Before You Buy AI, Do This: Community Bank’s Blueprint for Getting AI Right

There is a version of the AI conversation happening at banks across North America right now that goes something like this: the board mandates AI, leadership starts evaluating vendors, and someone ends up in a demo before the institution has done the work to know what problem they are actually trying to solve with AI. It is a familiar arc, and it is one of the reasons AI investments at banks and financial institutions so often stall.

That is not the version of the story I heard from Matt Mayo, Chief Risk Officer at Community Bank, during our recent discussion with the Mass Bankers Association. Matt has been a banker for 29 years, most of them at Community Bank, a $5 billion institution serving Mississippi, Alabama, the Florida Panhandle, and Memphis, Tennessee. He walked our audience through how his bank came to deploy AI in its mortgage operations. The story is worth hearing. Matt and his team did not start with AI. They started with process discipline. AI was the tool he reached for after months of foundational work that ensured an AI tool was the right solution for their problem.

If you are a banking leader reading this and wondering where to start with AI, Matt's playbook is worth studying. Here is what it looks like.

Start with efficiency, not with AI

Matt's team did not initiate an AI project. They initiated an efficiency project.

"We recently went through the rate escalation cycle. The rate increases that took place a couple years ago, this constrained normal earnings," Matt said. "Large investment portfolio, short, but very low yield, a lot of fixed loans, and so as rates went up, we quickly got compressed. And so a big initiative really became looking at how to become more efficient."

That is a framing every banking leader will recognize. Rate compression and margin pressure predate AI. So does benchmarking against peers on efficiency ratios and assets-per-FTE and the recognition that a bank running on outdated processes cannot scale to meet customer expectations.

The important detail is the sequence. Community Bank started asking how to become more efficient before AI was a serious topic in banking. When AI did become viable, they already knew where their inefficiencies lived. An AI solution was not the initiative, but rather the tool that could solve a problem they had already defined.

Map the processes, then ask the front line

Before Community Bank looked at a single vendor, they did something most institutions skip: they mapped their processes.

"We consolidated our family of banks a couple years ago," Matt explained. "The first one was on pace to cross the billion-dollar threshold... but we never formally consolidated the processes. So our starting gate was engaging a third party to help us in our major workflows, our major processes: how we originate loans, how we service loans, how we onboard staff, how we maintain deposit accounts. All those things had never really been well defined."

They mapped the processes. First, they identified where the processes broke most often. Then they started measuring. Only once they had done these steps did they start looking for automation solutions. And when they did look, they used the process map as the diagnostic tool: which of the broken points are mechanical and repeatable enough to automate?

The other decision Matt made is worth noting because most banks get it wrong. He did not ask the managers where the problems were. He went straight to the people doing the work.

"We didn't ask the managers. I wanted the people that sat in front of the customer, or who did the GL reconciliation, or who were keying in the data. Because what gets to the managers gets filtered."

Matt described walking through the operations center one day and noticing an employee manually pulling last-four ZIP codes from the USPS site for every customer address change, then re-keying them into the core. It was work that had been going on for years, but no one had asked why. He fixed it in ten minutes with existing technology.

Then he made that story public inside the bank. He set up an internal channel for employees to submit their own "why do we still do this" questions. He crowdsourced the automation backlog from the front line, not the org chart. That crowdsourced list is still driving the priorities today.

The lesson: you cannot automate what you have not examined, and you cannot examine what your leadership team has never questioned. Front-line staff know where the friction is. They just need to be asked.

Pilot where you can measure

Once Community Bank had mapped the processes and started building a backlog of automation candidates, they had to choose where to pilot. Matt was deliberate about the choice.

"We chose to start in our mortgage area," he said. "The benefit of the mortgage process and their systems is they're very measurable. I have a hard time determining KPIs for how long things happen in my ops center. Our software just doesn't account for how we do things. Well, the mortgage system does. And so it's really easy to go in and see, it takes them on average, 30 to 35 days from beginning to end. Well, if I can cut 10 or 12 days off that process, we deliver a service faster for a customer, so they're happier. We make money sooner, because we start earning quicker in that cycle."

Two things made mortgage the right place to pilot. First, it is a standardized product. Loans sold on the secondary market look similar to each other in a way that commercial deals rarely do. That standardization gives the automation something reliable to learn on. Second, mortgage operations already have KPIs. Time to close, time to funding, cycle time by stage. Community Bank could measure the impact of AI in mortgage in a way they could not measure it elsewhere.

For any banking leader weighing an AI pilot, that is the question worth asking: where do we already measure well? The answer is not always mortgage. It might be indirect auto, or check processing, or a regulatory filing workflow. But whatever the answer is, it needs to be a workflow with hard numbers already in place. If you cannot measure the deployment, you cannot defend the investment to your board.

Where Saris fits in

By the time Matt's team started evaluating vendors, they had done the harder work. They knew their processes and their pain points, sourced directly from the people experiencing the pain. They had chosen a measurable product vertical. The vendor conversation started from a specific problem, not from general curiosity or a board mandate about AI.

Saris fit in because the mortgage workflow Community Bank was trying to automate is exactly what Saris is built for. Documents arriving over the weekend without complete information, forcing the MLO to send a follow-up on Monday and adding days to the cycle. Income calculations done manually with subjectivity between MLOs and underwriters. Compliance checks and appraisal ordering handled repeatedly across every deal.

"When somebody's sitting there making an application on a Saturday night, if an agent can look at that application, validate the fields that are complete, validate documents that are attached are in fact what they are, and put the deal in our mortgage loan origination system, we cut down what could easily be five to seven days of back and forth just to get to a starting point," Matt said.

That is what Saris does in the mortgage arc: reads the file, verifies the fields, cross-checks documents, runs the income calculation, and surfaces exceptions to a human for review. Every action is logged, every finding is source-cited, and the underwriter still owns the credit decision. Community Bank piloted it in exactly the workflow they had chosen deliberately, and they are measuring it against the KPIs they already had in place.

What banking leaders can take from this

Matt's playbook is not particular to Community Bank. It is a framework any banking leader can run against their own institution.

Start with efficiency, then map your processes with enough discipline that you know where they break. Ask the people on the front lines where the friction lies. From there, pick a pilot workflow where you can measure the impact. Once you start evaluating vendors, the conversation will look very different. You will know what you are looking for, what a good answer sounds like, and how you will measure success.