New customer — account opening. Screening across World-Check, Dow Jones, ComplyAdvantage, EU/UN/OFAC lists. Low-confidence match: name similar to a watchlist subject, personal data not consistent. No automatic action. Alert for manual review (48h).
AGENT · AML SCREENING REAL-TIME
AML screening on customers and transactions happens in real time.
AML Screening Real-Time compares customers and transactions against the main commercial watchlists (World-Check, Dow Jones, ComplyAdvantage) and against public lists (EU, UN, OFAC). Immediate alert to the AML team for positive matches, automatic transaction freezing pending review.
AML Screening Real-Time at work.
Review complete. Documents consistent, data differ from the watchlist subject. False positive — closing with reason. Account opening authorised.
Decision recorded in the AML registry. Evidence available for regulatory inspection.
Why it exists.
Anti-money laundering is one of the most stringent obligations in banking. EU AML directives — the AMLD series and the forthcoming AMLR, with AMLA as the new supervisory authority — require customer screening at relationship opening and continuous transaction screening against national and international watchlists. The volume of customers and transactions at mid-to-large banks makes manual screening unworkable.
How it screens in real time.
The agent activates on the two core events: a new customer at account opening and a new transaction above the configured threshold. For each event it identifies the parties involved, compares them against the configured commercial and public watchlists, and classifies positive matches by confidence. High confidence: immediate alert and automatic freezing. Intermediate confidence: alert for manual review, transaction on hold.
The final decision stays with AML.
The final decision — authorise, block, close the relationship, file a suspicious activity report — stays with the bank's AML team. The agent sets the table for screening, classifies by confidence, prepares the evidence for regulatory inspection. Filing the SAR stays with the team.
The functions that govern the bank's AML process.
AML officer
Recovers time from manual screening — the most voluminous work in the AML team. Capacity shifts to evaluating matches that require judgment, not to mechanical screening execution on account-opening volumes.
Customer Acme Capital S.p.A.
World-Check no match
Dow Jones no match
ComplyAdvantage no match
EU · UN · OFAC no match
Screening complete · audit recorded
Compliance officer
Has the inspectable trace of every screening for regulatory inspection on the quality of the AML process. Every screening, every match, every team decision is in the registry, readable with a standard SQL client.
KYC case KYC-2024-091 In verification
ID document VERIFIED
Biometrics · SCA PSD2 LINKING OK
Beneficial owner BUSINESS REGISTRY
Case forwarded to AML Screening
Risk manager
Sees aggregate AML exposure. Suspicious activity reports are grounded in structured evidence — customer data, relevant transactions, suspicion rationale, screening evidence — not in scattered notes.
Counterparty check 3 entities
Customer · Acme Capital no match
Beneficial owner · M. Rossi no match
Counterparty · Aurora Ltd verify
1 multi-source verification required
One new customer, screening across five watchlists, one traced decision.
Account opening, screening across five watchlists.
At a mid-size bank, a new customer opens a current account. The agent activates. It runs screening across all configured watchlists (World-Check, Dow Jones, ComplyAdvantage, EU lists, UN lists, OFAC). It identifies a low-confidence match: name similar to a watchlist subject, personal data partially consistent.
Low confidence: no block, manual alert.
Classification: low-confidence match, no automatic blocking action. Event recorded in the audit registry. Alert to the AML team for manual review within standard procedures (48 hours). Integration with commercial watchlists is built during delivery.
False positive confirmed, account opening authorised.
The AML officer evaluates: the customer presents consistent documents, the residence information does not match the watchlist subject. The officer closes the match as a false positive with documented reason, authorises the account opening, and records the decision in the AML registry. The full event stays in the audit registry for regulatory inspection.
Declarative AML rules written by the bank's AML team.
The rules are declarative. The bank's AML team defines in readable format which watchlists to query, confidence thresholds for each match category, automatic freezing rules, and suspicious transaction patterns. Rules live in the bank's repository, versioned, and validated at agent startup. Prerequisite for operation: the bank must have an active subscription to the chosen commercial watchlist provider (World-Check, Dow Jones, ComplyAdvantage, LexisNexis, Refinitiv, or equivalent). Integration with commercial watchlists and with the bank's management systems is built during delivery via dedicated adapters.
- Language
- TypeScript (Node.js)
- LLM model
- customer's choice: Anthropic, OpenAI, Mistral, open source models hosted internally, AWS Bedrock for a private model
- Built-in controls used
- pii-detector, credential-detector, tool-param-validator
- Native channels
- Slack, Telegram, OpenAI-compatible HTTP
- Prerequisite: active bank subscription
- watchlist provider (World-Check, Dow Jones, ComplyAdvantage, LexisNexis, or Refinitiv)
- Commercial watchlist integration
- dedicated adapter built during delivery
- Public watchlist integration (EU, UN, OFAC)
- dedicated adapter built during delivery
- Bank management and AML system integration
- dedicated adapter built during delivery
- AML rules
- declarative, versioned, written by the AML team
- Memory
- persistent per instance, pgvector + PostgreSQL FTS on historical patterns
- Registry
- append-only, queryable with a standard SQL client (AML regulatory audit-ready)
Frequently asked questions about the agent.
For high-confidence matches at a risk level configured as critical, yes: automatic freezing pending the AML officer's review. For intermediate or low-confidence matches, the agent records and sends an alert; the blocking decision stays with the team. The automatic freezing rule is declarative and configured by the bank's AML team.
The AML officer evaluates manually, closes the match with a documented reason, and records the decision in the AML registry. The reason is inspectable in regulatory audit. The agent tracks the customer's false positive patterns (e.g. common names that frequently generate partial matches) to support calibration of confidence rules.
No. Filing a suspicious activity report to the relevant financial intelligence unit is the AML team's responsibility, following the bank's procedures. The agent provides structured material — customer data, relevant transactions, suspicion rationale, screening evidence — to support the reporting decision.
The typical pattern for AML Screening Real-Time is 16-22 weeks. Discovery 3-4 weeks (mapping AML rules, watchlist provider, management system), writing rules with the AML team 5-6 weeks, commercial watchlist and bank management system integration 6-8 weeks, compliance officer validation 2-3 weeks, hand-off 1-2 weeks.
From a 30-minute conversation to the squad in production.
A 30-45 minute conversation to understand how AML Screening Real-Time would configure to the bank. Which watchlist provider, which management system, which confidence thresholds for the AML process.