There is no single correct mediation setup. There is, however, a small set of defaults that work for 80% of apps and a small set of experiments that almost always pay back the time spent running them. This is what we ship to new accounts and what we tell their growth team to test in week one. Each section below covers placement, cadence, floor strategy, and the mistakes we see most often.
Rewarded video: the highest-value format
Rewarded is the only format where the user explicitly opts in. eCPMs are 4–8x interstitial, completion rates sit above 90%, and user satisfaction stays neutral or positive when the reward is meaningful.
Default cadence: unlimited offers, no global cap. The cap belongs on the reward economy, not the ad. Run an experiment doubling the reward value for the first three plays of a session — most apps see 15–25% lift in opt-in rate with negligible inflation.
Rewarded: placement, floors, and the three mistakes we see most
Placement matters more than people think. The best-performing rewarded placements are at moments of natural friction — out of lives, premium content gate, double rewards. The worst placements are reward buttons floating on the home screen with no friction to absorb. Friction is what makes the value exchange feel fair.
Floor strategy: bid floors on rewarded should be aggressive in Tier 1 and Tier 2, conservative in Tier 3. Run a country × daypart matrix and recompute weekly. The single biggest revenue leak we audit is one global rewarded floor.
- Mistake 1: capping rewarded — every cap below 10/session leaves money on the table for engaged users
- Mistake 2: pre-loading more than one rewarded ad — memory pressure shows up as crashes within 48 hours
- Mistake 3: confusing 'reward granted' with 'video completed' — always wait for the SDK completion callback
Interstitial: the format most teams over-serve
Interstitial is where teams lose retention without realising. The temptation is to show one after every level, every screen transition, every loading state. Don't.
Default cadence: minimum 90 seconds between interstitials, hard cap at 6 per session, and never within 30 seconds of app open. Then run two experiments: one tightening the gap to 60 seconds, one loosening to 120. Pick the variant that maximises Day 7 retained ARPDAU, not Day 1 ARPDAU.
Interstitial: the floor strategy that actually works
Interstitial floors are where we see the most lazy thinking. Most teams pick one floor — often $3 — and ship it. The floor that maximises revenue is almost never the one that maximises eCPM, because aggressive floors trade fill for price. A $5 floor with 60% fill loses to a $3 floor with 95% fill every single time on rewarded-adjacent inventory.
The right shape is a step function: low floor in low-eCPM markets, higher floor in premium markets, capped at the 75th percentile of historical winning bids. Recompute monthly, not weekly — interstitial bid distributions are noisier than rewarded and a weekly recompute oscillates.
Native: the format worth the integration cost
Native units take longer to ship because they require design work. They also have the lowest user-experience cost of any format and the best long-term retention impact when placed thoughtfully.
Default placement: in-feed, every 8–12 organic items, with the ad styled to match the surrounding content but always carrying a visible 'Sponsored' label. Run an A/B test on the label position — top-left vs. corner badge — measuring both CTR and complaint rate.
Native: when to invest, when to skip
Native pays back in apps with feed-shaped content — social, news, e-commerce, content discovery. It rarely pays back in single-screen utilities or game-loop apps where there is no organic feed to slot into. If you cannot describe the feed in one sentence, do not ship native.
The hidden cost is creative QA. Native creatives ship in dozens of aspect ratios and brand voices; the bad ones cluster around adult, dating, and shock-CPI campaigns. Always enable category-level blocks on day one and review complaints weekly for the first month.
Banner and app-open: the appendix
Banner and app-open are the two formats most often misused, but they do have a place. Banners work in apps with long, idle, single-screen sessions — utilities, reference apps, calculators — where they earn small but steady revenue without disrupting flow. App-open ads earn well in apps users return to many times a day, where one impression on the cold start is a fair trade.
The test for both formats is the same: hold them in a 50/50 experiment for two weeks and measure Day 14 retained ARPDAU. If retained ARPDAU is flat or up, keep them. If it is down, no amount of eCPM lift will compensate.
First 14 days: the experiment calendar we ship
If you adopt only this section, you will outperform 80% of accounts. Each experiment is a single change, gated on a 10% holdout, evaluated on Day 7 retained ARPDAU.
- Day 1–3: ship the rewarded country × daypart floor matrix
- Day 4–7: test rewarded double-reward for the first three plays of a session
- Day 5–10: test interstitial cadence at 60s vs 90s vs 120s
- Day 8–12: enable native in your highest-traffic feed, every 10 items
- Day 11–14: trim the waterfall to top six bidders plus two backfill networks
What we do not recommend by default
- Banner ads in apps with sessions under 5 minutes — they hurt more than they earn
- App-open ads outside of utility apps — retention impact is severe
- Mid-session interstitials during gameplay — they correlate with churn even when ARPDAU looks fine
- Reward inflation — doubling rewards globally feels good for a week and is impossible to roll back
Every ad format has a right placement, a right cadence, and a right floor. Almost no one ships all three at once — the experiments above are how you find the combination that fits your app.
What to do this week
Pick the format that earns the most for you today and ship one experiment from the calendar above on it. One change, one holdout, two weeks. The next experiment will almost design itself once the first one reads out.
