'At Risk' Subscribers: When the Segment Container Changes the Headcount
A streaming service with 4.2 million subscribers has built its churn-prevention programme on one CJA segment called 'At Risk': watched fewer than 120 minutes of content in the last 30 days AND visited the cancellation page. It was built at the hit container. The retention team reports 310,000 at-risk users, but the CRM team running the same logic on raw data counts 640,000. The programme also depends on a calculated metric, 'Engagement Rate', defined as content plays divided by sessions, which returns values above 100 for some users and blanks for others. An audience built from the segment and published to Experience Platform for email activation refreshes every 12 hours, yet the campaign team says the audience file is always smaller than the Workspace number. Leadership wants the programme defended in next quarter's budget review.
1. Explain the container error behind the 310,000 versus 640,000 gap and specify how you would rebuild the segment.
Read the worked answer
Two errors compound: container and time window.
Container. A hit-level match on 'watched < 120 minutes AND visited the cancellation page' only returns hits where both conditions are true simultaneously - effectively a cancellation page view that also carries watch-time, which almost never exists. That is why the count is roughly half the CRM figure. Confirm by rebuilding the same logic at person level; the delta should collapse immediately.
Rebuild at person container: include persons where total content minutes in the last 30 days < 120, and at least one cancellation-page event occurred in that same 30 days. If order matters (watched, then visited cancellation), use a sequential segment rather than a flat AND - the sequence is a better churn signal than co-occurrence.
Scope it. Add a subscriber-status condition at person level, otherwise already-churned users inflate the population and the retention team targets people who have left.
Version and freeze the definition. A segment used to defend budget must not change silently between quarters; document the logic and the date it was locked.
2. The 'Engagement Rate' calculated metric is returning values over 100 and blanks. Diagnose it and propose a defensible replacement.
Read the worked answer
'Content plays / sessions' is not an engagement rate; it is a plays-per-session ratio, and it breaks in three predictable ways.
- Zero denominator. Persons with plays but no session in the data view produce blanks or division errors. Guard the metric with a conditional that returns 0 when sessions = 0.
- Container mismatch. Plays counted at hit level and summed over sessions double count when a play event also starts a session. Numerator and denominator must share a compatible container.
- Values above 100 mean the denominator is undercounted - usually the app does not send session-start into the same dataset as the play event. Check the app dataset's session definition before adjusting the metric.
Better definitions, pick one: a person-level rate - persons with at least one content play divided by persons with at least one session, bounded 0–100% - or average minutes per active day, which is more interpretable for a streaming business.
Document the chosen formula in the component description and add a guardrail segment for denominator-zero persons so the blanks stop being invisible.
3. Why is the published audience smaller than the Workspace segment, and what do you tell the campaign team?
Read the worked answer
An audience is not a segment with a different button.
Likely causes, in order of probability:
- Identity. The audience publishes against a specific identity namespace (for example hashed email) and only persons resolving to that identity are mailed; the Workspace segment counts everyone with an ECID.
- Evaluation cadence and lookback. A 12-hour refresh evaluates at a point in time with its own lookback, so people who qualified 13 hours ago are missed and people who no longer qualify are dropped.
- Data view scope. If the audience is built on a view with different component settings or identity configuration, the population differs by construction.
- Ingestion latency. The file the CRM receives reflects an earlier evaluation than the number someone read in Workspace this morning.
Diagnose by running the segment in Workspace with the same identity namespace and the same lookback.
Then tell the campaign team the smaller number is expected and correct: the gap is the unaddressable portion. Report it as a metric - addressable reach rate - rather than letting each team argue about whose number is right.