Built by payment engineers who processed $2B in transactions — and got tired of watching fraud prevention fail in ways every fraud tool vendor said were impossible.
Between 2019 and 2022, InferX co-founders Priya Rajapakse and Marcus Chen ran payment infrastructure at a mid-size fintech processor handling $700M annually. Their fraud stack was three tools duct-taped together: a rules engine, a third-party velocity checker, and a manually updated blocklist.
In Q3 2021, a card-testing ring rotated through 4,000 accounts over 11 days, probing with $1 transactions before cashing out. The rules engine had no velocity window short enough to catch it. The velocity checker flagged 300 accounts — but 2 hours too late. Gross fraud loss: $340,000. Recovery: $18,000.
Every fraud detection vendor they evaluated had been built for e-commerce merchants. Their ML models were trained on cart abandonment signals and email marketing engagement. They had no concept of BIN-level risk, interchange velocity, or the network relationships between processor-level entities.
InferX was incorporated in March 2023 to build the tool they couldn't buy.
A model trained on payment processor data catches 40% more fraud than a general-purpose fraud model at the same false positive rate. Domain matters. We build nothing that isn't specific to the payment processor context.
Every millisecond between transaction submission and scoring is a window for card-testing attacks. Sub-50ms scoring is not a performance benchmark — it's a security requirement. We designed every part of the pipeline around this constraint.
Black-box fraud scores create operational debt. When a dispute arrives and your analyst cannot explain why a transaction was declined, you have a compliance problem. Every InferX decision ships with a human-readable explanation.
A regional processor handling $120M/month in card-not-present volume had a 1.8% fraud rate and a rules engine generating 12% false positives. After 90 days with InferX: fraud rate dropped to 0.6%, false positives to 1.9%. Chargeback ratio fell below Visa's dispute monitoring threshold for the first time in 18 months.
A B2B payments platform processing ACH and card transactions for 800 SMB merchants faced escalating synthetic identity fraud. InferX's network graph identified that 23% of flagged accounts shared device fingerprints from a 14-IP subnet. False positive rate on legitimate SMB accounts: 0.3% — within acceptable operational limits.
An embedded finance provider launching card issuing for 3 neobank clients needed fraud scoring before their first transaction. Starting from scratch with InferX: production scoring live in 5 business days using the pre-built card issuing risk model as a starting baseline, with custom thresholds set per neobank client.
Incorporated in March. Completed Seed round in August. Hired the initial engineering team of 6 from payment infrastructure backgrounds.
First payment processor scoring live transactions. Sub-50ms latency achieved in production under real transaction load for the first time.
Shipped the fraud network graph module. Immediately identified a card-testing ring spanning 4 processor customers simultaneously — previously invisible to each processor individually.
Crossed 1 billion scored transactions. Launched the Growth plan and the self-serve developer portal. Team expanded to 12.
Talk to the team. We know the problems you're dealing with because we dealt with them ourselves.
Get in Touch Meet the Team