Fill rate is the share of ad requests that come back with an ad. It is easy to measure, easy to chart and easy to improve. Those three qualities are exactly why it deserves some suspicion.
How fill rate gets raised
If you want more requests to be filled, the quickest route is to lower the price at which you will sell. Drop the floor, add buyers who bid low, accept anything that arrives. The number climbs almost at once.
Nothing in the metric notices what happened to the price. A request filled at a very low bid counts the same as a request filled at a very good one. A team can celebrate a record fill rate in the same month revenue falls.
This is how a healthy looking dashboard hides a weak business. Every request has an answer, every slot has an ad, and the monthly payment is smaller than it was.
What pays the bills
Money arrives per impression, but a publisher lives per visit. A reader opens a page, scrolls, perhaps opens another, and over that visit a handful of ads are shown. What matters is what that whole visit earned.
Revenue per session and revenue per page capture this. They fall when prices fall and rise when prices rise, and they move with the number of ads shown. A metric that includes both price and volume is much harder to flatter than one that counts only whether a slot was filled.
When an empty slot is the right answer
Suppose a slot is offered at a modest floor and the best bid is a fraction of that. Selling it would raise fill rate and add a sliver of revenue. It would also tell every buyer that this slot clears at that price, and bids tend to settle where past auctions have cleared.
Leaving the slot empty, or serving something of your own, keeps the price signal intact. Next time a better buyer may be there. Some unsold impressions are the market working properly, because they mark the line below which the inventory is not for sale.
A sold impression is only good if the price is one you would repeat. If you would not accept it a hundred times in a row, you should not accept it once.
More slots, lower prices
The same logic applies to layout. Adding an extra ad unit to a page increases the number of impressions for sale. It does not increase the number of advertisers who want to reach that reader, nor the amount they have to spend.
With the same budgets chasing more slots, each slot tends to sell for less. The page gets slower and more crowded, readers scroll away sooner, and fewer of the ads are seen at all. Buyers notice that a placement is rarely viewed and bid accordingly. The reader pays in attention, and the publisher often pays in revenue.
The test is easy to run. Add a unit to a few pages and leave others alone, then compare what each group earned per visit and how long readers stayed. If the busy pages earn the same or less, the extra slot added only clutter.
Count what the page earned, and forget what fraction of its slots were full.
What to watch instead
No single number tells the whole story, so it helps to keep a small set side by side:
- Revenue per session, which shows what a visit is worth.
- Revenue per page view, which shows whether layout changes helped or hurt.
- Average winning bid by placement, which shows whether you are selling cheaply.
- Share of impressions that were actually viewable, which shows whether buyers have reason to pay more.
- Pages per session and return visits, which show whether the ads are costing you readers.
Fill rate can stay on the list as a diagnostic. If it drops suddenly, something may be broken. It should not be a target.
Choosing the target
A metric that people are rewarded for will be improved, by whatever means are closest to hand. If the target is fill, the closest means is a lower price. If the target is revenue per session, the closest means is better demand, better placement and calmer pages. That is where Yield aims its own work.