What is an advertising waterfall?

How a waterfall works

The publisher ranks their demand sources, historically by average historical eCPM. When an impression becomes available, the top source is offered it first. If it declines, the second is asked, then the third, and so on until one fills or the sequence is exhausted.

It was the standard mobile mediation model for years, and it is a reasonable design given the constraint it was built under: not every network could be asked simultaneously, so they had to be asked in some order.

Why it loses money

The ordering is based on what a network paid *on average*, historically. It has no idea what any network would pay for *this* impression right now.

So a network ranked third that would have paid £8 for a particular impression never gets asked, because the network ranked first accepted it at £3. The publisher systematically leaves the difference on the table, and cannot see it happening — the waterfall reports a fill at £3 and nothing about the £8 that was never bid.

The second cost is latency. Each unsuccessful call takes time, and a deep waterfall can add hundreds of milliseconds before an ad appears, which itself reduces fill and degrades the user experience.

What replaced it

In-app bidding — the mobile equivalent of header bidding — asks every demand source simultaneously and takes the highest actual bid for that specific impression.

The gain is both revenue and speed: no sequential calls, and the price reflects real-time competition rather than historical averages. Most large publishers have migrated, often keeping a short waterfall as a fallback beneath the auction for sources that do not support bidding.

The general lesson transfers well beyond mobile: any allocation based on historical averages leaves money on the table wherever real-time competition is possible.