- Overview
- Financial Crime Compliance (FCC) solutions
Financial Services Solutions user guide
Tara - Transaction Screening Alert Review
Tara, the Transaction Screening Alert Review agent, detects real-time transactions or payments to or from sanctioned individuals, entities, or jurisdictions during onboarding and throughout the customer relationship lifecycle, helping to mitigate risks to financial institutions. Screening primarily occurs on the government-issued sanctions lists, such as the Office of Foreign Assets Control (OFAC), but can also include other external and internal bank sanctions lists.
Financial institutions use sanctions screening or watchlist filtering technologies to compare inputs, including transactions, with items on a sanctions list. The outputs of those technologies are alerts or hits. The agent is designed to ingest each alert generated by a sanctions screening system and determine if it is a false positive by examining each hit within each alert.
Tara, the Transaction Screening Alert Review agent, supports multiple input methods, including API input, uploaded CSV files, and data from a connector. It is also designed to generate output results and reports, including HTML, quality control, and decision reapplication reports. In addition to determining whether an alert is a false positive, a detailed narrative is included in the response for audit trail and justification purposes.
Watch a quick demo of how Tara, the Transaction Screening Alert Review agent, works!
High-level processing flow
At a high level, Tara, the Transaction Screening Alert Review agent, processes alerts through the following steps:
- Ingestion: retrieves input data from loaded files, API calls, or a connector and ingests alerts, hits, message information, and watchlist entity data.
- Data Parsing: parses the payment message content according to its specified format. This step searches through message hits, matches screened entities with message content, composes entity fields using the type's dictionary, and updates message objects for further processing.
- Data enrichment (optional): connects to information outside of the user's sanctions screening system to obtain additional data to assist with alert review.
- Machine Learning Model: the Named Entity Recognition (NER) model and the Decision model support adjudication at the message and hit levels. The Decision model takes alert input and returns decision features for further adjudication. This component also relies on a rule engine that applies default or custom rule files to support adjudication based on business requirements.
- Decision and justification: makes an adjudication decision at the message and hit levels, determining whether hits are false positives or require escalation. This step can include a detailed narrative in the response.
- Screening system/ case manager: returns a response for the connector or API after the final decision is made.
Key agent components
Tara, the Transaction Screening Alert Review agent, is composed of several core components that support your workflows:
- Connectors: provide data through a connector input method and support integrations through the LexisNexis Bridger XG connector, Firco Continuity connectors, and enrichment connectors such as Orbis and GLEIF.
- Alert Parser: based on the format of incoming alerts and hits, identifies and parses key inputs for the models, including messages and watchlist entities.
- Configuration: UI-based configurations for rules, input type, model selection, data enrichment options, reapplication data, and output.
- Decision Reapplication (optional): checks whether a decision on the same payment message already exists in historical data and, based on configurable rules, reapplies that decision when allowed. This enhances the efficiency and consistency in alert processing.
- Machine Learning Pipeline: uses the NER model, the Decision model, and rules to produce adjudication decisions at the message and hit levels.
- Analytics: uses pre-built dashboards to monitor execution metrics and provide actionable insights into alert trends, hit trends, automation rates, average hits per alert, average time spent, and decision distributions.
Model input
The models rely on three primary categories of input data:
- Payment Message: payment message content parsed according to its specified format, such as FUF, SWIFT MT messages, SWIFT MX messages based on ISO 20022, and SWIFT NPP messages.
- Alert/ Hit: alert and hit data processed together with payment message content and watchlist entity information.
- Watch List Entity and Supporting Information: name, identifying data, entity type, and supporting information for the watchlist entity associated with each hit.
Model overview
Tara, the Transaction Screening Alert Review agent, includes two models that are part of the payment sanctions screening execution pipeline:
- NER model: uses natural language processing and deep learning to identify entities in tag content and classify them by type, returning the results to the execution flow.
- Decision model: contains a proprietary Name Matcher submodel, a libpostal address parser submodule, an ID matching submodule, and more. The model takes alert input and returns decision features to the execution flow.