AI prompts a day and that more than 300 AI use cases have been approved. Separately, American Banker reported that 19,000 of the bank’s developers now use AI for real-time coding assistance, which the bank says is lifting their productivity by more than 20 percent. Both figures are self-reported by Bank of America (via its own CEO on the earnings call and via bank disclosure to the press) rather than independently measured productivity audits, and neither source breaks out how the 20 percent productivity gain was calculated. This is a distinct claim from Bank of America’s Erica chatbot user-count figures already covered in the July 16, 2026 radar run; this item concerns internal engineering tooling, not customer-facing usage.

On Citizens Financial Group’s Q2 2026 earnings call (July 16, 2026), CFO Aunoy Banerjee reaffirmed targets for the bank’s “Reimagine the Bank” program, a three-year, $300 million technology and AI overhaul: about $100 million in annualized pre-tax benefit by the end of 2026, $200 million by the end of 2027, and $450 million by the end of 2028. The program includes AI in credit-card contact centers, automated complaint logging, and AI-assisted commercial deal research. These are the bank’s own forward targets, not audited results, and the two sources agree on the figures with no conflict. The underlying “Reimagine the Bank” initiative and its 2028 target were first disclosed in March 2026; this item reports Citizens reaffirming the same figures on its most recent earnings call, not a new program.

On JPMorgan Chase’s Q2 2026 earnings call (July 14, 2026), CEO Jamie Dimon said the bank has already used AI to cut headcount by 30 to 40 percent in some discrete areas, and that most of the people affected were offered jobs elsewhere in the bank. Dimon paired the admission with a caveat that investors should not expect this to meaningfully widen JPMorgan’s margins, arguing that in a competitive banking market efficiency gains from widely-adopted technology tend to flow to customers rather than staying with any one firm. This is the CEO’s own on-the-record claim from the earnings call, not an independently audited figure, and neither source names which specific units or functions saw the 30 to 40 percent reduction.

On State Street’s Q2 2026 earnings call (July 16, 2026), CFO John Woods said equipping software developers with AI tools is expected to produce a 30 to 40 percent increase in productivity, as part of a broader transformation plan under which headcount was already down approximately 3 percent year over year. Woods separately said roughly 500 million dollars of the transformation program relates to headcount reductions, funded by one-time severance costs, with net headcount expected to decline only in the low single digits after reinvestment into other initiatives. This is management’s own forward-looking estimate from the earnings call, not a measured outcome, and only one source carried the quote at time of writing.

Thomson Reuters told staff on July 13, 2026 that it is cutting up to 500 engineering roles, about 5.2 percent of its 9,400-person operations and technology unit and roughly 1.8 percent of its 27,100-person global workforce, as it pushes AI further into its legal, tax and regulatory products. The company said it plans to hire more than 250 net-new engineering roles globally over the next two years, with the large majority described as senior and AI-native. Thomson Reuters framed the move as reallocating engineering capacity rather than a pure headcount cut, and did not disclose a net change to total engineering headcount once the 250-plus new hires are added back.

On FactSet’s Q3 FY2026 earnings call (July 1, 2026), CEO Sanoke Viswanathan said the company has reduced operator touch time for data table extraction by more than 50% “where we have fully implemented” new AI tools, and that coding agents now author 27% of committed code among engineering teams using them, with coding-related token use up 5x and AI-written committed lines up almost 10x quarter over quarter. These are the company’s own claims from its earnings call, not independently audited, and the touch-time figure applies only to teams with full rollout, not FactSet-wide. Included this run despite being 16 days old (outside the 7-day preference, inside the 30-day cap) because it carries a specific, quotable, named-company figure and had not appeared in an earlier radar run.

On its Q2 2026 earnings call, Fifth Third Bancorp said colleagues executed more than 1 million AI prompts in June alone, and that within its technology organization the AI code-prompt acceptance rate for new code was 45 percent during the quarter while over 87 percent of unit testing was automated by AI. The bank also shipped a new AI-powered navigation interface in its mobile app. Figures are self-reported on the earnings call, not independently verified, and the bank did not disclose a baseline pre-AI unit-testing automation rate for comparison.

Netflix co-CEO Ted Sarandos disclosed on the company’s Q2 2026 earnings call that generative AI tools were used in the production of roughly 300 titles this year alone. He cited one documentary containing 17 minutes of AI-enhanced footage produced twice as fast and at half the cost of conventional methods. These are company-disclosed figures from an earnings call, not independently audited, though corroborated across independent trade coverage. Netflix says the savings are being reinvested into content rather than cutting the budget.

