Beyond Hype: How to Make AI Work

By Aaron Colcord, CFO and founder of Voyager AI, a WBA Associate Member

Let’s get real for a second. Community and regional banks aren’t dragging their feet on Artificial Intelligence (AI).

The real problem is that banks are complicated. It’s like running a restaurant with a 50-page menu but only two chefs in the kitchen. Eventually, you stop offering the fancy dishes, not because you don’t want to sell them, but because you just don’t have the staff or the time to cook them.

Everyone in tech says, “AI is the solution!” You can’t go one to two minutes without someone saying the word AI now. It’s also not a secret, we probably have a subscription to ChatGPT, Claude, or even Google’s Gemini on our phone and we’re giving it our own personal information in return for a better cake recipe (You are probably wondering, is this a true story? Not mine, but yep).

Did I name one of the elephants in the room? That basic ability to keep our bank information completely safe inside the vault. We didn’t copy customer information over to Google when search engines first showed up. Even today, you can do a basic Google search for internal bank information and find absolutely nothing.

Or how about this other elephant?

If you ask a standard AI model a question on Monday, and then ask it the exact same question on Tuesday, you might get two completely different answers. That’s totally fine if you’re asking it to write a funny poem about a cat. It’s absolutely terrifying if you are asking it to evaluate a $5 million commercial loan.

Imagine a bank auditor knocking on your door to ask why a specific deal was structured the way it was. If your answer is, “I don’t know, the computer just felt like it today,” you’d be fired before lunch. Banks cannot run on guesses or probabilities. You can’t audit the missing person. You need cold, hard certainty.

So, how do we make AI safe for a bank to actually use? We create a hard architectural separation between processing data and making decisions.

Think of AI like a super-smart, caffeinated intern. This intern is amazing at doing the exhausting busywork. They can speed-read messy tax returns, dig through 50-page PDFs, and instantly pull out the key numbers you need. But, and this is the important part, you do not let the intern make the final call on the deal.

To make a system reliable, the actual decision-making part (the rules engine) cannot be run by an AI brain.

Instead, once the AI intern gathers the data, it hands those numbers over to a strict, perfectly predictable calculator. This is where we write the actual math and logic. There is no guessing. There is no “AI magic.” If you feed the exact same financials and risk profiles into this strict rules engine one thousand times, it will spit out the exact same answer one thousand times.

This hybrid approach changes everything. You use AI to do the exhausting work of organizing messy data, but you use strict, hard-coded logic to execute your lending rules.

Sorry, it’s not groundbreaking. We have been trusting computers to do this exact work for decades. AI is the interface and the helper to the human which leads to…

The Problem With the “Loop”

But if the AI is the intern and the rules engine is the boss, where does that leave the actual human lender?

This is where the tech industry gets it completely wrong. Most tech companies talk about putting a “human-in-the-loop.” Sound comfortable and reasonable. We still have the human checking the system, not part of the system, they are sitting on the loop. Your highly experienced lending team babysitting the robot.

You need your lending team’s experience. You need a way for them to grow their experience. That is what will make your bank grow. Your unique way of approaching the business problem. Your relationships.

Putting the Human in the Lead

We need to throw out the loop and adopt a “human-in-the-lead” approach instead.

In a human-in-the-lead system, the technology works for the lender, not the other way around. The AI intern gathers the messy files. The deterministic rules engine crunches the policies and immediately flags what is missing or what hurdles exist.

Then, the technology gets out of the way.

The human lender takes the lead. Armed with all the right information instantly, they can apply their years of institutional knowledge, look the business owner in the eye, and figure out how to make the deal work. They aren’t spending hours searching for a missing schedule in a tax document; they are spending their time advising a local business on how to expand their warehouse.

This is all about the fundamentals of what being a bank is really about: trust and relationships. It’s about putting the customer first. The whole point of computers is to handle the tedious stuff so people can be better people with each other. When your systems handle the heavy lifting safely and predictably, your team kills the “long no.” More importantly, they get back to providing incredible service and building the real relationships that community banking is built on.