The report covers a survey run between April and June 2025 among five hundred and fifty-six executives at financial institutions across some twenty markets, with more than a hundred employees and revenue above ten million. It reports five areas where they claim results — productivity, customer experience, growth, marketing and security — and gives the second half to agents: slightly over half say they have them in production, and nearly half are putting more than half of their future AI budget there.
What I took from it
The detail that decides how everything else reads sits in a footnote: unless stated otherwise, the figures cover only those already using this technology in production. That does not invalidate the data, but it changes the sentence: it is not that three in four companies see a return, it is that three in four of those who already deployed say they do. Since then, when a client arrives with a figure from a report, the first thing I look for is who was asked.
What I do use are the questions, transcribed verbatim in the notes. They work as they are for a diagnostic conversation, and they are better phrased than most questionnaires in circulation.
And one figure genuinely useful as an expectation anchor: the declared time from idea to production is measured in months, not weeks.
What I left out
The conclusions and the customer stories, which are the vendor's sales material. And a general warning about the genre: a report where the return is declared by the executive who approved the investment measures enthusiasm, not profitability. Read it for the questions and the methodology, not for the percentages on the cover.