Request demo

Tara: Transaction Screening Analyst

This article focuses on the configuration, and running of Tara: Transaction Screening Analyst.

Banks and other financial institutions hire large teams to monitor real-time transactions and reduce false positive alerts. Tara optimizes the repetitive tasks involved in real-time transaction monitoring and allows analysts to focus on high-value work.

Tara is designed to ingest each alert generated by a financial institution’s sanctions screening system and she determines if it is a false positive. She helps to identify real-time transactions or payments to or from sanctioned individuals, entities, or jurisdictions, not only during onboarding but also throughout the customer relationship lifecycle that may pose a risk to financial institutions. 

Tara is also trained to escalate potential true hits to a pre-defined team within the institution. All flagged transactions or payments are then reviewed manually by analysts to clear false positives.

Configure Model

Input

Tara allows input submission through WorkFusion APIs, and Flat CSV file uploads as shown below:
 

Model Selection

In the Prepare Your Use Case box, the ML model tab allows users to select a PSS classification model for the Business Process execution. It shows all the available models. By default, it selects the most recent version. 
 

Output

Tara generates an understandable and consistent audit trail report for review on a single screen. 
 

Section 2: Start Screening

The following steps to start the Transaction Screening from the Control Tower. 
  1. In the Control Tower menu, click Business Processes, and then click the name of the required process to expand the process definition.
  2. Next, click the process name. The BP window appears.

3. Go to the Data tab, and click Upload Data. The Upload your main data window appears.
4. In the window, click Add, select the CSV input file to be screened.
5. After the parameters are saved, go to the Run tab, and click Run This Process.
The information window shows a message ”Business Process started” successfully.
 

Section 3: Review results manually

1. The messages and hits analytics tables store execution results, and contain decisions based on their contents.
2. HTML reports are available on Minio if the user chooses to produce them.
3. Review the screening results.

Section 4: Reference Input and Output data

Input data

Tara supports the following types of input for transaction screening:
  • Streaming API Input
  • File Input
  • Template flat file CSV Input

Streaming API Input

The system accepts input data from the built-in streaming service. Users can define and update the API endpoint. Business processes stay alive even when all records are processed.

Check the box to use the specified external source

File Input

This input feature supports the file upload feature. The business process will finish when all records are processed.
  • File-in-original format contains a JSON request
  • This format requires two columns for execution; one is a request_json_object and another is an optional column which is a decision_gold_json_object.
  • The File-in-flat format contains an unfolded request_json_object.
  • The business process treats each row as a separate request.

Using CSV flat-file Templates for Input

Using a CSV flat file as the template is the easiest input method for uploading input data for Transaction Screening.

image

 
 

Request Datastore

Tara accepts standardized input in JSON-formatted strings and saves it into the “pss_request_processing_v1” datastore.

Example of pss_request_processing_v1 datastores:

Column

Description

system_id

request_id

request_json_object

decision_json_object

decision_gold_json_object

message_json_object

request_insert_time

last_updated_time

decision

status

client_id

DB generated ID

Unique UUID generated before adding record to datastore

Original customer data for adjudication in JSON format representing a Message object.

Model Decision in JSON format representing the MessageDecision object.

Gold Model Decision in JSON format representing MessageDecision object. This is used for model training and calculating statistics.

Processed or enhanced copy of customer data for adjudication in JSON format representing a Message object. Data from this column will be sent to the model for processing.

Insert data timestamp.

Timestamp of last update.

Final Message decision

Processing status. This is updated during record processing.

User who sent the request for processing

Output data

The table below specifies Tara Business Process output data fields:

Alert Classification Model

Field

Description

Type

model_result

Model response with structured data

Category: Reason code (Different for each client)

Explanation: Human-readable decision explanation

Score: Classification model confidence of the alert to belong to the predicted topic

Model-decision attribute: – contains full model decision in JSON format

Json