Re-engagement campaign paying for users who would return anyway

The reported $3.10 CPA is an attributed cost, not an incremental one: any targeted lapsed user who opens the app inside the 7-day window is credited to the campaign, including users who would have returned unprompted - and the 18% organic lift from the content release makes that very likely here. Test it:

• Geo holdout. Pick matched metro pairs with similar install base and historical reactivation rates. Run the campaign in half, pause it in the other half for four weeks, and compare reactivations per lapsed user between cells. Incremental CPA = spend ÷ (reactivations in test cell − reactivations in control cell, scaled to equal audience size).
• Use a platform-run conversion lift experiment if available for re-engagement; it achieves the same separation without manual geo matching.
• Segment by lapse depth. Compare the 30–60 day lapsed cohort with the 90+ day cohort. Users lapsed 90+ days are far less likely to return unprompted, so the incremental rate should be materially higher there. If the campaign's incremental lift is concentrated in the recently lapsed, it is buying credit for intent that already existed.
• Control for the content release. If it launched everywhere simultaneously, the geo holdout is your only clean control; otherwise use difference-in-differences against a pre-period.

Until this reads out, treat 19,000 reactivations as an upper bound on impact, not a result.

Several mechanisms can make paid re-engagement and acquisition pay twice for the same user:

• Stale audience membership. A 900,000-user list built once will drift. Exclude anyone active in the last 7–14 days and shorten the membership window so the list reflects current lapse status. Rebuild on a schedule rather than once.
• Watch minimum list sizes and match rates after exclusions. Over-excluding can drop a list below eligibility, which silently kills delivery.
• Exclude the active-user list from the acquisition campaign where the platform allows it, or at minimum report new versus returning users so acquisition isn't funding re-engagement.
• Separate the campaigns by audience logic. Acquisition should run without lapsed-user lists; re-engagement should target the lapsed list only. Mixing objectives in one campaign blurs both the algorithm and the reporting.
• Reconcile acquisition conversions against MMP new-user installs. A persistent gap suggests acquisition is converting existing users.

Also question the event itself: 'first session within 7 days' is a weak signal for a streaming app and invites credit for incidental opens. 'Watched 10+ minutes' or 'resubscribed' would align the bid with value and reduce the pool of trivially attributed conversions.

Hold at $2,000 per day until the holdout reads out, because the campaign is budget-limited on a metric that may be largely organic. Tripling spend on an unverified incremental CPA is the most expensive way to learn the answer.

Decision rule to agree in advance:

• If incremental CPA is at or below the $4.00 target - for example, $56,000 of spend producing about 14,000 incremental reactivations - raise budget in 20–30% steps and re-measure marginal CPA after each step. Re-engagement audiences saturate quickly, so expect the marginal CPA to rise and plan to stop when it crosses target.
• If incremental reactivations are far lower - say 4,000, giving an incremental CPA near $14 - cut the budget sharply or re-target to the 90+ day lapsed segment, where incrementality is structurally higher.
• Rebalance toward the content wave. A meaningful share of the 18% organic lift did not need paid support. Shifting some budget into surfacing that content to lapsed users through owned channels is cheaper than paying $3.10 for opens that were coming anyway.

Finally, fix the event before scaling. Bidding to 'first session' caps the value the system can learn; moving to a deeper engagement or resubscription event will change both the CPA and the scale ceiling, and it is better to make that change before adding budget than after.