Bernard Worthy’s Post

Human underwriters read bank statements top-down... AI reads them all at once. Here's a real-world demo we ran on a retail sunshade business applying for a microloan. They came in with 3 months of bank statements + the documents every CDFI asks for on intake. What the AI surfaced in the first pass: 🟢 Seasonality in deposits, expected for the business type 🟢 Lumpy revenue, but consistent year-over-year 🟢 Transfer activity suggesting the account functions as a pass-through 🟢 Cash flow inflated by inter-account movement, masking real revenue Time to flag the pass-through → 30 seconds. A human underwriter would catch the same pattern on page 4 of the statement, halfway through a 45-minute read. But sometimes, they never see it. Then we stress-tested: 🟢 Deposits down 15% (still passed) 🟢 Higher debt load layered on top (still passed) The AI handles the analyst's grunt work, where talented folks key numbers off tax returns into spreadsheets for hours. The credit decision belongs to the underwriter. For most community banks under $10B, that's still how small business deals get done. What's eating the most hours on your underwriting team each week? If it's the prep work before the decision, AI can take most of that load.