PiR2-IT
Start an engagement
PiR2-IT
Start an engagement
An agentic AI and hyperautomation platform for anti-money laundering. Specialised agents enrich alerts, evaluate typologies, resolve beneficial ownership and draft investigation and SAR narratives, while every consequential decision stays with a human. Implemented in 16 financial institutions.
Alerts and transactions from core banking, SWIFT/SEPA and the existing monitoring system are streamed through an event bus and normalised into one standard payload.
A supervisor agent routes each case to specialised agents — enrichment, typology and risk, narrative and case packaging — with short-term state and a vector memory of past resolutions.
Agents call governed tools over APIs / MCP: customer profile, transaction ledger, sanctions and PEP screening, counterparty graph and adverse media.
High-confidence false positives are closed with a full rationale; everything else goes to an L2 analyst as a pre-filled case file. Every step is logged and explainable.
An alert from the monitoring system is normalised into a standard payload: alert ID, rule, entity, transaction and timestamp.
The enrichment agent pulls the KYC profile, 30/60/90-day ledger history and the counterparty graph in parallel.
The typology agent checks the business rationale, re-screens sanctions and PEPs and scores known laundering typologies.
The narrative agent writes an audit-proof summary: trigger, findings, risk factors and recommended disposition.
Auto-close only at ≥95% confidence and low risk; otherwise escalate to an L2 analyst with the full case file.
Analysts and the MLRO take the decisions that matter — escalations, SARs and high-risk customers.
Immutable logs of every step; regular benchmark tests on precision and recall.
Baselines, typologies and rules are updated as behaviour and regulation change.
Agentic AML Hyperautomation is a PiR2-IT product in production and support, implemented in 16 financial institutions and continuously refined. Agents prepare evidence and recommendations; suspicious activity reports, high-risk escalations and model changes are approved by people.
References, case studies and demonstrations are available on request, under NDA.
No. It works on top of the existing system (for example Actimize, Oracle Mantas or NetReveal): alerts are streamed in, and agents do the enrichment, triage and case preparation.
Only when confidence is at least 95% and the risk score is below the institution’s threshold. Everything else is escalated to an L2 analyst with a pre-filled case file. If a tool fails or times out, the case is escalated by default.
A compliance officer. Agents draft the narrative, citing transaction IDs, rules triggered and typologies; a person reviews, signs off and submits.
Every agent reasoning step, tool call and model prompt/response is written to an immutable audit log, and the models are tested regularly against historical false positives and confirmed SARs.
In 16 financial institutions. Client names are confidential; references are available on request, under NDA.
Yes. An interactive demo on synthetic data runs at aml.pir2-it.com: alert triage by agents, L2 case files, SAR drafting with MLRO approval, UBO resolution and a hash-chained audit log. A demonstration on your own data runs under NDA.
PiR2-ITIndependent consulting for institutions, regulated sectors and high-trust programmes.