Why it matters
Mehrotra credits three forces for the retreat: model providers raising prices, an experimentation phase that has ended, and teams finally seeing enough value to operationalise it. A cost-discipline account from a regulated buyer, not a vendor.
Tokenmaxxing read
His filter is frequency, not sophistication: a dazzling use case that fires once a month loses to dull high-volume work with a measurable baseline. Experiments are timeboxed and killed when they miss, so the spending decision lands before any tokens burn.
Source takeaway
A Q&A, so every claim is the bank’s own framing with no independent verification and no spend figures. Useful for the operating model — evangelise, educate, enable, execute, and reuse over reinvention — not for benchmarking what U.S. Bank pays.
