Floor prices (minimum eCPMs) are the highest-leverage and most-misused lever in ad mediation. Set them right and you protect your inventory's value without losing fill. Set them wrong and you either leave money on the table or quietly suppress your own fill rate. Here is a practical way to set floors, and the two mistakes that cost publishers the most.
What a floor actually does
A floor tells the auction: do not sell this impression below X. Any bid under the floor is discarded. That means a floor has two opposite effects at once — it raises the average price of impressions that do sell, and it reduces the number that sell at all (fill). Your job is to find the point where the price gain outweighs the fill loss on net revenue. Net revenue, not eCPM, is the only number that matters; a higher eCPM with lower fill can easily mean less money.
How to set them (a method that works)
Floors are not one global number. Demand differs by country, format, and platform, so a single floor is wrong everywhere at once. A workable approach:
- Segment first. Set floors per country (at least tier-1 vs the rest) and per format. A US rewarded floor and an India banner floor have nothing to do with each other.
- Anchor to your own data. For each segment, look at the historical eCPM distribution of filled impressions. A reasonable starting floor sits a little below the median realized eCPM for that segment — high enough to refuse genuinely cheap demand, low enough that most real demand still clears.
- Move in small steps and watch net. Adjust by 10–20%, wait for meaningful volume (days, not hours), and compare net revenue and fill against the prior period. Keep the change only if net went up.
- Prefer bidding where you can. Real-time bidding sets the price per impression automatically, which reduces how much manual floor-tuning you need. Reserve aggressive manual floors for waterfall lines.
Mistake 1: setting floors and never revisiting them
The most common and most expensive mistake. Demand prices drift — seasonally, and as networks change. A floor that was perfect in Q4 (high demand) becomes too high in Q1 (soft demand): the market fell beneath your floor, fill collapses, and net revenue drops even though you changed nothing. A stale floor is indistinguishable from "the market got worse" unless you check. Floors are not set-and-forget; they need a periodic review against current realized eCPM.
Mistake 2: chasing eCPM instead of net revenue
It is tempting to raise floors because the eCPM number goes up and feels like progress. But if fill falls faster than price rises, you earned less total money with a prettier eCPM. This is the trap dashboards encourage, because eCPM is shown big and fill is shown small. Always evaluate a floor change on impressions × eCPM = revenue, per segment. If you can't see net revenue move, the eCPM change told you nothing.
A quick sanity check
After any floor change, ask three questions per segment: did fill drop more than a few points? did net revenue rise or fall? is the realized eCPM of filled impressions now bunched right at the floor (a sign the floor is binding and probably too high)? If fill cratered and revenue fell, revert. If revenue rose with only a small fill cost, keep it.
Tracking realized eCPM distributions per segment and catching floors that have gone stale is exactly the kind of slow, per-segment bookkeeping Mediation One does for you: upload your CSV and it flags segments where a floor is suppressing fill or lagging the market. The free audit is one CSV upload — no SDK, no signup, nothing stored.