Auto-CPM (AdMob's "optimized" CPM, MAX's auto-CPM, LevelPlay's automatic optimization) is on by default and works great for the median publisher. It also leaks revenue for a real subset of publishers — usually the ones with non-typical traffic shape. This article is for that subset.
How auto-CPM actually works
At a high level, the platform watches your realized eCPM and clearing prices, then adjusts the effective floor (and sometimes the network ordering) to maximize a blended objective — usually expected revenue per impression. The math is well-tuned for the median case where:
- Traffic is steady week-over-week (no major UA shifts)
- Country mix is stable
- Demand shape is "normal" — bidders behave roughly as expected
The four cases where auto-CPM hurts you
Case 1: You just had a UA shift
Auto-CPM averages historical data. If your country mix shifted from 70% US to 30% US last week, the algorithm is still pricing as if the average is somewhere between the old and new mix. That under-prices your remaining US impressions and over-prices the new LATAM ones. Manual floors recover faster.
Case 2: You have a steady direct-deal layer
Direct deals with hard floors don't fit the auto-CPM model — the algorithm sees the direct line as just another bidder with a flat clearing price and doesn't reason about why it's there. Result: it can under-prioritize your direct line and send impressions to lower-paying programmatic.
Case 3: You have a tiny, niche audience
Auto-CPM needs volume to converge. If you're a niche app with 5,000 DAU concentrated in one country, the algorithm doesn't have enough signal and tends to over-floor (killing fill) or under-floor (killing eCPM) for weeks. Manual floors set near the 80th percentile of your last 30 days' clearing price work better at low volume.
Case 4: You have unusually high eCPM in one country
If JP eCPM on your app is 4x the global average (common in puzzle, casino, JP-targeted IP), auto-CPM tends to spread the floor too aggressively and ends up capping JP fill. Manual country-level floors recover this.
How to detect that auto-CPM is hurting you
Run this test: take one ad unit in one country, disable auto-CPM, set a manual floor at the 80th percentile of the last 30 days' realized eCPM for that combination. Wait 14 days. Compare blended eCPM and fill.
- If both go up: auto-CPM was hurting you on this combination. Roll out to other high-value countries.
- If eCPM goes up but fill drops: your manual floor is too aggressive. Drop the floor to the 70th percentile and re-test.
- If both go down: auto-CPM was helping. Re-enable it and look for problems elsewhere.
The pragmatic policy
Leave auto-CPM on by default. Manually override floors in two specific situations:
- High-value country / format combos where you have enough volume to set a sane manual floor (typically your top 1–3 ad units in your top 1–2 countries).
- For 30–60 days after any major UA shift, until the auto-CPM model has re-converged on your new traffic shape.
Mediation One's daily diagnosis will flag both situations automatically — if your country mix shifts more than 15% week-over-week, or if a specific country/format combination is being under-priced relative to its 80th percentile clearing rate, you'll see it in the next morning's report.