AML / Fraud Detection
Calibrated, explainable alerts that satisfy a regulator - not a black box that floods your analysts with false positives.
TRANSACTION STREAM - LIVE SCORING
Payments · 2.1M/yr
↳ legacy rule would have alerted on #3 - suppressed (calibrated risk 0.08). Fewer false positives is the product.
ALERT · €9,850 → Vela Consult SIA · priority 1
calibrated risk score · precision@100: 0.42 vs 0.05 under legacy rules
Reason codes - feed the SAR narrative
🔴 a human decides - always. The model ranks, never files.
How it works
Connect in your environment
Read-only feed from your core banking / PSP data - on-prem or your VPC. Transaction data never leaves your infrastructure.
Ensemble scoring, calibrated
Supervised model on your confirmed cases + anomaly detection for novel patterns + counterparty-graph features. Scores mean what they say.
Reason codes per alert
Every alert carries the pattern that fired it, in analyst language - pre-filling the SAR narrative instead of a blank form.
Shadow-run before switch-over
2-4 weeks alongside your rules engine, comparing catch-rate and false positives. Legacy rules switch off one at a time, with compliance sign-off.
The Four Guarantees™ - this build
Measured value
~30% of currently-missed fraud caught + ~40% fewer false-positive review hours. Precision@k eval gate at your analyst capacity.
Defensible
SHAP reason codes per alert feeding SAR narratives; regulatory-grade hash-chained audit trail - every decision replayable for the supervisor.
Self-correcting
Fraud drifts fast - PSI/KS + concept-drift monitoring is core here, not hygiene. Analyst dispositions feed the retraining loop.
Yours & everywhere
On-prem / VPC deployment for regulated buyers; full source handover; MCP tools for your investigators' agents (🔴 disposition stays human).
The number, sized honestly
Reference buyer: Baltic mid-market bank or fintech, ~€500M flow, ~2M transactions/yr, AML team of 3-6 reviewing ~400 alerts/week at ~95% false positives.
Three ways to own it
| Tier | What you get | Price | |
|---|---|---|---|
| Scaffolding ★ | The full repo - ensemble + calibration + reason codes, regulatory-grade audit log, Annex IV draft, MCP server. Reference run on public AML data. | €2,990 | |
| PoC ★★RECOMMENDED | Trained on your transaction history - false-positive reduction and detection quality measured against your analysts' verdicts. Code + quality report yours. On a regulator-grade problem, no guarantee until your data proves it. | from €9,000 | |
| Implementation ★★★ | Production: live scoring with reason codes, case-tool integration, threshold monitoring, scheduled retraining, monthly precision report - the agreed number (conservative) attaches here. | from €27,000 |
★ = engagement depth. PoC is the recommended path: quality proven on your data before production money. The PoC carries no performance guarantee by design; the agreed number (conservative) attaches at Implementation, informed by the PoC report.
What we don't promise
Ready to see your own number?
Request the build: within 48h you get a personal reply with the value sized to your volume.
No commitment · reply within 48h · your data stays in the EU