| raw | raw/mastering-meta-ads-2024-a-10-step-blueprint-for-app-install.md |
|---|---|
| url |
TL;DR
Admiral Media’s Juan David lays out a 10-step creative-testing framework for Meta app-install campaigns: form a hypothesis, group creatives into concepts, run separate ad groups so each creative gets its own spend, and judge winners on cost-per-trial and thumb-stop rate. The hard part is SKAN’s 88-install-per-campaign threshold on iOS, which makes testing expensive — Aggregated Event Measurement (AEM) is the budget workaround. This source belongs to the app-masters-youtube batch.
Biggest lessons
- Structure creatives as concepts, not random ads. Run ~3 concepts × 5 variations (15 ads). Keep variants within a concept similar so you isolate the winning factor; make concepts very different from each other.
- iOS SKAN forces an 88-daily-install threshold per campaign to unlock postbacks (trial/subscription data). At a $7.20 US CPI that’s ~$634/day per adset — a real budget gate. Android has no such restriction.
- Use AEM (Meta) / ADC (TikTok) when budget is tight. One client ran a 7-day, $1K test optimizing to installs, grouped ads by concept, then promoted winners into the BAU campaign — yielding 12% lower CPI, 44% lower cost-per-trial, 133% more installs.
- One creative per ad group, same targeting. Isolating creatives stops the algorithm from dumping all spend on one ad and starving the rest of data.
- Optimize for the event you actually want. Optimizing for installs gets cheap installs but weak ROAS; optimize toward conversion value if you want revenue.
- Judge creatives on thumb-stop rate and CTR (~1% target), not just CPI. Hooks that work: “this is why your X doesn’t work,” “this is going to blow your mind” — play with audience psychology.
Why it matters
- Anchors the wiki’s paid-ads-scaling thinking for Meta specifically — concept-based creative testing plus the SKAN math every iOS advertiser hits.
- Reinforces the cross-source theme that creative is the lever, not targeting, and that you should never stop testing.