Skip to main content
Version: 4.0.0

Analyze effectiveness

Tara uses prebuilt analytics dashboards to monitor specific metrics for actionable insights into a Business Process (BP).

Dashboards​

To view Tara dashboards, go to Control Tower > Analytics > Dashboards > Tara: Payment sanction screening overview.

note

The Payment sanction screening overview dashboard calculates metrics for finished transactions only. Thus, not all transactions are considered on the dashboard during BP execution.

The out-of-the-box analytics includes the following business metrics:

  • Alerts trends
    • Resolve
    • No decision
  • Hits trends
    • Resolve
    • No decision
  • Alert automation
    • Alert automation rate
  • Hit automation
    • Hit automation rate
  • Average hits per alert
  • Average time spent
  • Hits
    • True positive
    • False positive
    • Missed false positve
    • Missed true positive
  • Alerts
    • True positive
    • False positive
    • Missed false positve
    • Missed true positive
  • Hit decision code distribution
  • Content format analysis
    • Resolve
    • No decision
    • Error
  • Entity type analysis
    • Resolve
    • No decision
  • Tag name analysis
    • Resolve
    • No decision

  • True positive: both the gold decision and the AI Agent's decision are NO_DECISION.
  • False positive: both the gold decision and the AI Agent's decision are RESOLVE.
  • Missed true positive: the gold decision is NO_DECISION, but the AI Agent's decision is RESOLVE.
  • Missed false positive: the gold decision is RESOLVE, but the AI Agent's decision is NO_DECISION. The automation rate is calculated as the number of false positives divided by the sum of missed false positives and false positives, provided that gold values are available. Otherwise, it is calculated as the count of RESOLVE divided by the total number of cases.

Data Stores​

Alert-level analytics Data Store​

The pss_messages_analytics_v4 Data Store contains alert-level analytics data.

The available columns are as follows:

  • campaign_uuid: unique identifier of the BP installation
  • run_uuid: unique identifier of the BP execution
  • transaction_uuid: unique identifier of a request
  • client_id: client executing BP
  • message_id: identifier of an incoming alert
  • message_type: type of an incoming alert
  • alerts_count: number of hits in the alert
  • message_decision: alert decision as a JSON string
  • message_review_decision: alert gold decision as a JSON string
  • report_link: generated link of an HTML report
  • last_modified: timestamp indicating when a record was last modified
  • filtered_out: flag if alert was filtered out
  • decision_reapplied: flag if a decision on the alert was made by decision reapplication
  • model_id: decision model identifier
  • ner_model_id: Named Entity Recognition (NER) model identifier
  • request_insert_time: timestamp indicating when a record was received
  • status: completion status of a record
  • tara_version: Tara version used during request execution
  • message_subtype: value from the message subtype field

Hit-level analytics Data Store​

The pss_alerts_analytics_v4 Data Store contains alert-level analytics data.

The available columns are as follows:

  • campaign_uuid: unique identifier of the BP installation
  • run_uuid: unique identifier of the BP execution
  • transaction_uuid: unique identifier of a request
  • client_id: client executing BP
  • message_id: identifier of an incoming alert
  • alert_id: identifier of an incoming hit
  • entity_type: type of a screened entity
  • alert_decision: final hit decision
  • alert_decision_code: hit decision code or a name of the rule executed to make a decision
  • alert_decision_description: hit decision description; currently, the same as the code
  • alert_decision_explanation: hit decision explanation, a verbal detailed explanation of the code
  • alert_review_decision: gold hit decision
  • last_modified: timestamp indicating when a record was last modified
  • filtered_out: flag if a hit was filtered out
  • decision_reapplied: flag if a decision on the hit was made by continuous learning
  • model_id: decision model identifier
  • ner_model_id: NER model identifier
  • content_format: message content format
  • tag_name: name of the tag where the hit occurred

Enrichment data analytics Data Store​

The pss_de_analytics_v4 Data Store contains enrichment data analytics.

The available columns are as follows:

  • type: enrichment data provider
  • time: call execution time
  • message_id: identifier of an incoming alert
  • num_hits: total number of hits in an alert
  • num_hits_searched: number of hits that executed the call
  • num_hits_decision_changed: number of hits for which the decision was changed because of the call
tip

Also, see View CLDS Data Stores.

Last updated on