Proximitty's bet: regulator-grade AI agents run commercial loan servicing

An AI-native commercial loan servicing platform founded in San Francisco in YC's Winter 2026 batch: autonomous agents ingest and parse borrower documents, including blurry scans and handwritten notes, spread financials on each lender's business logic, generate credit memos, monitor covenants and handle collections for FDIC-insured banks, credit unions and fintechs.

That banks will trust regulator-grade AI agents to run loan servicing - chasing documents, spreading financials, tracking covenants - instead of scaling headcount.

CEO Wye Yew Ho advised banks and fintechs on risk strategy at McKinsey and later led growth at Taptap Send as it scaled from $75M to $200M ARR; CTO Zi Zhang led security infrastructure at Bloomberg. During YC's Winter 2026 batch the pair founded Proximitty in San Francisco to automate the back office of commercial lending.

Proximitty's agents chase borrower documents, reconcile discrepancies, spread financials, generate audit-ready credit memos, monitor covenants and handle collections; its Servicing Studio records a workflow or standard operating procedure and turns it into a production agent. The company claims roughly 5x faster underwriting, about 80% lower servicing operating expense, and one analyst covering 6,500 loans versus a typical 800, and says it plugs into core systems from Fiserv, FIS, Jack Henry, Finastra, Temenos, nCino and Moody's with a three-week onboarding arc. It presented at FinovateSpring 2026, with YC, CRV, Cohen Circle and Uphonest Capital among backers and angels who lead teams at Stripe and OpenAI.

Early and live: as of September 2026 Proximitty remains an active two-person San Francisco startup selling to community and regional banks, credit unions and non-bank lenders, with a SOC 2 report in its observation period and published metrics drawn from early deployments.

The founders were the customer: Ho advised banks on risk at McKinsey and grew Taptap Send from $75M to $200M ARR, so the team had felt the pain of covenant breaches caught too late.

Document chasing and financial spreading is exactly the bounded, rule-heavy labor agents can do now, and banks already budget whole teams for it, making the pitch a cost line a CFO can defend.

Auditability is the product in regulated lending: audit trails, reproducible decisions and no training on customer data answer the objection that kills most banking AI.

Distribution respected the installed base: integrating with Fiserv, FIS, Jack Henry and nCino instead of replacing the core turned an 18-month migration into a three-week onboarding pitch.

A proof point beat a pitch: one customer's 15-to-2 headcount story, shown at FinovateSpring 2026, gave early-stage Proximitty credibility with publicly traded fintechs and banks.

Sell capacity, not features: Proximitty's customers defend a budget item that gives one analyst the reach of eight, and regulated buyers trust software that shows its work.

As of September 2026 Proximitty remains an active two-person San Francisco company in YC's Winter 2026 batch, selling to community and regional banks, credit unions and non-bank lenders. Its published economics - 6,500 loans per analyst, 5x faster underwriting and roughly 80% lower servicing operating expense - are company-provided figures from early deployments, and its SOC 2 is still in observation. Against incumbents nCino, Abrigo, Moody's and Numerated, the open question is whether end-to-end agentic servicing displaces the point tools banks already run.