A payments processor’s fraud detection model was accurate in testing but nearly useless in production, because the batch data pipeline feeding it only updated every six hours — long after fraudulent transactions had already cleared.
Standing Up Real-Time Data Pipelines for Fraud Detection at Scale
Service Provided
Capabilities Utilized
Streaming Data Architecture, Real-Time Infrastructure, Model Serving
Client challenge
The existing data infrastructure was built for daily reporting, not real-time decisioning, and re-architecting it touched systems considered too critical to risk disrupting. Meanwhile fraud losses were mounting on transactions the model would have correctly flagged, if only it had seen the data in time.
Our solution
We built a streaming data pipeline running alongside the existing batch infrastructure rather than replacing it outright, capturing transaction events in real time and feeding them to the fraud model within seconds of a transaction occurring. The batch pipeline remained in place for reporting and reconciliation, so nothing considered mission-critical was touched during the transition.
We rolled the real-time path out to a small percentage of transaction volume first, running it in shadow mode against the batch system to validate accuracy and latency before it was trusted to make live blocking decisions.
Results
Fraud detection latency dropped from a six-hour batch delay to under 400 milliseconds, letting the model block fraudulent transactions before they cleared rather than flagging them after the fact. In the first quarter on the real-time pipeline, the processor prevented an estimated $3.4 million in fraud losses that would previously have gone undetected until after settlement.
Additional Case Studies
Launching a Usage-Based Pricing Model for an AI Developer Platform
An AI developer platform’s flat-tier pricing was leaving money on the table with its largest customers
Establishing an Enterprise AI PMO to Track ROI Across 30 Initiatives
A global enterprise had 30 active AI initiatives across its business units but no consistent way
Building the Product Operating Model for an Agentic SaaS Vendor
A vertical SaaS vendor had shipped an early agentic feature that could autonomously complete customer workflows,
