You Can’t AI Your Way Out of Bad Data

What three bank CEOs said on a Saturday morning, and the one statement that flipped everything upside down.

 

If Adele can show up in full glam at 8:00am on a Thursday morning, we can manage Saturday morning at 8:30am with an AI strategy agenda.

A couple weeks ago I had the privilege of joining Anne Tangen, Meg McIsaac, and Julie Thurlow for a panel at the Massachusetts Bankers Association Annual Convention. The room was full — genuinely, not “conference full” — of community bank leaders who showed up to talk about data and AI on a weekend morning.

That alone told me something.

The session was titled “Preparing Your Data for AI, Deposit Defense, and What Comes Next.” A broad bill. But these three CEOs kept bringing it back to one idea, one fundamental truth:

You can’t AI your way out of bad data.

I want to tell you why that line landed the way it did.

The banks asking the best AI questions did the boring work first

Here’s what struck me as these women described their journeys: the decisions that positioned them well for AI weren’t AI decisions at all. They were data decisions made two, three, sometimes four years ago — unglamorous ones. Getting reports out of silos. Agreeing on a single version of the truth. Pushing data beyond the C-suite to the branch managers, lenders, and tellers who are actually closest to customers every day.

One of the panelists described what happened when she pushed data to the front line of her bank. Not reports emailed on the fifth of the month, but real-time information built for someone who shouldn’t need a training session to understand what they’re looking at. Curiosity showed up almost immediately, but it wasn’t curiosity about numbers in the abstract. A branch manager could see deposit outflows by product, by officer, by day. She didn’t have to wait for someone to pull a report — she could see it, form a theory, and adjust before it hit the bottom line. A lender could look at their portfolio and spot where to push harder while there was still time left in the month.

The questions haven’t changed. Why are we behind on deposits? What’s driving fee income? Those questions have always existed in community banking. What changed is that people could finally see the answers themselves and do something about it. Data sitting in a system nobody opens isn’t a data strategy. The banks building real data cultures aren’t just giving people access. They’re giving people clarity. And clarity is what makes everything that comes next, including AI, actually work.

Where AI actually lives right now

There was a real tension in the room: fear versus curiosity. Fear of replacement. Curiosity about relief. The CEOs handled it practically: the way to move people from fear to curiosity is to show them something concrete. A repetitive task that disappears. Forty hours a month reclaimed. A lender who preps for renewal calls in half the time.

The most useful AI conversation, they said, isn’t about transformation. It’s about relief. What’s burning time, and what could take it off your plate?

That framing is also accurate. For most community banks right now, AI is operating at the individual productivity level — a draft email, a summarized document, a starting point for a query. Useful, but not yet embedded in the places where community banks actually win or lose. Underwriting quality. Deposit retention. Officer performance. Credit risk.

Why? Because AI at that level requires a data foundation clean and consistent enough that people trust what it’s telling them. And the shift from making decisions purely on instinct to making decisions where data is in the room too — that’s a muscle every institution is always training. No one ever finishes it.

The instinct doesn’t go away, and it shouldn’t. It’s often what separates a great community banker from a good one. The goal was never to replace it. The banks that figure out how to pair that instinct with data they actually trust — that’s where the edge gets sharper.

The line that defined the hour

Midway through, one of the CEOs said something more impactful than any pitch you’ve ever heard about AI:

She talked openly about a tool her bank had implemented that didn’t deliver what they hoped. And her take was generous — it wasn’t the technology’s fault. The data foundation underneath it wasn’t where it needed to be, and what AI does in that situation is reflect exactly what it’s given. The lesson she walked away with? Sequence matters. Get the data right first. Not because AI isn’t worth investing in, but because the investment performs so much better when the foundation is there.

Clean, accessible data isn’t a precondition for an AI strategy. It is the AI strategy.

What I appreciated most about how she framed it was the absence of regret. It was advice, not a warning. The banks that wait for AI to mature before worrying about their data are going to find themselves building the foundation and the house at the same time — wondering why neither is moving as fast as they need it to.

And here’s what’s worth remembering: community banks aren’t behind on this. The largest institutions carry legacy complexity that makes data governance genuinely hard to untangle. A $2B bank with clear data ownership, unified reporting, and dashboards that reach the front line has something many larger institutions are still trying to build. The size that once felt like a disadvantage is starting to look a lot like agility.

The crossroads are real, but it’s not grim

The panel ended with urgency, not alarm. AI in community banking isn’t a question of whether. It’s a question of sequence, and how ready you’ll be when the use cases that actually matter start to mature.

Used on a clean foundation, AI can close gaps with larger institutions faster than almost any other lever available. Used as a substitute for that foundation, it exposes the gaps instead. The worst version of this story isn’t a bank that moves slowly on AI. It’s a bank that moves quickly on AI without the data underneath it and spends the next two years rebuilding internal trust in the technology.

The banks asking the right AI questions today are almost always the banks that made data decisions two or three years ago. Not because they saw AI coming. Because they decided their people deserved better information, and they built accordingly.

That decision is still available. And the institutions that make it now will be the ones positioned when AI becomes the standard operating layer across the workflows where community banking is actually won or lost.

A few of us wanted to keep going well past sixty minutes on a Saturday morning. IYKYK. If that sounds like you, we just launched the KlariVis Data and AI Institute C-Suite Summit — a small executive gathering in Roanoke this August for exactly this kind of conversation. More details at klarivis.com.