Rebuilding Tide's Legacy Risk Engine
Reset a legacy risk engine around fresher fraud patterns and broader signals.
- Risk engine
- Model retraining
- Device intelligence
- Fraud precision
- Roadmap reset
Saurav Mahapatra
Lead Product Manager at Tide
From their time as
Senior Product Manager
Tide • 2023 - 2026
Overview
Saurav inherited a risk engine that retrained annually and relied on a limited signal set. He reset the roadmap around more frequent retraining and additional data, raising precision from about 4.5% to nearly 12%.
The Story
Saurav inherited Tide's Initial Financial Crime Risk Engine after it had been maintained through an annual retraining cycle. The model incorporated new fraud typologies during retraining, but its feature set had not expanded with newer device, IP, VPN, and behavioral signals.
He reframed the work around the cost of stale fraud patterns. Fraud tactics shift over time, and an annual cycle could put a model into production using signals that were no longer sufficiently predictive.
Saurav worked with engineering and data science to move retraining to a six-month cadence. He also changed the integration approach so the model could enrich the onboarding payload with data from other services, including GeoIP and device intelligence, rather than waiting for the orchestrator to expose every signal.
He tested whether the changes improved detection quality. Precision rose from about 4.5% to close to 12%, reducing the share of legitimate members incorrectly flagged for review.
Ownership Snapshot
Broad role
Product manager for a legacy financial crime risk engine.
Goal
Improve fraud detection quality and reduce false positives.
Direct ownership
Reset the roadmap, retraining cadence, and signal strategy.
Team execution
Worked with engineering and data science on model and data integration changes.