The Evident AI Index ranking in 2026: what the ranked banks have actually proven

Every October a bank’s comms team gets a new line for the annual report: its place in the Evident AI Index. Boards read the table as a scoreboard of who is winning with AI.

The problem for anyone using it that way is in the method. A readiness ranking scores inputs. The question a board, a buyer or a regulator actually has is about outputs: which of these banks has a result someone other than the bank has measured?

What the Evident AI Index ranking is

The Evident AI Index ranking is an annual table of 50 of the world’s largest banks, scored by Evident Insights on four pillars: talent, innovation, leadership and transparency.

Evident is plain about what it is. In its own words, the index “is an ‘outside-in’ assessment based on public data” [source].

How is the Evident AI Index scored

The Evident AI Index is scored on four weighted pillars, and talent carries almost half of the total. Evident lists talent at 45%, innovation at 30%, leadership at 15% and transparency at 10% [source].

Innovation covers “research, patents, ventures, ecosystem”, and leadership is scored “in public communications and strategy” [source]. No pillar is named for a measured business result. That is a design choice, not a flaw: the index measures capacity, and capacity is easier to see from outside than outcome.

Evident AI Index banks 2026: the top of the table

The 2026 table, published in October, puts JPMorganChase first, Capital One second, Royal Bank of Canada third, CommBank fourth and Wells Fargo fifth [source].

Further down, Goldman Sachs is 12th, ING 22nd, Danske Bank 31st and Nordea 43rd, all as listed on Evident’s own index page [source]. Those five banks are the ones TIN has checked a public AI claim on, which is why they are the ones this post reads.

What the ranked banks have actually proven

The obvious reading of the table is that a higher rank means more proven value. The record on these five banks says something narrower: a higher rank means more published activity, and the published figures are almost all the bank’s own.

JPMorganChase, ranked 1st. The bank says LLM Suite “went from zero to 200,000 onboarded users within eight months” [source]. Trade press relays the same count, which traces back to JPMorgan, and the productivity claim does not reconcile across outlets: “three to six hours per week” in one ([source]), “an hour saved here and three hours there” from the bank’s own chief analytics officer in another [source].

The bank’s most famous AI number is older. COIN is said to do work that “once required 360,000 hours of work each year by lawyers and loan officers” [source]. That figure comes from a single 2017 Bloomberg article attributing it to JPMorgan’s own designers, and it has never been independently measured or updated.

Goldman Sachs, ranked 12th. The one third-party finding in this set is a regulator’s. New York’s Department of Financial Services investigated the Apple Card, which Goldman underwrites, and said its review “did not produce evidence of unlawful discrimination against applicants under fair lending law” [source]. That is a clearance, not a performance result.

ING, ranked 22nd. ING told investors its chatbot handled 46% of chats without human support at the end of 2023 and 75% at the end of 2025 [source]. The trend is clear and the measurer is ING.

Danske Bank, ranked 31st. The headline numbers carry their own warning. Teradata’s 2017 release marks its 50 and 60 percent figures with an asterisk that resolves to “Pre-production measures” [source]. An analyst firm later put the production cut in false positives at 20 percent to 30 percent [source].

Nordea, ranked 43rd. Nordea’s own Capital Markets Day slides report 5.2 million chats with about 78% handled by its chatbot, over October 2024 to September 2025 [source]. The bank does not define “handled”, and nobody outside it has measured the figure.

How the ranked banks compare

Ranks are from Evident’s 2026 index page [source]. Figures are from the sources cited in the section above, each traced in the case file linked below.

What TIN checked

Each row above is a TIN case file, checked against the public record before it went live, and each page states its own limits.

• JPMorgan’s LLM Suite: a first-party adoption count relayed by two trade outlets; productivity shown as a disputed range.
• JPMorgan’s COIN: a myth-check; the 360,000 hours is a single-origin 2017 corporate claim.
• The Apple Card investigation: the regulator found no unlawful discrimination and faulted service and transparency.
• ING’s chatbot, Danske’s fraud engine and Nordea’s chatbots: each stops short of green, because the bank or its vendor is the only measurer.

The pattern holds from first place to 43rd. Rank does not sort these banks by how well their AI claims are evidenced; on this sample, the quality of the evidence looks much the same everywhere.

Bottom line

The Evident AI Index ranking is a good map of where banks are building AI capacity and a poor proxy for where AI has delivered a measured result, and Evident’s own method page says as much. The number-one bank’s two most cited AI figures are both its own. Use the rank to pick whom to watch, then ask each bank the question the table cannot answer: who, other than you, measured this? The principle travels beyond banking: a ranking of inputs tells you who is trying, never who has succeeded.