Missing middle layer in fintech fraud
Trisha Kothari, CEO of Unit21, on the fraud problem in fintech
The real bottleneck in fintech fraud is usually not model quality, it is the gap between raw signals and the operations team’s ability to act on them quickly. In practice, analysts often have identity checks in one tool, transaction data in another, and a basic admin dashboard that shows history but cannot block payments, link related accounts, or turn repeat patterns into reusable rules. Unit21 was built around that missing middle layer, where non technical risk teams can investigate alerts, connect many data types into one user profile, and decide before money leaves the system.
-
Older bank tooling was built for fixed products like checking or cards. Fintech fraud looks different because risk depends on the exact product and behavior, like whether a user is borrowing, paying invoices, moving crypto, or changing account settings, which means teams need customizable workflows rather than one generic fraud score.
-
The practical gap is between dashboards and action systems. A tool like Retool can show account details, but deeper fraud work needs alert queues, case management, linked entity analysis, and the ability to hold or block transactions in real time. Unit21 and similar platforms productize that operations layer so companies do not have to build it themselves.
-
The category is splitting into adjacent layers. Alloy is strongest at onboarding and identity decisioning, where a company checks whether an applicant is real and should be approved. Sardine contributes device and behavior signals. Unit21 sits closer to the ongoing decisioning and investigation layer after identity checks, when a user has already entered the system and fraud risk is unfolding across transactions and account behavior.
This stack is moving toward a single operating system for fraud and AML, where detection, investigation, and action live in one workflow. The winner will be the platform that lets a risk team test rules, explain decisions, share fraud intelligence across customers, and stop bad payments before settlement, without waiting on engineers every time fraudsters change tactics.