AI Decision Layers for Financial Services
Charles Birnbaum, partner at Bessemer Venture Partners, on the five waves of fintech
The main value is not the model itself, it is the software layer that turns raw AI into something a bank or insurer can actually put into production. In financial services, every AI output has to fit existing approval rules, audit trails, vendor controls, and regulator facing workflows. That creates room for companies that package AI inside a controlled underwriting, fraud, or onboarding system, instead of selling a general model alone.
-
Insurance is an early wedge because underwriting already runs on messy documents, repeated judgment calls, and expensive human review. Sixfold AI was described as focused on underwriting, and Unqork has since pushed deeper into insurance with AI underwriting products that centralize workflows and automate document extraction and summarization.
-
The pattern matches other fintech infrastructure winners. Alloy combines identity checks, fraud signals, rules, and audit ready workflows in one configurable layer, serving both startups and large banks. That is the same basic architecture, software that lets customers adopt new data and AI tools without rebuilding compliance operations from scratch.
-
Insurance already has adjacent proof points. Shift Technology sells AI systems for underwriting risk, claims fraud, and compliance risk, showing that buyers prefer vertical tools trained on insurance workflows, not generic copilots. The spend goes to products that reduce manual review time while improving loss ratios and keeping decisions explainable.
This is heading toward AI control planes for regulated workflows. The winners will be the companies that own the decision layer, gather data from many vendors, record why each decision was made, and plug into old bank and insurer systems. That makes them harder to replace than a standalone model, and pushes them toward core infrastructure status.