Roughly 40% of mid-market SMB lenders now run AI-assisted underwriting in production. Performance data shows decision speed up 60%, default rates unchanged.
AI-assisted underwriting has crossed from pilot to production at scale in 2025–2026. Industry surveys suggest 40% of mid-market SMB lenders now use ML scoring in some part of the decision pipeline; some are fully automated below certain ticket sizes.
Performance data is encouraging. Decision speed has improved 60% on average for AI-assisted files. Default rates on AI-decisioned files are tracking flat to slightly better than human-decisioned counterparts — once normalized for ticket size and credit tier.
Stip resolution, edge cases, fraud review, and any deal above defined dollar thresholds remain human-driven. AI is augmenting throughput, not replacing analyst judgment on complex files.
Twelve-month default rates on 2026 originations are tracking 80 bps below the 2023 vintage at the same seasoning point, with the largest improvement in services and B2B.
Funders that adopted soft-pull pre-qualification in 2025 are reporting 2.3x higher application-to-funding conversion and materially lower CAC than peers still gating with hard pulls.
Falling SBA fees and faster PLP processing have changed the math. For tickets under $500K with weaker collateral, 7(a) is now structurally cheaper than conventional term once total cost of capital is measured.
