A price floor is the lowest bid a seller will accept for an impression. Most people describe it as a fence that keeps cheap bids out. It does that, but its more useful job is to tell buyers something.
What a floor says
In an auction, every bidder is trying to guess what the impression is worth to the seller. They cannot see the seller's costs or the other bids. They can see the floor.
So the floor is read as a statement. A high one says this inventory is valuable and the seller knows it. A low one says anything will do. Buyers adjust their behavior to match, usually without being aware of it.
A floor that is too low
Take a floor so low that nearly every bid clears it. It does nothing. The auction runs exactly as if there were no floor, and the winner pays whatever the second highest bidder offered.
That is fine when many buyers want the impression, because competition pushes the price up on its own. It is a problem when only one buyer shows up. In a second price auction a lone bidder pays the floor, however much they were willing to pay. A buyer who would have paid a good deal more pays almost nothing, and the seller never learns the difference.
A floor that is too high
The opposite mistake is easier to see. Set the floor above what any buyer will pay and the impression goes unsold. The page loads, the slot sits empty or falls back to something worthless, and the reader sees nothing of value either.
A high floor also teaches buyers. If they keep losing at a certain price, they either stop bidding on that inventory or learn to bid just over the floor and no higher. Neither helps the seller.
There is a quieter cost too. Buyers keep records of where their bids win. Inventory that rarely clears gets less attention from the people who tune bids, and attention is hard to win back once it has gone elsewhere.
Why a floor goes stale
The right minimum price is a statement about demand, and demand moves. Consider a few ways it shifts:
- By hour. A buyer running a campaign for commuters wants the morning and has little interest at midnight.
- By season. Budgets swell before big shopping periods and shrink afterward, and the same impression is worth different amounts in each.
- By placement. A slot at the top of an article, seen by nearly every reader, is a different product from one at the foot of the page.
- By audience. A reader who is about to buy something is worth more to the right advertiser than a reader who is browsing.
A floor typed in once reflects the market on the day somebody typed it. A month later it is a guess about a market that has moved on.
Holiday weeks make the point. Demand can double in the days before a big shopping event and collapse the day after. A floor that was sensible on Monday can leave real money on the table by Friday, and nobody notices, because an auction that sells for too little still looks like a sale.
A floor set once is a price from a day that has already gone.
Letting floors follow demand
The better approach starts from what buyers have actually paid. If winning bids on a placement cluster in a certain range at a certain hour, the floor can sit just beneath the range. It filters out the bids that were never going to be competitive, and it protects the sale that would otherwise clear for far less than the market would bear.
The floor then needs to move. Raise it when demand is strong and bids are dense. Lower it when the room is thin, so the impression still sells. Some impressions will go unsold at a good floor, and that is a fair price to pay for not selling everything cheaply. A floor adjusted this way is a running estimate of what the impression is worth right now.
A simple test
Look at any floor you run and ask when it was last changed, and by what evidence. If the answer is a long time ago and a hunch, the floor is a habit. Habits are cheap to keep and expensive to carry.
A second question is whether anyone could tell if the floor were wrong. If you only learn about a bad floor when revenue visibly drops, it has been wrong for a long time already. Good floors come with a record of bids that cleared, bids that fell short and impressions that went unsold.
Yield sets floors from observed bids, by placement and time, and keeps them moving. The method is simple. The discipline is in never treating the number as finished.