• Version 1.5 [2024-09-25]
  • Split event and user property Measurement Protocol payloads: Send event and user property Measurement Protocol payloads separately, allowing for proper event session attribution and audience segmentation.

• Version 1.4 [2024-09-17]
  • Event Inclusion in Measurement Protocol Payload: Updated the Measurement Protocol payload to include the event and predictive value for Value-Based Bidding (VBB) scenarios by default. This enhancement allows the pipeline status to be reflected directly in Google Analytics.
  • Cron Job Update to 6-Hour Interval: Adjusted the default prediction pipeline cron job frequency to run every 6 hours. This update improves pipeline efficiency, especially during intermittent failures, and benefits users leveraging streaming or fresh daily exports, enabling near real-time analysis capabilities.
  • Up to 30-Hour User Scoring Lookback Window: Modified the user scoring process to consider up to a 30-hour lookback window, rather than a single table partition. This change ensures that all users are scored, accounting for potential rollover periods between prediction pipeline runs, particularly for simultaneous streaming and daily export users.

• Version 1.3 [2024-09-16]
  • Remove event and session ID from payload: The event and its parameters have been removed from the Measurement Protocol payload, leaving only the user property. This change ensures that the payload is sent silently without impacting reporting. The user property will continue to be set for all users to support audience segmentation.

• Version 1.2 [2024-08-14]
  • 24-Hour Data Filtering: Added a new filtering mechanism that selects only the users within a 24-hour window for aggregation. This logic ensures that only recent users are considered for component analyses, which are then used by CRMint's measurement protocol worker to send events to Google Analytics.
  • User Scoring Data Insertion: Implemented a new query that inserts user scoring data into the scored_users_log table. This data includes user IDs, timestamps of the last scoring event, user property values, and event values.
  • Hourly Prediction Pipeline Runs: The prediction pipeline now runs every hour instead of every day by default. This change is intended to help score users more quickly within the 24 hour session attribution window.

• Version 1.1 [2024-06-20]
  • Optimized Table Selection Logic: Refined the logic for identifying the correct table for training and prediction when both Daily and Streaming exports are enabled.

• Version 1.0 [2024-05-17]
  • Exact Matching Enhancement: Transitioned the matching logic for Event, Product, Churn, and User Property marketing objectives from regular expressions to exact matching. This refinement ensures precise targeting and accurate identification of relevant data points for analysis.
  • Random Timestamp Offset Implementation: For scenarios where new user properties have not yet been added to the BigQuery export table, the measurement protocol timestamp offset is now randomly set between 5 and 100 microseconds. This adjustment enhances the accuracy of audience refresh processes by preventing potential timestamp conflicts in new pipelines.

• Version 9 [2024-05-09]
  • Decoupled Event Data Processing: Segregated the logic for extracting the latest event table date, visitor IDs, session IDs, and timestamps. This adjustment boosts pipeline transparency and operational clarity. Accompanied by explanatory comments for an in-depth understanding of the functionality.
  • Advanced Timestamp Allocation: Introduced a dynamic calculation for the timestamp_micros field within the measurement protocol payload, leveraging an alphabetical ordering of user properties for offset determination. This enhancement ensures a more precise assignment of event timestamps, vital for the accurate refresh of audience segmentation membership.
  • Performance Insights: Implemented a new job to analyze conversion rates based on user property predictions, segmented into High, Medium, and Low categories. This enhancement provides valuable insights into the model's effectiveness in identifying high-value users.

• Version 8 [2024-05-03]
  • Enhanced GA4 Event Tracking: Incorporated session_id into the GA4 measurement protocol payload, enabling direct association of GA4 events with their session source/medium within the GA4 UI for improved data coherence and user journey analysis.
  • Refined Instant Vertex Pipeline Timing: Adjusted the timing of Instant Vertex pipelines by adding +2 microseconds to timestamp_micros, strategically designed to prevent data collisions with Instant BQML pipelines, thereby enhancing the reliability of audience refresh logic.

• Version 7 [2024-04-01]
  • Streamlined Event Propensity Pipeline: Eliminated the automatic generation of conversion events within the Event Propensity pipeline. This modification is intended to minimize confusion among users not engaged with Value-Based Bidding (VBB). The feature remains accessible and can be activated for tasks related to VBB as required.

• Version 6 [2024-02-29]
  • Introduced GA360 Fresh Daily Export Support: Added support for the GA360 Fresh Daily export option.

• Version 5 [2023-11-21]
  • Refined wildcard query logic to target 'events_20*' prefix, reducing the inclusion of views and enhancing query specificity.
  • Updated all date format specifiers in BigQuery queries from %Y (four-digit year) to %y (two-digit year) to align with the revised wildcard filter logic, ensuring accurate table selection and data querying consistency.
  • Implemented the boosted tree regressor algorithm for BigQuery Machine Learning models. Hyperparameters have been optimized, including setting BOOSTER_TYPE to DART and adjusting SUBSAMPLE to 0.7 for improved generalizability.

• Version 4 [2023-11-07]
  • Remove session_id and engagement_time_msec from GA4 measurement protocol payload.
  • Remove session_id calculations from Calculate Fields query.

• Version 3 [2023-10-24]
  • Updated the Event & Churn propensity training, prediction, and value-based bidding logic to include a regular expression filter. This filter omits events with names ending in '_iBQML' or '_vertex', which are associated with propensity score calculations, ensuring they are not included in the analysis.
  • Set data split method to `AUTO_SPLIT` in BigQuery ML training query hyperparameters.

• Version 2 [2023-08-07]
  • Updated audience boundaries job to define boundaries based on AUDIENCE_SEGMENTS variable, not user counts.

• Version 1 [2023-06-27]
  • Create Version Changelog to manage pipeline upgrades.
  • Remove conversion event creation from non-Event Propensity marketing objectives.