Stopping Fraud Before It Clears
From flagging fraud after the fact to stopping it mid-transaction.
Overview
Vantor Payments processes transactions for mid-sized merchants across North America. Their existing fraud rules were static and rule-based, catching obvious cases but missing anything novel — and generating enough false positives that legitimate customers were getting blocked.
The Challenge
The client's rule-based fraud system couldn't adapt to new fraud patterns and was declining roughly 1 in 20 legitimate transactions, frustrating customers and costing revenue.
- ›Fraud rules required manual updates and lagged new attack patterns by weeks
- ›False positive rate high enough that merchants were losing legitimate sales
- ›No feedback loop between confirmed fraud cases and the detection system
- ›Scoring ran as an overnight batch job, so fraud was often caught after settlement
The Solution
We built a real-time risk-scoring model trained on the client's transaction history, deployed as a low-latency inference service sitting directly in the payment path.
- ›Trained a gradient-boosted ensemble on 3 years of labeled transaction data
- ›Deployed inference as a sub-200ms service in the live authorization path
- ›Built a feedback loop so confirmed fraud and disputes retrain the model weekly
- ›Gave risk analysts a case queue with model explanations for every flagged transaction
Results
Fraud losses dropped sharply while false declines fell even further, and the model now improves continuously instead of waiting on quarterly rule updates.
-58%
Confirmed fraud losses
-74%
False decline rate
180ms
Median scoring latency
Weekly
Model retraining cadence
“We used to find out about fraud patterns after they'd already cost us money. Now the model catches the pattern before it becomes one.”
More case studies
Logistics & Freight Tech
Keeping a Multi-Carrier Freight Platform Running, Around the Clock
A tiered, SLA-bound support model that gave Sea Technologies' FEED platform 24/7 resilience without pulling engineers off the roadmap.
Retail & E-commerce
Syncing Inventory Across 140 Stores in Real Time
A real-time inventory sync layer that eliminated stock discrepancies between Cresthill's stores, warehouses, and online storefront.