AdMob mediation waterfall optimisation: rank ad sources by what they actually deliver
Most mediation waterfalls are ordered by what networks claimed when they were added, then never revisited. The fix isn't better auction settings — it's reading the mediation report against observed performance and reordering on evidence. This is that workflow.
Claimed eCPM versus observed eCPM
The gap between what an ad source says it pays and what it actually delivers is where waterfall money leaks. A network with an attractive headline rate that fills 4% of requests contributes less than a duller one filling 40%. Ordering by claimed rates therefore optimises the wrong number — and once a waterfall is set, almost nobody re-derives it.
The first useful question is simply: "rank the ad sources in each mediation group by observed eCPM over the last 30 days." That single ranking usually disagrees with the configured order somewhere.
The four questions worth running
- 1"Rank ad sources by observed eCPM per mediation group." The evidence-based ordering, versus what's configured.
- 2"Where are the fill-rate gaps?" A source high in the waterfall that rarely fills is a latency and revenue cost on every request that passes through it.
- 3"Split earnings and eCPM by iOS versus Android." They routinely behave differently enough to deserve different waterfalls, and one platform is often subsidising the other in a blended view.
- 4"Rank my apps by earnings and by eCPM, and flag the weakest with meaningful impressions." Portfolio-level attention allocation, which no per-app dashboard puts in front of you.
By ad format, not just by app
Format is the segmentation most publishers skip. Rewarded, interstitial and banner have very different economics, and an app that looks mediocre in aggregate is often one strong format carrying two weak ones. Splitting by format before concluding anything about an app avoids retiring a format that was actually working.
What to do with the answer
Reordering a waterfall, adjusting a floor or removing a source stays a decision you make in AdMob — the integration is read-only, so the agent produces the ranking and the reasoning, not the change. That division is deliberate: the analysis is the part that's tedious and skippable, and the configuration change is the part worth a human thinking about it.
Make it a rhythm
Waterfall economics drift as networks' demand changes, so a one-off audit decays. Run the source ranking monthly on a schedule, alert on fill-rate collapse, and keep the portfolio view on a self-refreshing dashboard so a bad month is visible before it's a bad quarter.
Frequently asked questions
- How do I optimise an AdMob mediation waterfall?
- Rank ad sources by observed eCPM per mediation group over a recent window rather than by their claimed rates, check fill-rate gaps for sources sitting high in the waterfall, and split by iOS versus Android since they often deserve different orderings.
- Why is claimed eCPM misleading?
- Because contribution is eCPM times fill. A source with a high headline rate that fills a small fraction of requests delivers less than a lower-rate source that fills consistently — and it costs latency on every request that passes through it.
- Should I use one waterfall for iOS and Android?
- Usually not. They routinely differ enough that a blended view hides one platform subsidising the other. Splitting the report by platform is the check worth running before deciding.
- Can the agent change my mediation groups?
- No — the AdMob integration is read-only. It produces the ranking and the reasoning; reordering sources or adjusting floors stays your decision in AdMob.
- How often should I revisit the waterfall?
- Monthly is a reasonable rhythm, since network demand drifts. A scheduled ranking plus an alert on fill-rate collapse catches decay before it compounds into a bad quarter.