An eCPM floor sets the minimum price per 1,000 impressions accepted for a placement. Bids below it are rejected, leaving the impression unfilled; bids at or above it compete normally. In auction terms, it is your reserve price.
A zero or very low floor is not a neutral setting. It accepts whatever the market offers, including prices below what competition might produce. The real question is how to set floors correctly for each placement, format, and geography.
Why Floors Matter in RTB Auctions
In a first-price or second-price auction without a reserve, a demand partner can win by bidding just above the next competitor. Where competition is weak, the clearing price may fall to the highest amount that bidder will pay. A floor blocks that outcome by defining the minimum acceptable result.
The signal effect matters too. Networks that learn your inventory always accepts low bids may shade prices downward over time. A credible floor tells demand partners to bid at or above market rate. This is basic auction design, not punishment: publishers receive fair value, while buyers get a clear price signal instead of racing downward.
A floor set too high can reduce fill rate and net revenue. The target is the point where eCPM gains from rejected sub-floor bids exceed revenue lost from unfilled impressions. Low floors can preserve fill when that matters most, but this balance varies by placement and changes with demand.
Setting Floors Too Low: The Cost of Inaction
A $0.10 floor on rewarded video in South Korea, where advertisers routinely bid $3 to $8 for premium mobile inventory, captures only a small part of available value. The floor is barely functioning as a reserve. Fill reaches every impression, but eCPM remains a fraction of what a floor could produce through upward bidding pressure.
The cost of pricing below market is hidden in daily reports. Fill rate stays high, revenue looks stable, and no error appears. The missing comparison is what the placement could have earned at the right floor. That counterfactual explains why publishers with stable low floors often see unexpected gains after optimization.
A utility app publisher working with Sova Apps, part of our early-access cohort, raised rewarded video floors on South Korean traffic from $0.50 to $2.00 after seeing nearly every bid exceed $0.50. Fill rate fell slightly, but eCPM more than doubled for that segment, producing a significant net revenue improvement. The bid cluster was already visible in reporting; the floor simply had not followed it.
Setting Floors Too High: The Fill Rate Risk
A floor well above market rate leaves most bid requests unfilled. That affects more than revenue from those impressions: it also lowers total ad event volume and can affect engagement and user experience when app mechanics depend on ad availability, such as rewarded video for in-game currency.
This failure is common when a high-value geo's floor is applied to all traffic. A floor suitable for Japanese rewarded video rejects most bids from markets with median CPMs at a third of that level. High-value traffic fills while lower-CPM traffic nearly stops, which can look like a mediation configuration issue instead of a floor calibration issue.
Use geo-aware floors instead. Adrop's floor manager supports different floors by country or region within one placement configuration. A placement can use a $4.00 floor for Japanese traffic and $0.80 for Indonesian traffic, with each geo participating independently in its auction.
How to Calibrate Floors Methodically
Begin with the bid distribution for each placement. What range arrives, and what are the 10th, 50th, and 90th percentile bids? If the current floor sits below the 10th percentile, it does little. Moving it to the 25th or 30th percentile rejects some bids while preserving most fill, and those rejected bids are the lowest-value ones.
Adrop's dashboard displays bid distributions by placement and geo. Find the cluster: if 80% of bids sit between $1.20 and $4.00 while the floor is $0.30, the floor is far below actual demand. A $1.00 or $1.10 floor would reject few real bids while signaling market-rate pricing.
After raising a floor, wait at least five to seven days to assess it. Daily CPM variance can be significant, so short windows mislead. Compare weekly averages rather than day-over-day movement.
A/B Testing Your Floors
For placements with enough volume for statistically meaningful data, a floor A/B test is more reliable than a sequential before-and-after comparison. Seasonality, weekday effects, and campaign starts or ends can distort sequential tests. In a concurrent test, both groups experience those variables.
Adrop's floor A/B testing routes a configurable share of traffic through an experimental floor while the rest uses the control. After a week, compare eCPR, or revenue per request. If the higher-floor group wins, send the experimental floor to 100% of traffic. If not, the floor was too aggressive for that placement.
Floor optimization is ongoing, not a one-time change. Demand shifts with advertiser budget cycles, seasonal campaigns, and competition within an app category. Floors calibrated six months ago may now be below or above market. For most publishers, reviewing bid distributions against current floors quarterly is a reasonable maintenance cadence.
Format-Specific Floor Guidance
Floor ranges differ sharply by format. Rewarded video generally commands the highest CPMs across app categories and geos because opted-in viewers are engaged, high-intent users advertisers value. One floor for rewarded video and banners miscalibrates both: rewarded video is likely too low, while banners are likely too high.
Interstitial floors usually fall between banner and rewarded video in most markets. Banners still need floors, but at lower levels, and their fill rate reacts more strongly because banner CPM distributions are narrower. Many publishers use no banner floor or a very low one, concentrating yield work on higher-value formats.