Custom data science for profile enrichment blueprint

The custom data science for profile enrichment blueprint illustrates how data can be used to train, deploy, and score models to provide machine learning insights into Experience Platform and the Real-Time Customer Data Platform from data science and machine learning tools.

Modeled insights can be ingested into Experience Platform to enrich the real-time customer profile. Examples of machine learning insights include lifetime value scoring, product and category affinity, propensity to convert, or propensity to churn.

Use cases

  • Extract insight and discover patterns from customer data, train and score models from this data.
  • Enrich the Real-time Customer Profile with model driven insights and attributes for more granular personalization and optimized journeys.
  • Train and Score models to determine customer insights such as customer lifetime value, propensity to convert or churn, product and content affinities, and engagement scores.


Reference Architecture for the Custom Data Science for Profile Enrichment Blueprint


  • For detailed guardrails and end to end latencies on ingesting data science results into Experience Platform and the Real-time Customer Profile refer to the data ingestion guardrails and latency diagram referenced in the deployment guardrails document.

Implementation steps

  1. Create schemas for data to be ingested.
  2. Create datasets for data to be ingested.
  3. Ingest data into Experience Platform.

For model results to be ingested into Real-time Customer Profile be sure to do the following prior to ingesting data:

  1. Configure the correct identities and identity namespaces on the schema to be sure that ingested data can stitch into a unified profile.
  2. Enable the schemas and datasets for profile.

Implementation considerations

  • In most cases model result should be ingested as profile attributes and not experience events. The model results can be a simple attribute strings. If there are multiple model results that are to be ingested, it is recommended to use an array or map type field.

  • The daily profile snapshot dataset which is a daily export of the unified profile attribute data can be leveraged to train models on profile attribute data. Profile snapshot dataset documenation can be accessed here.

  • For extracting data out of Experience Platform the following methods can be used

    • Data Access SDK

      • Data is in raw file form
      • Profile experience event data remains in its un-unified raw state.
    • RTCDP Destinations

      • Profile attributes and segment memberships can be egressed.