Sprout Social, a social media management SaaS company, began notifying employees on July 15, 2026 of a workforce reduction of approximately 260 positions, or 20% of staff, per an 8-K filed the same day. The board approved the plan on July 8, 2026. The company’s own language ties the cuts to aligning its cost base with “strategic priorities, including ongoing investments in AI-powered social intelligence,” not to AI directly performing the eliminated roles; CEO Ryan Barretto framed it as acting “from a position of strength” while beating guidance, rather than as a distress cut. The caveat: neither the 8-K nor press coverage states which functions are cut or claims AI is doing that work now, so this is a funding-reallocation story, not a demonstrated AI-for-labor substitution.

Bank of America’s Q2 2026 earnings release (filed July 14, 2026) reports 24.6 million active users of its AI assistant Erica, up 23% year over year. Consumer Banking’s broader digital metrics in the same release show 70% of total sales were digitally-enabled (2.2 million digitally-enabled sales) and 4.4 billion digital logins, up 7%; these are Consumer Banking-wide figures, not broken out as Erica-specific, so they are reported here as context rather than folded into the Erica number. The 24.6 million and 23% figures come directly from Bank of America’s own earnings materials (tier 1, first-party), corroborated by independent press (CX Dive) reporting the same figures the same day.

On JPMorgan’s Q2 2026 earnings call (Tuesday, July 14, 2026), CEO Jamie Dimon said AI has cut staffing by up to 40% in some parts of the bank, with 30% to 40% of the people previously doing certain work in specific departments replaced by AI. He did not name which departments. Dimon added that most affected employees were offered other roles within the firm, and that the efficiency gain will not translate into dramatically better margins because competing banks are deploying the same technology and gains flow to customers through pricing rather than to JPMorgan’s operating margin. The department(s) affected and displacement-vs-reassignment split are not independently broken out; the 30-40% figure is Dimon’s own characterization on the earnings call, not an externally audited number.

Klarna’s AI customer-service agent now does work equivalent to roughly 850 human agents, up from 700 a year earlier, and the number of external human agents working with Klarna fell from 3,000 to 2,300 over the same period, per Forbes. The walk-back: Klarna found human agents were still needed for complex or sensitive cases and hired a team of around 100 highly skilled human operators to cover them. Note the “cut too aggressively” framing attributed to CEO Sebastian Siemiatkowski appears only in the article’s summary, not as a direct quote in the body; that specific admission could not be independently confirmed verbatim and is omitted here. The headcount and workload figures above are directly quoted and confirmed.

On ManpowerGroup’s Q2 2026 earnings call (July 16, 2026), North America president Becky Frankiewicz said nine months into deploying AI-powered, early-in-the-funnel interview tools across its staffing operations, the company is seeing “a 67% decrease in our time to fill that has been material for us.” She also said 30% of candidate interviews are now taking place outside normal business hours through the same automation, and that ManpowerGroup expects the AI-powered screening and go-to-market capability to scale to 70% of revenues by year-end 2026. This is the company’s own figure from its own earnings call, not an independently audited number, and no dollar-cost or headcount effect was disclosed alongside the time-to-fill percentage.

On UnitedHealth Group’s Q2 2026 earnings call (July 16, 2026), UnitedHealthcare’s Patrick Conway said the company’s digital prior-authorization product hit a 96% first-pass approval rate and saved 69,000 administrative hours. A second executive, identified on the transcript as Sandeep, gave the underlying volume: about half a million prior authorizations processed year to date, of which roughly 69,000 were for external entities to UHG. The figure is company-reported on its own earnings call, not independently audited, and the transcript does not state the time window “year to date” covers beyond 2026. The same call included adjacent AI claims (90% reduction in clinician-reported cognitive burnout from ambient documentation; ambient listening live for 70% of employed Optum Health providers) that were not included here for lack of a single clean quotable figure per item within this flash’s scope.

On its Q2 2026 earnings call (July 16, 2026), UnitedHealth Group said AI-based ambient listening technology, used to reduce clinical documentation burden during patient visits, is currently available to 70% of employed Optum Health providers, with a target of exceeding 90% by year end. This is a company self-reported adoption figure from an earnings call, not an independently measured time-savings or accuracy outcome; UnitedHealth also said it is targeting elimination of 30% of total prior-authorization volume in the same period.

