The Destination That Delivered Yesterday's Audience
A travel company activates a Cart Abandoners - 24h audience to a demand-side platform (DSP) and to an on-site personalization tool. The DSP destination is configured for batch activation (daily file drop); the personalization tool uses streaming/edge activation. The audience rule is abandoned_cart = true AND last_cart_event < 24 hours ago. Marketing reports that cart-abandon emails arrive 30+ hours after abandonment and include users who already purchased. The personalization tool, by contrast, behaves correctly. Spend is significant and the DSP charges per record.
1. Explain the behavioral difference between the two destinations and why the DSP audience is stale and over-inclusive.
Read the worked answer
The two destinations have fundamentally different activation cadence:
- Streaming/edge (personalization tool): evaluates audience membership close to real time and reflects current profile state. When a user purchases, membership drops quickly; when they abandon, they qualify quickly. Hence correct behavior.
- Batch (DSP): the audience is evaluated and exported on a schedule (e.g., daily). That introduces two failures:
- Latency: a user who abandons at 23:00 may wait until the next batch run + DSP ingestion → 30+ hours.
- Staleness/over-inclusion: the export is a snapshot. A user who purchased after the snapshot but before the DSP acts is still in the file. The DSP has no way to know they converted unless you send a removal/negative audience.
The 24h rule is also evaluated against last_cart_event - in a batch model, the clock and the export time diverge, so the rule's intent (fresh abandoners) is not honored at delivery time.
Conclusion: the destination architecture, not the rule, is the root cause. A 24-hour time-boxed audience is a poor fit for daily batch activation.
2. Propose a remediation plan that respects the DSP's batch constraint and reduces wasted spend.
Read the worked answer
You can't make a batch DSP real-time, so design around the cadence:
- Add a suppression/negative audience: export a
Purchased in last 48haudience to the same DSP as an exclusion. This is the single highest-impact fix for over-inclusion and wasted spend. - Shorten the window and align to batch: if the DSP runs daily, a 24h rule guarantees some records are already stale at delivery. Consider a rule that tolerates the cadence (e.g., abandon in last 12h evaluated at export) and time the export so records are fresh when the DSP acts.
- Increase batch frequency if the destination supports multiple daily runs - reduces latency without changing architecture.
- Move time-sensitive abandonment to a streaming destination (e.g., the personalization tool or an ESP with real-time triggers) and reserve the DSP for broader prospecting where 24h precision isn't required.
- Monitor cost: track records delivered vs. converted; the per-record DSP charge makes suppression directly ROI-positive.
Judgement: match the audience's time sensitivity to the destination's activation method. Don't put a 24-hour audience on a 24-hour batch pipe and expect freshness.