Every year, we attend banking conferences and hear advice echoed from stage after stage: “Get your data together to use AI.” It’s become a mantra in the industry. Everyone agrees it’s important, and then most go home and nothing meaningfully changes.

Data matters. And the leaders I talk to know it matters. The part that many miss today is understanding what “getting your data together” really means. That gap, between the slogan and the substance, is good news. It means the biggest opportunity to improve your efficiency ratio is still in front of you.

When most people talk about getting your data together, they mean integration. Pull it from your core, your loan origination system, your CRM, your general ledger, and get it into one place, maybe a dashboard. Consolidated reporting is better than fragmented reporting, but if the end goal is using artificial intelligence to drive your efficiency ratio, then integration is step two of a ten-step journey.

The good news? The next steps are clearer than you might think.

What “getting your data together” really means is something far more granular and far more valuable. It means refining your data. It means embedding business context, encoding the business logic of your specific bank. It means having your data ready to use AI. Once you understand that distinction, the path forward becomes surprisingly actionable.

A properly built data foundation is the unlock to using AI. It’s the key to doing the work you already do, but quicker, easier, and smarter.

Context matters because an LLM is extraordinary at processing language, identifying patterns, and generating output. What it cannot do, on its own, is understand that when your bank says, “Primary Relationship” or “Class 3 commercial real estate,” it means something slightly different than when the institution next door says the same thing. It doesn’t know your policy exceptions, committee preferences, or the dozen small decisions your best banker makes without even thinking about them.

That’s business logic. And every bank’s logic is different, which is a strength. Your institutional knowledge is a competitive asset. The data foundation doesn’t just store your data. It teaches AI how your bank actually works, preserving and scaling the expertise your team has built over years.

Take credit memo writing. It’s the example that illustrates the opportunity clearly, and once you see it, the same pattern shows up across the institution.

A credit memo is a structured synthesis of financial data, borrower history, market context, policy compliance, and risk assessment. It’s organized in a way that tells a clear story to a credit committee. It requires knowledge, context, and consistency.

A commercial banker and credit analyst might spend hours on a single credit memo. Not because the intellectual work is that complex, but because the assembly is. Pulling data from multiple systems, cross-referencing financials, ensuring the narrative aligns with current policy, formatting it correctly, reviewing it for completeness. Your bankers shouldn’t be spending their best hours on assembly.

Now the exciting question: what is stopping AI from writing that credit memo?

Frontier LLMs can synthesize, analyze, and write at extraordinary levels. What’s stopping them is context and business logic. The AI doesn’t know your bank’s specific credit policy. It doesn’t know how your committee likes to see deals presented. It doesn’t know that your chief credit officer always wants to see the debt service coverage calculated a certain way, or that your institution has a specific appetite for owner-occupied CRE that differs from the industry norm.

But if you’ve built the data foundation, if you’ve done the important work of refining your data, embedding your context, and encoding your logic, then an AI agent can draft that memo in your format, with your logic. The commercial banker reviews it, applies judgment, and moves on to the next relationship. The work that took hours now takes minutes. Multiply that across your team, across your branches, across a year, and you’re looking at a meaningful shift in your efficiency ratio from a single AI capability: writing credit memos.

So what does this look like in practice? I’d suggest a framework that starts with the work, rather than the tech.

Decompose. Pick a role: commercial banker, credit analyst, branch manager, compliance officer. Map the activities that role performs. Be specific. Instead of “lending,” think “spreading financials from tax returns.” Instead of “compliance,” think “reviewing BSA alerts and documenting decisions.” You’ll be surprised how clarifying this exercise is.

Categorize. Which of those activities require judgment, relationships, or creativity? Which are information processing: data retrieval, synthesis, pattern matching, structured output? When you free your people from assembly work, they can focus on the high-value activities that drew them to banking in the first place.

Build the foundation. This is the work that matters most and gets talked about least. Refine your data. Contextualize it. Encode your business logic. Build the data infrastructure that turns your raw institutional knowledge into something an AI can reason with. This is a strategic investment, and it’s the step that separates institutions using AI from institutions talking about AI.

Deploy with precision. Start with one AI agent — conversational, accessible, embedded in your workflow — that takes on one specific, high-leverage activity. Credit memo drafting. Loan covenant monitoring. Exception tracking. Whatever generates the most time savings for the most people. Prove the value, build trust, then expand.

Every CEO I talk to cares about their efficiency ratio. It’s one of the clearest measures of how well an institution converts revenue into profit. But here’s what I’d encourage every leader reading this to consider: the efficiency ratio of the future won’t be driven by the same levers as the efficiency ratio of the past.

It will be driven by how effectively your people are equipped to do their work. It will be driven by whether your institution has built the data infrastructure to let AI do what AI is uniquely qualified to do, so your people can do what they’re uniquely qualified to do. The institutions that invest in foundational, context-rich data work now will operate at a level of efficiency that sets them apart.

That’s the future driver of efficiency ratio. It’s not a dashboard. It’s not a chatbot bolted onto your website. It’s a data foundation that knows your bank as well as your best banker does, and an AI layer that puts that knowledge to work, every day, at scale.

The opportunity is here. The path is clear. And the institutions that move now will be the ones that define what community banking looks like for the next generation.