Brazilian fintech lender Creditas has cut headcount from over 4,000 at its 2022 peak to fewer than 1,800, a more than 50% reduction, while its loan portfolio and revenue both doubled over the same stretch. The company’s shareholder VEF describes this as a “fivefold productivity gain,” with 40% of that improvement occurring in the preceding six months through AI implementation. The figures come from VEF’s Q2 2026 investor presentation on its Creditas stake, not an independent audit, so the AI attribution and the productivity-gain math are the company’s own framing.

Sprout Social’s board approved a plan, disclosed in a July 15, 2026 SEC 8-K, to cut approximately 260 employees, about 20% of staff, with the filing stating the move is designed to “streamline the Company’s organizational structure and align its cost base with its strategic priorities, including its ongoing investments in AI-powered social intelligence.” The company estimates $18.0 million to $20.0 million in pre-tax restructuring charges, mostly severance. CEO Ryan Barretto’s public statement did not use the word “AI” itself, only that “the way companies need to operate and invest has changed”; the direct AI linkage comes from the regulatory filing’s own language, not a spoken quote. Honest caveat: this is a company reallocating headcount and cost base toward AI investment, not a claim that AI performed the eliminated roles’ work.

This is a walk-back/failure item, not a win. FedScoop reports that the VA’s disability-claims backlog remains heavy, with 80% of the Veterans Benefits Administration’s 600,000 pending claims stuck in an evidence-gathering phase, while the agency leans on automation to keep up. A VA Office of Inspector General review (April 2026) found that 8,000 automated Pension and Fiduciary Service decisions or letters omitted favorable findings and had incomplete evidence summaries. VA acting deputy CIO Robert Orifici insists “every disability claim is decided by a trained VA employee, not by AI or automation,” but Rep. Nikki Budzinski (D-Ill.), ranking member of the House Veterans’ Affairs Subcommittee on Technology Modernization, noted staff cuts to 2,700 claims examiners since January 2025, raising the question of whether human oversight is keeping pace with automation. The 8,000 figure is the Inspector General’s own audit finding, not a company or vendor estimate, which makes it a stronger-than-usual data point for this beat.

On Citi’s Q2 2026 earnings call, CEO Jane Fraser said nearly nine out of ten Citi employees are now using the bank’s internal AI tools, tying the adoption figure to faster product launches such as Payment Express. This is a company self-report on its own earnings call, not an independent measurement, and the transcript gives no breakdown of what counts as “using” an AI tool or how usage was measured.

Sumner County Schools in Tennessee found that student groups averaging 25 or more CourseMojo AI tutoring activities improved ELA proficiency on state TCAP assessments by an average of 3.7 percentage points, and eighth graders in that heavy-use band rose 8.7 percentage points. Groups with little or no platform use saw a 0.2 percentage point decline over the same period. The figures are drawn from the district’s own state assessment data as reported by WSMV, an independent outlet, though the comparison is correlational (higher-use students may differ from low-use students in other ways) and was not run as a controlled study.

On Wells Fargo’s Q2 2026 earnings call, CEO Charlie Scharf said headcount fell to 197,000, down 15,000 from a year earlier and the 24th consecutive quarterly decline. Asked about the outlook, CFO Mike Santomassimo said “technology and AI helps us get out aspects of that… faster than maybe in the past, but we expect that we will continue to see more efficiency from here,” tying further headcount efficiency to AI without giving a separate AI-only figure. This is a self-reported executive characterization on an earnings call, not an independently measured AI-attribution study.

Chubb’s AI-driven Global Platform, which the company says handles 45,000 underwriting submissions a month, has cut policy endorsement cycle time from 22 days to 8, with a stated target of under 1 day in 2026. Both figures trace to Chubb’s own December 2025 investor presentation, relayed here by an industry research site (Emerj) rather than a fresh company release, so the underlying data point predates this coverage by about seven months and is company-self-reported, not independently verified.

