Good Morning!

AI pricing changed under your feet this year. What used to be a flat subscription fee is now billed by the token, and costs are spiking. RBC's usage jumped 500 percent in a year. But cost isn't your only problem.

New research out this week says most banks are pointing AI at the wrong people entirely, and it has nothing to do with budget.

Let’s dive deep.

ANKUR PATEL Founder & CEO, Multimodal

TRENDING AI NEWS FOR CB

Your AI bill is about to get a lot less predictable

American Banker reported this week that banks are scrambling to control AI costs as vendors shift from flat subscription pricing to token-based billing. RBC's token usage jumped 500 percent year over year. PNC's CEO has said publicly that token costs are eating into the productivity gains AI was supposed to deliver.

Why it matters for your bank: If your AI contracts are still priced like software subscriptions, renegotiate before the next renewal, not after the invoice shocks you.

A third of banks can't confirm they can turn off a bad AI model

A Wolters Kluwer survey of 230 banking professionals found that 34 percent weren't sure their institution had a kill switch for its AI models. Another 38 percent flagged AI failure reporting as their least-prepared area.

Why it matters for your bank: Examiners are already asking these questions in routine exams. Not knowing the answer is worse than a bad one.

Nearly half of community banks have deployed generative AI

A Cornerstone Advisors survey found 49 percent of community banks have deployed generative AI, with credit unions slightly ahead at 59 percent.

Why it matters for your bank: If half your peer group is live, "still evaluating" is no longer neutral. It's falling behind.

DEEP DIVE

Your newest teller might already be your best AI user

New research making the rounds this week found something most banks are getting backward. Stanford economist Erik Brynjolfsson studied over 5,000 workers using AI. Average productivity gain: 14 percent. Novices gained 34 percent. The most experienced barely moved. AI didn't make your best people better. It made your newest people almost as good, instantly.

A similar pattern showed up outside financial services, too. Experienced software developers were measured 19 percent slower using AI on their own code, while believing they were faster. The pattern holds everywhere researchers have looked: AI helps most where experience is thinnest.

Here's the part that should change how you buy. MIT studied 300 enterprise AI deployments and found 95 percent produced no measurable financial impact. The 5 percent that worked bought a focused tool built for the job instead of building one internally. Vendor-bought projects succeeded roughly two out of three times. Internal builds succeeded about one out of three.

For a community bank watching AI costs spike this week, this is good news dressed as a warning. Your newest hire gets the biggest lift. Your veteran loan officers get almost nothing. Point the tool at the right person, and buy it instead of building it.

Three things to do this week:

  1. Name the newest person on your lending or teller team. Ask what tool would compress their learning curve.

  2. Before your next AI purchase, ask whether it's a focused vendor tool or an internal build trying to do everything. The data says buy.

  3. Pull your last AI pilot's results. If you can't name a measurable outcome, you have a demo, not a pilot.

FROM MULTIMODAL

Good AI implementation is slow at first

Nothing about integrating a new AI system into a legacy core banking environment is seamless. The data is messy. The fields don't map cleanly. The loan workflow has exceptions that only three people in the building understand.

Good implementation looks slow at the front end, honest about the constraints, and built with the people who run the workflow, not just the people who signed the contract. The banks that get to fast, reliable deployment are the ones that front-loaded the discovery. They sat with the ops team. They mapped every exception. They built for the actual system, not an idealized version of it.

"Seamless" is a sales word. The real word is methodical.

Data point this week

49%

of community banks have now deployed generative AI, according to Cornerstone Advisors' 2026 survey, roughly triple the share that had deployed it in 2025.

Source: Cornerstone Advisors' 2026 survey.

BEFORE YOU GO

Is your newest hire or your most experienced person using AI more effectively right now?