AI for Community Banks: What to Use Now, What’s Coming, and How to Stay Out of Trouble

If you’ve been to a banking conference in the last year, you’ve heard the same thing I have: AI, AI, AI. Your board is asking about it. Your peers are claiming to use it. Your competitors are putting it in their marketing. And yet, when you ask people what they’re actually doing with it, the answers are quite vague. We all know AI matters; many just don’t know what to do with it. The good news is that the gap between asking the questions and putting it into practice is pretty small.

What’s Working Right Now

Think of today’s AI as a tireless junior analyst who never sleeps, never complains, and occasionally needs to be corrected. Used that way, it’s already paying for itself at community banks across the country. The highest-return uses we’re seeing today:

• Drafting and editing. Loan committee memos, board packet narratives, marketing copy, customer letters. AI gets you to a solid first draft in minutes instead of hours. (Full disclosure — I used Claude Opus to get the first draft of this very article.)

• Meeting summaries and action items. Microsoft Copilot handles your Teams meetings, and a device like Plaud captures the in-person ones. Either way, a 60-minute meeting becomes a one-page summary your team will actually read.

• Policy and procedure work. Updating an outdated policy, comparing your procedures against a regulatory change, or generating a first draft of a new SOP.

• Loan file preparation. Summarizing tax returns, financial statements, and credit memos so your underwriters spend their time analyzing rather than transcribing.

None of these replace your people. They provide faster answers for your people. This is especially true for pattern recognition with financials and large amounts of data.

Two Skills That Multiply Your Results

Two things separate banks getting real value from AI from banks getting mediocre results.

1. The first is prompting. Your team needs basic training on how to ask. A vague request gets a vague answer. A clear request — with context, an example of the output you want, and the audience in mind — gets something genuinely useful.

2. The second is picking the right model for the job. Tools like Microsoft Copilot now let you choose which underlying AI model handles your request — Claude Opus, GPT, or Auto. The smarter models (like Opus) are better for complex analysis, nuanced writing, and anything regulatory. Auto is fine for quick everyday tasks.

What to Be Ready For

The next wave is moving from “AI helps me write things” to “AI helps me decide things” — with a human still in the loop.

• Underwriting assistance that flags risks and missing documents before a file hits the underwriter’s desk

• Fraud and AML pattern detection that surfaces unusual behavior faster than rules-based systems

• Internal knowledge agents trained on your own policies so a new hire can ask “what’s our process for a new depositor?” and get the right answer in seconds

• 
  

  Compliance monitoring that reviews marketing materials and disclosures for issues before they become findings

Don’t Skip Vendor Management

Before anyone on your team types a customer name into a chatbot, you need an answer to a simple question: which AI tools are we allowed to use, and which ones are we not?

Treat AI tools like any other vendor. Vet them for SOC 2, data handling, where the data is stored, and whether your inputs are used to train someone else’s model. Enterprise tools like Microsoft Copilot are built with business-grade controls. Public ChatGPT is not — and it shouldn’t be touching customer data. Then write a short, plain-language AI use policy that names the approved tools and gives clear examples of what’s okay (drafting an internal memo) and what’s not (pasting customer NPI into a public chatbot).

One Last Thing

AI can hallucinate — it’ll confidently make things up, including citations and regulations. The newer models are getting better at this, and some are specifically built to reduce it. You can also help by telling the AI directly: “If you don’t know, say you don’t know — don’t make it up.” That single sentence in your prompt does more than people realize.

So, before you let AI anywhere near anything customer-facing or examiner-facing, have a human check it.

If you’re not sure where to begin, you’re not alone. Forward in Technology is rolling out an AI Readiness and Governance service soon to help community banks with managing their AI usage.  If that would be useful, reach out — we are happy to talk through it, no pitch required.

AI isn’t going to replace community bankers. But community bankers who use AI well are going to have a noticeably easier time than those who don’t.

by Andrew Johansen, President, Forward in Technology