Enable stitching

If you are going to use XDM schema fields for persistent ID or person ID, ensure that identities are marked properly in the schema for the event dataset. See Identity namespace overview.

Verify identity coverage for both persistent ID and person ID:

• Persistent ID

  Query 7 days of data where your persistent ID field is not null and divide by a query of 7 days of data for all events in your dataset. This percentage should be above 95%.

  Example of a query you could use for verification:

  code language-sql
  SELECT COUNT(*) AS total_events, COUNT({PERSISTENT_ID_FIELD}) AS events_with_persistentid, ROUND(COUNT({PERSISTENT_ID_FIELD}) / COUNT(*), 2) AS percent_with_persistentid_not_null FROM {DATASET_TABLE_NAME} WHERE TO_TIMESTAMP(timestamp, '{FORMAT_STRING}') >= TIMESTAMP '{START_DATE}' AND TO_TIMESTAMP(timestamp, 'FORMAT_STRING') < TIMESTAMP '{END_DATE}';

  Where:

  • {PERSISTENT_ID_FIELD} is the field for the persistent ID. For example: identityMap.ecid[0].
  • {DATASET_TABLE_NAME} is the table name for the event dataset.
  • {FORMAT_STRING} is the format string for the timestamp field. For example: MM/DD/YY HH12:MI AM.
  • {START_DATE} is the start date. For example: 2024-01-01 00:00:00.
  • {END_DATE} is the end date in standard format. For example: 2024-01-08 00:00:00.

• Person ID

  • For graph-based stitching, ensure that the identity graph contains fragments that link ID values from your chosen persistent ID namespace and person ID namespace. You could run a test by going to the Experience Platform Identity graph viewer and query the graph by some sample persistent ID values. Verify to see if these persistent ID values are linked to person ID values in the graph.

  • For field-based stitching, query 7 days of data where your person ID field is not null and divide by a query of 7 days of data for all events in your dataset. This percentage should ideally above 5%.

    Example of a query you could use for verification:

    code language-sql
    SELECT COUNT(*) AS total_events, COUNT({PERSON_ID_FIELD}) AS events_with_personid, ROUND(COUNT({PERSON_ID_FIELD}) / COUNT(*), 2) AS percent_with_personid_not_null FROM {DATASET_TABLE_NAME} WHERE TO_TIMESTAMP(timestamp, '{FORMAT_STRING}') >= TIMESTAMP '{START_DATE}' AND TO_TIMESTAMP(timestamp, 'FORMAT_STRING') < TIMESTAMP '{END_DATE}';

    Where:

    • {PERSON_ID_FIELD} is the field for the person ID. For example: identityMap.crmId[0].
    • {DATASET_TABLE_NAME} is the table name for the event dataset.
    • {FORMAT_STRING} is the format string for the timestamp field. For example: MM/DD/YY HH12:MI AM.
    • {START_DATE} is the start date. For example: 2024-01-01 00:00:00.
    • {END_DATE} is the end date in standard format. For example: 2024-01-08 00:00:00.

Person ID coverage: The coverage of the selected person ID used for identification during the stitching process (live and replay).

• For the best field-based stitching results, a person ID (user info) should be sent on at least one event for each persistent ID (device info).
• For the best graph-based stitching results, a (persistent ID, person ID) relation should be present in the identity graph for each persistent ID.

Person ID coverage is shown as a percentage and compared to what is recommended on a stable development or on a production setup. The higher this coverage value is, the better stitching results are obtained with the selected person ID.

Persistent ID coverage: This value is used for identification during the stitching process (live and replay), in case a person ID value cannot be detected. Events with no persistent ID and no person ID are dropped from the data. For best stitching results, a persistent ID should be present on all events.

Persistent ID coverage is shown as a percentage and compared to what is the minimum recommended on a stable development or on a production setup.