At an internal town hall on July 2, 2026, Mark Zuckerberg told staff that AI agent development “hasn’t really accelerated in the way that we expected” and that the company’s reorganization was “not as ‘clean’ as planned” with bets on the new structure that “haven’t come to fruition yet” (Yahoo, July 7, relaying a Reuters recording). The admission follows Meta having “notified roughly 8,000 employees in May 2026, about 10% of its then-80,000 person workforce.” Read this as a candid internal walk-back on the pace of Meta’s AI-agent push, paired with a headcount figure, not a claim that the layoffs were caused by a specific deployment. Single source (Yahoo, citing a Reuters recording); the town-hall remarks are not from a Meta primary and should be treated as reported speech.

Aviation trade outlet OAG reports that more than 2,700 Southwest Airlines developers are already using Kiro, AWS’s agentic coding service, as part of a broader push to move the airline “from a largely on-premises environment to a cloud-based, AI- and agent-enabled architecture by 2028,” a transformation described as supporting more than 70,000 employees and 134 million customers a year. This is a scale/adoption figure (developer headcount using the tool), not a measured productivity or cost outcome; no hours-saved, error-rate, or dollar figure is disclosed for the rollout itself.

Physicians at Washington University School of Medicine and BJC Health System using the Abridge AI ambient scribe platform saw a 20% reduction in “pajama time,” the after-hours work spent drafting patient notes, with notes generated in about 30 seconds to a minute. The figure comes from an institutional press release describing “early studies” without stating sample size, comparison method, or an independent evaluator, so it should be read as a preliminary, company-adjacent claim rather than a peer-reviewed result.

Taco Bell and voice AI vendor Omilia extended their partnership, with Taco Bell’s own announcement putting the drive-thru voice AI footprint at “over 890 restaurants across 38 states to date” as of July 8, 2026. The same announcement claims transaction time “on par with, and in some cases faster than, traditional ordering methods” and that locations using the voice AI “reported higher employee retention compared with those where it hasn’t been deployed,” but neither claim carries a percentage or measured figure, only the deployment-scale count above. Honest caveat: this traces to one company/vendor announcement carried by multiple outlets (QSR Magazine, Fox Business), not an independent measurement, and the only hard number in it is the restaurant/state count, not a performance outcome.

OpenAI’s “OpenAI for Business” channel published a customer video (with Bain) featuring Bain Capital portfolio company Zelis, a healthcare payments and claims company, “using ChatGPT to streamline claims processing and help reduce costs across the healthcare system.” The one hard figure is an adoption count, quoted from James Mackey of Bain Capital: “Having 2,700 employees, all of whom now have an enterprise ChatGPT license, has just been a giant force multiplier for the business.” Two honest caveats: the post pairs Mackey’s 2,700-employee statement with the Zelis example but does not explicitly state that 2,700 is Zelis’s headcount, and there is no published operational outcome for the claims-processing use (no cost-reduction percentage, no claims-throughput or time figure). Separately, Zelis’s own product numbers (over $8B in claims savings, $155B priced annually, built on AWS Bedrock) are about its pricing platform, not this internal ChatGPT deployment, and must not be conflated. Single source; the primary is auth-walled and could not be re-fetched live, so this is a fresh lead to expand and validate, not a confirmed outcome.

Blend Labs announced that its Autopilot AI agent for mortgage pre-underwriting has gone live with its first five commercial lender customers, including Onity Mortgage, one of the largest mortgage servicers in the nation, after a four-month early-access preview in which lenders activated Autopilot across more than 25,500 real loans. Blend said loans routed through Autopilot during pre-underwriting moved faster and converted at higher rates than comparable loans on standard workflows, but the press materials do not disclose a specific percentage or time figure for that claim, so this is a volume figure rather than a fully quantified efficiency outcome, and it is Blend’s own vendor announcement rather than an independent report.

Commonwealth Bank of Australia cut more than 40 customer service roles and replaced them with an AI voice bot, then reversed the cuts after the system “was unable to cope, which led to an increase in calls” (CNBC, 2026-07-01, citing Australia’s finance sector union and an ABC report). The union called the rescission “a massive win.” The underlying cuts date to 2025; the fresh reporting is CNBC’s July 2026 retrospective on AI-layoff reversals. Single independent source in this flash; the union statement and ABC report are the primary trail for a deeper case file.

IBM automated its human resources functions with AI that “handled around 94% of routine requests but was unable to meet the other 6%, which included ethical dilemmas” (CNBC, 2026-07-01). The company then announced plans to triple its US entry-level hiring across all business units in 2026. Its chief human resources officer, Nickle LaMoreaux, framed the rehiring as a pipeline problem: “If we don’t continue to invest in entry-level hires, what happens in three-five years? There’s no pipeline.” The 94 percent figure is company-originated and relayed by CNBC; single independent source in this flash.

