Every ad you show the same user is worth less than the one before it, and the decline is steeper than most publishers assume.
Measured across a mobile game running on a major mediation platform, eCPM falls between 3.9 and 5.1 percent with each consecutive impression a user sees. By the twelfth impression it sits 39 to 52.7 percent below the first.
The uncomfortable half is the cohort split. The low-value cohort, starting at 3 to 6 dollars, decays at 3.9 percent per impression. The high-value cohort, starting at 100 to 200 dollars, decays at 5.1 percent and ends 52.7 percent down.
Premium inventory burns faster. The publishers with the best audiences lose the most value to ad frequency, and they are usually the ones adding demand partners, because the headline eCPM still looks healthy while per-impression value collapses inside the session.
That is not a demand problem. It is an ordering problem, and it is the one lever on this list that costs nothing and requires no partner.
Why the usual advice does not move the number much
The standard answer to low eCPM is add partners and raise floors. Both are things you negotiate once, and both are weaker than they sound.
On floors, the two organisations that run the plumbing say so themselves. Prebid’s documentation: “Prebid does not recommend setting static floors. They are blunt tools and you’ll forget to update them.” Google’s AdMob help: “A higher eCPM floor value will likely cause a decrease in an ad unit’s fill rate”, and “Setting an eCPM floor value is likely to affect overall monetization”. Neither publishes a recommended number anywhere.
Even the vendors selling dynamic flooring claim 3 to 5 percent yield improvement, with 8 to 11 percent for high performers. Compare that with a 39 to 52.7 percent decay curve inside a single user’s session and the priority order writes itself.
Frequency, format and order, in that sequence
The same practitioner data shows eCPM never fully resets after a break. Even 48 hours later it recovers only to roughly 1.3 to 1.7 times the value of consecutive impressions, with iOS plateauing around 1.27 to 1.38 and Android continuing to climb to 1.48 to 1.67.
It also shows the format ranking inverting with depth. Around the sixth or seventh rewarded impression, interstitials begin to out-earn rewarded, because they start lower and decay on a flatter curve.
Both findings come from a single publisher’s apps, so treat them as a hypothesis to test rather than an industry constant. But the hypothesis is cheap to test and nobody is testing it, because the whole category is looking at partner counts.
The supply-path lever nobody counts as a yield lever
Pixalate’s Q1 2026 Mobile Seller Trust Index found that 16.2 percent of global authorised mobile app inventory is resold through arbitrage, against 20.8 percent on web and 4.1 percent on CTV. By country, roughly 48 percent of ranked SSPs in Mexico, 47 percent in Australia and 46 percent in Japan operate primarily as resellers.
That is inventory you authorised, being intermediated by a party taking margin out of your eCPM. It is an ads.txt hygiene question, not a fraud question, and it is entirely within your control.
Five things worth doing this month
Where the exchange relationship changes the ceiling
Most of the levers above are things a publisher does alone. The one that is not is demand density: how many genuine buyers see each impression, and how many intermediaries sit between them and you.
For example, at AdSpin the exchange runs a proprietary bidder on OpenRTB with a large base of owned and operated inventory alongside direct publisher supply, which keeps the path short and the declared signal accurate. Neither of those raises eCPM by itself. What they do is remove the two things that quietly cap it: hops that take margin, and bid requests that under-describe the impression.
If your eCPM chart looks stable while your ARPDAU does not, that is worth diagnosing against a direct connection at adspin.io.
