A floor discards bids below a threshold. Set too low it earns nothing; set too high it leaves inventory unsold. The only reliable method is testing on a portion of traffic and judging by total revenue rather than rate.
What a floor does
Instructs the ad server to reject bids below a stated price for a defined slice of inventory.
Two effects, always both. The average price of what sells rises, because the cheap end was excluded. And the proportion that sells falls, because some impressions now attract no acceptable bid.
Whether that is a gain depends entirely on the shape of the bids you were receiving, which differs by placement, by geography, by device and by time.
Which is why a floor copied from another publisher is a guess about somebody else's bid distribution.
Why the obvious reasoning fails
The intuition is that a floor makes buyers pay more. Frequently it does not.
In a second-price arrangement the winner pays slightly above the next bid, so a floor can raise what the winner pays. In a first-price arrangement, which is now common, the winner pays their bid, and a floor mostly determines whether they participate rather than what they pay.
Which means the mechanism differs by how the auction settles, and a floor that worked historically may behave differently now.
The reliable statement is narrower: a floor removes the bottom of the distribution. Whether removing it helps is an empirical question about your inventory.
When a floor is justified
- You are receiving many very low bids that sell inventory for almost nothing.
- A placement is genuinely premium and priced as though it were not.
- You have direct deals that programmatic demand should not undercut.
- A specific buyer or channel is consistently paying below what others pay for the same slot.
The third is the strongest case and the least contentious. Protecting a directly sold rate from being undercut by the open market is what floors are genuinely good at.
When it is not
Where publishers apply one and lose.
Low traffic, where reduced fill removes more revenue than the rate improvement adds and the numbers are too small to tell.
Applied uniformly across all inventory, when the right floor for a top placement and a footer slot are entirely different.
Applied to a geography with structurally lower rates, which simply stops that audience monetising at all.
Set from a benchmark rather than from your own data, which is the most common version and the least defensible.
Testing properly
- Establish a baseline over a period long enough to cover normal variation.
- Apply the floor to a portion of traffic, not all of it.
- Hold it long enough for buyers to adjust, since immediate results overstate the effect.
- Compare total revenue, not rate and not fill individually.
- Segment the result by device and geography, since the answer differs.
- Try a second level before concluding, since the first attempt is rarely optimal.
The third step is where most tests go wrong. Buyers adapt over days, and a floor judged after forty-eight hours is being judged on the disruption rather than the settled state.
Differentiating by inventory
The refinement worth making once the basic test is done.
Different floors for different placements, since a slot in the first screen and one below the article are worth different amounts.
Different floors by geography, since applying a rate suited to one market to all of them will silently stop others selling.
Different floors by device, since mobile and desktop rates differ structurally.
That granularity is where floors move from a blunt instrument to a useful one, and it requires the ad server configuration to support it, which most do.
Knowing when to remove one
The decision nobody revisits.
Demand changes seasonally and structurally. A floor set during a strong quarter may be excluding most bids in a weak one, and nothing announces that.
Watch unfilled inventory. A rising share of impressions attracting no bid, with no other explanation, is a floor that has become too high for current conditions.
Reviewing floors on the same cadence as the rest of your setup, rather than setting and forgetting, is the practice that separates a working configuration from a historical one.
The measurement discipline this depends on is set out in CPM, RPM and eCPM without the confusion.
Frequently asked questions
What does a floor do?
Rejects bids below a stated price. The average price of what sells rises and the proportion that sells falls, always both.
Does a floor make buyers pay more?
It depends on how the auction settles. Where the winner pays their own bid, a floor mostly determines whether they participate rather than what they pay.
When is a floor clearly justified?
Protecting a directly sold rate from being undercut by open market demand. That is what floors are genuinely good at.
When does it backfire?
On low traffic where reduced fill outweighs the rate gain, applied uniformly across different placements, or applied to a geography with structurally lower rates.
How should it be tested?
On a portion of traffic against a baseline, held long enough for buyers to adjust, judged on total revenue rather than rate or fill alone.
When should a floor be removed?
When unfilled inventory rises with no other explanation. Demand shifts seasonally, and a floor set in a strong quarter can exclude most bids in a weak one.
West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.
Set a floor once and never revisited it?
We check what share is going unfilled, which is how a floor that used to work announces that it stopped.
Start a Conversation