How many AML typologies are covered out of the box?
130+ typologies across structuring, layering, smurfing, mule networks, trade-based laundering, cuckoo smurfing, money muling, terrorism financing, sanctions evasion, and proliferation-finance — each mapped to FATF, AUSTRAC, EgmONT, and MAS PMLA typology references.
What's the typical false-positive rate after AI triage?
Pre-triage rules-only alerting typically runs at 90–98% false positives in the industry. WIDTH's AI triage closes the obvious-true-negatives layer and reduces analyst-queue noise by 55–70% in production deployments while maintaining a documented 1:1 audit trail.
Does WIDTH support model risk governance (e.g., SR 11-7 / MAS FEAT)?
Yes. Every model that scores a transaction or alert is versioned, A/B-testable in shadow mode, and shipped with a model card covering training data lineage, feature importance, monitoring drift metrics, and the human-review thresholds. Aligned with US SR 11-7, MAS FEAT, IMDA AI Governance and EU AI Act high-risk requirements.
How are typology rules customised per jurisdiction?
Each rule lives in a versioned policy bundle. You can fork the global library per legal entity / jurisdiction and override thresholds, observation windows, look-back periods, and exemptions. AUSTRAC TTR, FinCEN CTR, and HKMA STR-specific thresholds ship as defaults.
What kind of audit trail is produced per alert?
Every alert carries: the underlying transactions, the rule + model + policy version that fired, the AI triage rationale, the human reviewer ID, the disposition, and the supporting evidence. The export is regulator-ready (PDF or signed JSON) and stable across replays.
Can we integrate our own watchlists and risk segments?
Yes. Customer risk segments, internal blocklists, PEP refresh lists, and any structured risk attribute can be loaded via API or batch SFTP and consumed directly by the typology rules and the AI triage layer.
