The daily publication · October 27, 2026 · Small business

More loans, same people: the bank that spent its AI hours on growth

By Tabron Merryhill · The Workflow

A bank just freed up 24,300 hours a year with AI. It did not cut a single job with them. It told its people to do twice as much business instead.

The idea

Most AI-for-labor stories end the same way. A task gets automated, a role gets cut, and the savings go to the bottom line. First Internet Bank of Indiana ran the same play and chose the other ending.

The bank put Bretton AI agents on its compliance research work. SBA compliance reviews that took eight to ten hours now take under one hour. That is an 87% cut in review time. Across SBA origination, enhanced due diligence, fraud reviews, and periodic customer reviews, the bank estimates about 24,300 employee hours freed a year. That is roughly 12 full-time people worth of hours.

CEO David Becker told the team to use the capacity to do twice as much business, not to shrink the team. The hours went back into lending, not out the door.

The example

Picture a small business owner waiting on an SBA loan. The slowest part is rarely the decision. It is the compliance research underneath it: checking the applicant, the documents, the risk flags, the sanctions lists. A human does that work one file at a time, and the queue grows while the owner waits.

First Internet Bank handed the research to agents and kept the humans on the judgment. The agent gathers and checks. A person still reviews before anything final. The loan moves faster because the research stopped being the bottleneck.

Customers Bank ran the same pattern in parallel. Its pilot closed select commercial loans in seven days against a typical 30 to 60. Same lesson, different bank: speed came from removing the research queue, not from removing the reviewer.

Here is the fair counterpoint, and it matters. The 24,300-hour figure is vendor-reported capacity, not payroll savings, and PYMNTS Intelligence data in the same piece finds 63% of enterprise leaders cite data quality as a barrier to AI. Agents fed messy records produce fast, confident, wrong research. The guardrail that makes this setup defensible is the one Kaufman Rossin stressed in its independent validation of Bretton's agents: mandatory human review before final disposition. The agents research. They do not decide. Any owner copying this pattern should copy that line too, because it is the whole difference between capacity and a compliance failure.

The next step

Find your own research queue this week. Pick one bottleneck process where someone on your team spends hours gathering and checking before a decision gets made. It could be loan paperwork, vendor onboarding, quote comparisons, or permit checks.

Time it before you change anything. Write down how long the research takes today and where it stalls. Then run a supervised pilot: let the tool gather, and keep a human sign-off before anything final goes out. Measure the time again after two weeks.

If the hours come back, decide in advance what they buy. More customers served, not fewer people employed. That decision is the strategy. The tool just makes it possible.

Want help finding the one bottleneck worth handing to AI first? A free 10-minute workflow conversation will map where the hours go and who signs off before anything final. No pitch, just the map.

Sources: PYMNTS, https://www.pymnts.com/news/banking/2026/first-internet-bank-bets-ai-means-more-loans-not-fewer-people/; Pulse2, https://pulse2.com/bretton-ai-profile-will-lawrence-interview/; Fintech Brainfood, https://www.fintechbrainfood.com/p/the-benchmark-will-eat-finance; Morningstar, https://www.morningstar.com/news/business-wire/20261019748783/kaufman-rossin-independently-validates-bretton-ais-kyc-aml-and-sanctions-agents. Opposing view: PYMNTS Intelligence, https://www.pymnts.com/news/banking/2026/first-internet-bank-bets-ai-means-more-loans-not-fewer-people/.