Amsterdam-based hotel operating-system provider Mews cut about 15% of its roughly 1,350-person staff (around 170 people) on July 7, 2026, framing it as a proactive restructuring toward becoming “AI-native” rather than a response to weak demand. Founder Richard Valtr is quoted saying AI is “now capable of handling much of the execution layer of that work intelligently and at scale, and more profitably,” arguing the cut roles belonged to “an era that is ending.” The move follows Mews raising $300 million in January 2026 at a $2.5 billion valuation earmarked for agentic AI that autonomously coordinates hotel pricing, staffing, and guest services. This is a company-framed walk-back-adjacent story: the causal claim (AI eliminating the need for these roles) is the founder’s own characterization, not an independently measured productivity or cost figure.

Salesforce announced Agentforce Help Agent with pay-per-resolution pricing: customers are charged only when the agent resolves an issue, generally available July 2026. The proof point is Salesforce’s own deployment: the company says Agentforce on help.salesforce.com “handled 4.3 million inquiries and resolved 70 percent of them” (announcement dated 2026-06-25), a figure CIO relayed as “resolved 70% of them” the same day. Both figures originate with Salesforce and are not independently audited; the deployer and the vendor are the same company, so read this as a vendor’s self-deployment claim with an unusually concrete pricing commitment attached. The pay-per-resolution model itself is the newsier fact: it prices the product on outcomes, which invites external measurement.

NHS England said an initial trial of its new NHS App AI triage tool at Wealden Ridge Medical Partnership, a GP practice in Sussex, cut the number of patients queuing on the phone by 29 percent while maintaining patient satisfaction levels. The tool asks adaptive questions to route patients to the right service, GP, pharmacy, A&E, community service or self-care. NHS England says the AI triage tool is due to reach more than 200,000 patients within 12 months, part of a wider GBP 10 billion NHS AI push that also includes rolling Microsoft Copilot out to over 500,000 NHS staff. The 29 percent figure is from a single-practice pilot, not yet a system-wide result, and comes from NHS England’s own announcement rather than an independent evaluation.

Cisco CFO Mark Patterson told Fortune that AI now produces 80 to 90 percent of the first draft of the management discussion and analysis (MD&A) sections of the company’s public filings. Cisco is separately rolling out personalized AI agents to all 90,000 employees starting around August 2026. The figure is a self-reported CFO statement, not an independently audited measurement, but the underlying claim is corroborated by a second independent trade outlet.

Ford says it spent the past three years hiring back 350 veteran engineers, some former employees and some from suppliers, after concluding that feeding its design requirements into AI tools alone was not producing high-quality vehicles. Charles Poon, Ford’s VP of vehicle hardware engineering, is quoted by TechCrunch admitting the company “mistakenly” assumed AI ingestion of design requirements would be sufficient. The returning “gray beard” engineers now lead troubleshooting and have retrained the AI tools to flag defects earlier in the assembly process. Ford says the fix helped it top J.D. Power’s Initial Quality Survey for mainstream brands for the first time in 16 years, and CEO Jim Farley is quoted by Forbes crediting the effort with generating “hundreds and hundreds of millions of dollars” of cost tailwind through lower warranty and recall spending. This is a walk-back: Ford, not an outside AI vendor, is the one describing where its own AI rollout underperformed, and the dollar figure is the company’s own estimate rather than an independently audited number. The story is dated June 28 to 30, 2026, outside the 7-day preference window but kept given its scale and the concrete, quotable figures.

Oracle’s fiscal 2026 SEC filing states that “the adoption and deployment of AI technologies across our operations have resulted, and may continue to result, in reductions to our workforce,” alongside a disclosed cut of 21,000 jobs during the fiscal year and $1.8 billion in severance and restructuring costs (versus $374 million the prior year). This is a rare case of a major company naming AI explicitly as a cause of layoffs in a formal regulatory filing rather than only in earnings-call color or press statements. The filing is dated outside this run’s normal 7-day freshness window (reported June 24, 2026, about 29 days before this run), kept as the genuinely significant exception the beat allows: it is, per the sources checked this cycle, the first instance of a major tech company attributing headcount reduction to AI adoption in a federal filing rather than a memo or interview.