Product-Centric Fraud in Fintech
Trisha Kothari, CEO of Unit21, on the fraud problem in fintech
This shift turns fraud prevention from a branch level rules problem into a product design problem. A fintech like Affirm wins by serving a specific use case, like paying over time at checkout, not by owning a local market, so its risk stack has to learn the patterns of that exact product, merchant mix, and user behavior. That is why one generic bank fraud system breaks down, and why tools like Unit21 are built as configurable decision layers for different money movement products.
-
Traditional banks were organized around accounts, cards, and branches, with fairly standard monitoring workflows. Fintechs are organized around products like BNPL, invoice payments, crypto trading, or embedded lending, each with different actions, data, and fraud failure modes. Unit21 was built to ingest those different signals and let risk teams tune rules without rebuilding the stack each time.
-
The concrete example is merchant level variance inside one product. At Affirm, fraud patterns differed sharply between categories like hair extensions and mattresses, which means the same payment flow can carry very different risk depending on what is being sold, how often, and to whom. Product specific underwriting and monitoring become part of the product itself, not a back office add on.
-
This also explains the market structure around Unit21, Alloy, and Sardine. Alloy is strong at identity checks during onboarding, Sardine adds device and behavior signals, and Unit21 sits as the core decisioning and case management layer that combines identity, transaction, and in app behavior into one user profile. In fintech, risk tools are increasingly assembled around the product workflow rather than bought as one monolithic bank system.
The next phase is even more product specific. As fintechs add instant payments, vertical lending, crypto rails, and AI driven support, the winning risk platforms will be the ones that can adapt rule logic and investigations to each workflow in real time, while also sharing fraud intelligence across customers so every new product does not have to learn the same attack pattern from scratch.