Hybrid waterfalls: what bidding still leaves on the table

Every mediation platform has spent five years telling publishers that the waterfall is a legacy artefact and full bidding is the destination. Most large game publishers still run a hybrid: a handful of bidders competing in real time, plus a set of traditional fixed-price lines above and below them. Inertia, people say. Missing bidder adapters. The better explanation is that the hybrid survives because bidders, left to themselves, do not pay what the impression is worth, and the fixed lines are the publisher's only instrument for making them.

That matters to a UA desk and not only to a monetisation one, for a reason that keeps getting skipped: ad revenue per user is half of the LTV that sets your bids on ad-supported titles. Leave eCPM on the table and your UA bids sit lower than they could, which means losing auctions to a competitor whose stack does not.

Why a bidder pays less than the impression is worth

In-app bidding on the major mediation platforms runs as a unified auction: each bidder returns a price for the impression, highest wins. In a pure first-price auction a rational bidder does not bid its true value. It bids somewhat below, and how far below depends on how much competition it expects. That is bid shading. Web programmatic has worked this way since exchanges moved to first price.

A bidder on a game's rewarded video placement estimates two things: what the impression is worth to its advertisers, and what the next-highest bid will come in at. Weak competition means it shades harder. The publisher sees only the winning price and has no direct way to learn the bidder's real value. Over enough time, a stack of bidders with no floors settles into a pattern where the bidders learn each other's behaviour and the clearing price sits below the level a competitive market would produce.

A fixed-price waterfall line changes that calculation. A traditional network line at a set eCPM above the bidders works as a hard floor: no bidder clears it, that line takes the impression. Bidders that shade too aggressively lose impressions they would otherwise have won, and their models learn to bid closer to value on that placement. Publishers running hybrid stacks are using the waterfall as price discipline on the auction, whether or not they would describe it that way.

The costs that the tidy diagram leaves out

The hybrid is not free, and the trade-offs deserve stating plainly.

Latency is the obvious one. Each fixed line in the sequence adds request time, and a placement that takes two seconds to fill loses a share of impressions to users who tap away. Lost fill has a cost, and on interstitials that cost can run past the eCPM you defended.

Reporting is the less obvious one. A hybrid stack produces two kinds of revenue data: auction prices from the bidders, which are real per-impression clearing prices, and network-reported eCPMs from the fixed lines, which are averages reconciled after the fact. Mix them into one ARPDAU figure, feed that into a UA LTV model, and you get a number whose error bars nobody has estimated. When the monetisation team moves a waterfall line and the UA team's LTV curve shifts a week later, UA usually reads it as a cohort change.

The third cost is strategic. Fixed lines depend on networks that still sell on a non-bidding basis, and that set is shrinking. Publishers using the waterfall as discipline lean on an instrument that may not exist in its current form for much longer. Demand for rewarded and interstitial inventory is not the constraint; the mediation platforms themselves report that most large publishers still run some fixed lines, which tells you plenty about how much publishers want that discipline. The constraint is whether publishers keep any way to extract a competitive price for that inventory once the last fixed lines go.

A decision rule for what stays fixed

Skip the blanket move to bidding and treat each placement as its own question. Keep a fixed line only where it does one of these jobs:

• It sets a floor the bidders demonstrably respond to; test that by pulling it for a week on a matched placement and watching whether bidder clearing prices drift down.
• It captures a demand source that does not bid, and that demand is materially better than the next bidder at that position.

One more case survives: a bottom line that backfills where the alternative is no fill.

Everything else goes to bidding, with an explicit floor set in the mediation platform instead of a fixed line pretending to be one. Both AppLovin's MAX and Unity's LevelPlay expose placement-level floors, and using them strips out the latency of a network call while keeping the discipline.

As an illustrative example, picture a hybrid-casual publisher with a rewarded placement running five bidders and four fixed lines. Pull the two lowest fixed lines and revenue does not move, while latency drops. Pull the top fixed line and bidder clearing prices fall by a visible margin over ten days as the bidders shade into the space it left. The publisher puts the top line back, or swaps it for a mediation-level floor at the same price. The numbers are invented; the test is the point.

What the UA desk should ask for

The practical request from UA to monetisation is one figure: the share of ad revenue by placement that comes from real auction clearing prices rather than reconciled network averages. If more than a quarter of the revenue feeding your LTV model gets reconciled after the fact, you are setting bids on a number carrying a lag and an error you have never measured.

Ask as well for advance notice of any waterfall change on the top three placements by revenue. A floor moved on a Monday turns up as a cohort shift in your day-7 ARPDAU by the following Monday, and knowing in advance costs far less than a week spent diagnosing a phantom creative problem.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• How platform fee changes flow into bids
• Dayparting is dead, except where it is not