The Silent Segment Split
A retailer's profile dataset merges CRM records and web events into a single XDM schema. Overnight, an identity namespace change causes 18% of profiles to become 'orphaned' - events land on anonymous ECIDs while authenticated CRM IDs sit on separate profiles. The segment 'High-Value Repeat Buyers' (purchase count >= 3 in 90 days) drops from 240,000 to 61,000 members. The nightly batch segment job runs at 02:00; the streaming segment for cart abandoners still fires correctly. The data team confirms no schema change and no new source connectors. Marketing wants to 'just re-run the segment' before a campaign launches in 6 hours.
1. Diagnose why the batch segment collapsed while the streaming segment still works, and decide whether re-running the segment will fix it.
Read the worked answer
The symptom split is diagnostic: streaming segments evaluate events in real time against whatever profile the event stitches to, so they keep working even when identity resolution is degraded. Batch segments evaluate the merged profile store, so orphaned profiles mean the purchase-count attribute is fragmented across two profiles and neither reaches the threshold. Re-running the batch job re-evaluates the same broken profile graph - it will produce the same 61,000. The root cause is identity stitching (namespace priority/merge policy), not segment logic. The fix is to correct the identity configuration and let profiles re-merge, then re-evaluate. Re-running first is wasted effort and risks masking the real problem.
2. Marketing insists on launching in 6 hours. What is your recommendation and what do you tell them?
Read the worked answer
Do not promise the full 240,000. The honest position: the segment is under-counting because profiles are fragmented; launching to 61,000 is safe but under-delivers. Options: (1) ship to the current 61,000 as a conservative audience, (2) delay 24h to fix identity and re-merge. I would not 'pad' the segment with an OR condition on anonymous IDs - that risks duplicate messaging to the same human (one ECID profile, one CRM profile) and inflates frequency. The developer's job is to surface the risk, quantify it (18% orphaned, ~179K missing members), and let the business choose. Document the decision.