What a passback chain actually costs you
Each additional layer looks free because it only earns on inventory the previous one could not fill. The arithmetic says otherwise.
Passback chains are built out of individually sensible decisions.
Your primary demand does not fill everything. Someone offers to take the remainder at no fixed cost — they only make money when they fill, so the downside appears to be zero. You add them. A year later the same logic adds a third layer, then a fourth.
The downside is not zero. It is just distributed across three places nobody looks at together.
Cost one: the take rate compounds
Each layer keeps a percentage. Individually the numbers sound modest — fifteen here, twenty there. Chained, they multiply rather than add.
An impression that a direct buyer would have valued at a dollar, passing through three intermediaries each keeping twenty percent, arrives at fifty-one cents. The publisher sees fifty-one cents and concludes the impression was worth about fifty-one cents.
Cost two: latency destroys inventory
Every hop is a network round trip. On desktop fibre this is an inconvenience. On mid-range Android over a congested mobile network — a large share of traffic in most markets we operate in — it is the difference between an impression that renders and one that does not.
We regularly measure double-digit percentages of impressions lost in the fourth layer of a chain simply because the user scrolled past, navigated away or closed the tab first. Those impressions do not appear as a loss anywhere. They appear as slightly lower inventory, which looks like a traffic problem.
Cost three: your best demand never sees it
This is the expensive one, and the least visible.
Chains are ordered by who bid first, not by who would have paid most. Inventory that a category-relevant advertiser would have bought at a strong price gets consumed by whichever generic layer happened to be earlier in the sequence.
The buyer who valued that reader most never got a chance to bid, so their absence never shows up in the report. You cannot miss revenue you never saw quoted.
How to find out what yours costs
The audit is not complicated, but it does require measuring the thing itself rather than the reports each partner produces about their own performance.
- Instrument the chain. Log time-to-render by layer, not just fill by layer.
- Compute revenue per thousand sessions. Layer-level eCPM will look fine even while the total collapses.
- Test removal, not theory. Drop the last layer on a traffic slice for two weeks and compare against a holdout. In most cases revenue per session is flat or up, because the impressions that were being lost to latency come back.
- Reorder by value, not by history. The sequence was almost certainly determined by the order in which contracts were signed.
The typical result of collapsing a four-layer chain into a properly configured unified auction is a mid-teens improvement in revenue per session. Not because anyone was cheating — because four intermediaries each taking a fair margin on the same impression is not a business model, it is an accident.