Why latency budgets are eating your yield
Every 100ms of auction delay costs you measurably. Here is the data, pulled from real publisher sessions.
Latency is the quietest yield killer in the stack, because it never shows up as a line item. Nobody sends you an invoice for the revenue a slow auction cost you.
The 100ms rule
Across the publisher sessions I sampled this year, every additional 100ms of auction time cost roughly one to two percent of viewable revenue. It compounds, and it compounds worst on exactly the mobile sessions you can least afford to lose.
Where the time actually goes
It is rarely the auction itself. The expensive milliseconds hide in:
- Wrapper initialisation that blocks the first auction.
- The consent round-trip, especially when the CMP loads late.
- A long tail of slow bidders you enabled once and never reviewed.
Setting a real budget
A latency budget is only useful if something enforces it. Pick a timeout, measure the bid-response distribution, and cut any bidder that consistently misses it without contributing revenue.
A bidder that wins two percent of auctions but slows every auction is not a revenue source. It is a tax.
The discipline
Review the bidder list every quarter. Latency budgets are not a one-time setup; they rot. The publishers who hold yield through a soft market are almost always the ones who treat auction speed as an ongoing operational metric, not a launch checklist item.