Transaction arrives
Real-time ingestion at the payment gateway layer. Device fingerprint, IP, behavioural signal, velocity history, and linked-account graph — all assembled within the first 10ms.
SENTR · SOLUTIONS · FRAUD DETECTION
Most fraud platforms stop at a risk score. SENTR closes the loop — connecting every detection decision back to investigation outcomes, dispute results, and model updates. Your stack gets smarter with every transaction it sees.
Current fraud stacks catch patterns they were trained on. Novel attacks, velocity abuse, and cross-channel combinations slip through. The only signal is a chargeback 30–90 days later. By then the fraud ring has moved on.
"We have 15 rules in our fraud engine. Half of them catch historical attacks. The new ones? We find out about them at chargeback time."
Four stages. One model. Every transaction resolved before the payment gateway decides.
Real-time ingestion at the payment gateway layer. Device fingerprint, IP, behavioural signal, velocity history, and linked-account graph — all assembled within the first 10ms.
Gradient-boosted trees, sequence models, and graph neural networks fire in ensemble. Cross-merchant fraud ring patterns that single-model detection misses are surfaced at this stage.
Score + confidence band + top contributing signals + human-readable rationale — all returned before gateway decision. EU AI Act Article 13 explainability satisfied at decision time, not as a post-hoc audit.
Every dispute outcome — won or lost — is fed back into the model as a training signal. Fraud intelligence compounds over time. The model gets harder to beat as your portfolio grows.
SENTR runs in parallel to your existing stack during shadow mode. Every decision above is produced in read-only mode — your live operations are untouched. After 50 days, you see every false negative your current stack missed.
Four sub-capabilities. All active in shadow mode from Day 1.
Every transaction scored before the payment gateway decision. No user-facing delay. Average latency 84ms across EU data centres. P99 under 100ms.
Gradient-boosted models catch single-transaction fraud patterns. Graph neural networks catch cross-account fraud rings. Both fire on every transaction. One score. One decision. No blind spots.
Every score surfaces the top contributing signals — velocity, device mismatch, IP anomaly, linked-account graph — with confidence weighting. EU AI Act Article 13 compliant at the decision level.
Dispute outcomes retrain the model automatically. Pre-transaction scoring improves with every dispute cycle. Fraud intelligence compounds month over month — no manual rule updates.
SENTR processes your live transaction data against the full fraud detection stack in parallel to your existing system — scoring every transaction against 500+ signals without modifying your live decisioning. After 50 days, you have a complete picture of every false negative your current stack produced.
Start your 50-day shadow mode evaluation →Free. No contract. No cutover. 8-day API connection — your engineering team stays in control. See the full 8-day integration timeline →
Book a 20-minute Architecture Session. We review your current fraud stack, score a sample of your transaction history against SENTR's ensemble, and scope your shadow mode integration. You leave with a clear picture of the gap — or confirmation there isn't one.
Book your Architecture Session →50-day shadow mode — free · No contract until Day 50 · No cutover to go live