Bretton AI agents slash bank compliance reviews by 87%

First Internet Bank of Indiana has cut compliance review times for Small Business Administration loan applicants by 87% using AI agents from Bretton AI, according to a case study reported by PYMNTS on October 26, 2026. Reviews that once took eight to 10 hours now take less than an hour, and the bank says the freed capacity will go toward doing more business, not reducing headcount.

The $6 billion-asset digital bank estimates the technology frees 24,300 employee hours a year across four compliance workflows. That figure equals the annual hours of roughly 12 full-time employees on a 2,080-hour work year, but the bank frames it as capacity gained rather than payroll saved. The workflows covered are SBA origination, enhanced due diligence, fraud reviews and periodic customer reviews.

The result matters beyond one bank because compliance work is the quiet bottleneck in small-business lending. Borrowers can meet every credit requirement and still wait while analysts chase records, verify ownership and screen for risk. When that step shrinks from a full workday to under an hour, the whole loan file moves faster.

How Bretton AI agents handle the compliance file

Bretton AI agents take over the research that analysts once did by hand, according to the case study. The agents retrieve corporate records, identify beneficial owners, check sanctions lists and fraud indicators, and then consolidate the findings into a structured report. Analysts review that report and take the cases that need human judgment, rather than starting each file from a blank page.

That division of labor is the point. The agents do the retrieval and the assembly; people do the decisions. It is a pattern showing up across enterprise AI this week, from intranet tools to factory floors. Interact Software, for example, just connected 700 intranet capabilities to enterprise AI assistants through Interact MCP, and Verosoft launched Verovia to connect industrial systems with governed AI agents. In each case, the agent earns its place by working inside existing systems and permissions, not by replacing them.

First Internet Bank was already growing into the bottleneck. It funded $610 million in loans under the SBA 7(a) program in 2024, up 44% from a year earlier, according to the bank, and chief executive David Becker has described the bank as a top 10 SBA 7(a) lender. More volume meant more files for the same analysts, which is exactly the kind of pressure that makes automation attractive and also makes staff nervous about what comes next.

When the bank's SBA manager asked whether the gains meant fewer people, Becker answered that the goal was double the business, according to the case study. That answer reframes the usual automation story: the prize is growth capacity, not a smaller team. For a digital bank competing nationally for SBA volume, speed is a selling point to borrowers and brokers alike.

Customers Bank shows the same pattern in commercial lending

First Internet Bank is not alone in aiming AI at the slowest part of the loan file. Customers Bank in Pennsylvania used AI in a pilot to close select commercial loans in seven days, against a typical period of 30 to 60 days, as PYMNTS reported. The bank's parent, Customers Bancorp, also estimated more than $15 million in annualized staffing capacity gains as it scales the process, with employees estimating cumulative savings of 76,000 hours across the bank's AI initiatives.

The compliance work is the same at both banks. Lenders must verify ownership, confirm identities and screen for risk before a loan moves forward. Both banks are shortening that step instead of adding staff to it, and both are counting the gain in capacity rather than headcount.

The pattern extends beyond banking. Oracle NetSuite unveiled SuiteAgents at SuiteWorld, industry-specific AI agents designed to monitor ERP workflows and recommend next steps for finance teams. The throughline is consistent: agents that watch ongoing work, gather context and prepare the next move, while people approve and act.

The catch: agents are only as good as the data they read

Speed like this depends on clean data, and most large companies are not there yet. PYMNTS Intelligence's April Enterprise AI Benchmark Report found that 63% of enterprise leaders cite data quality as a barrier to AI. The report surveyed 65 senior technology executives at United States firms with at least $1 billion in revenue. Nearly all said they were very or extremely confident their data governance supports enterprise AI, yet 85% said their data is fragmented or only moderately integrated, and only 15% said it is mostly integrated.

Financial services is no exception. In the May edition of the benchmark, 30% of financial services firms named data quality as their top barrier, according to PYMNTS Intelligence. A lender that closes in seven days competes against one that takes 30 to 60, and staying that fast takes data its agents can read.

That tension defines the next phase of bank AI adoption. The tools to cut a review from ten hours to under one already exist, and the case studies are now public. The harder work is the unglamorous part: integrating records, cleaning ownership data and keeping sanctions and fraud screening current, so the agent's structured report is one an analyst can trust. Banks that do that work get the speed. Banks that skip it get a faster way to produce a report nobody believes.

For First Internet Bank, the bet is explicit: use the 24,300 hours to write more loans. If SBA volume keeps growing at anything like its recent pace, the bank will find out quickly whether its agents can keep up with its ambitions.

Sources: PYMNTS on First Internet Bank and Bretton AI; Interact MCP brings intranet tools to enterprise AI assistants; Verovia industrial AI brings agents to factory floors; NetSuite SuiteAgents bring industry AI agents to ERP.