Process automation

The document and rules heavy ones, on TIN's record. Omega Healthcare runs UiPath automation across medical billing, coding and insurance correspondence, and JPMorgan built COIN to interpret commercial loan contracts. The work is repetitive, high volume and structured enough for extraction and decisioning to carry the routine cases while people keep the exceptions. Customer-facing chat and open-ended judgment are not where back office automation lands first.

Not on any deployment TIN's record can confirm. Klarna went AI-first on customer service, then its CEO said the cost-driven push had produced lower quality, and the company began re-recruiting human agents. Octopus Energy took the other route: its Kraken tool drafted support emails and a person reviewed and sent every reply. The deployment that kept a human on every reply had nothing to walk back.

Not by default. The Bank of Korea found AI adoption cut average work time, but the link between time saved and output growth was essentially zero: an efficiency stage, not yet a productivity stage. Firm-level cases show where freed time did become output. Delivery Hero cut account lockout recovery time with an n8n workflow, and Otto's replenishment forecasts are checked against whether the items sold. Each ties its claim to an event the business can count.

Rarely. Omega Healthcare's time and accuracy figures originate with Omega or UiPath, and its hours-saved figure exists at two dates that TIN shows side by side rather than merging. The famous hours figure for JPMorgan's COIN traces to a single Bloomberg article attributing it to the bank's own designers, and it has never been independently measured or updated. Read these as accurately quoted company claims, and ask who counted.

Not on any deployment TIN can independently confirm. The agent framing is ahead of the evidence: no insurer running autonomous claims agents in production has a measured outcome on the public record. The one claims process on the record that decided alone, a rule-based benefits system at the VA, was reviewed by its inspector general, who found a deficiency in almost every case sampled.

Not necessarily, on TIN's record. Otto's automated stock replenishment came with hiring more people, not fewer. Octopus Energy framed its AI drafting as workload equivalence, called job cuts unlikely and went on to announce thousands of new roles. Klarna, which froze hiring and stated its assistant's output in full-time agent equivalents, reversed course and began re-recruiting humans. A workload-equivalence figure is a model, not a count of layoffs.