The timeout is the single most consequential number in a header bidding configuration, and it is almost always set by copying someone else.
Understand what it actually controls: how long you are willing to make the user wait, in exchange for a chance that a slow bidder turns out to be the highest one.
Both directions cost you money
Too short and you cut off bidders mid-response. Their bids are discarded, so you lose competition and your average clearing price falls. Bidders on slower infrastructure, or geographically distant from your users, get systematically excluded.
Too long and every user waits. Ads render below the viewport after the user has scrolled past, viewability drops, and impressions that never rendered at all still show as attempts. Worse, ad requests contend for bandwidth with your content, so your LCP degrades — and a slower site loses users, which reduces the number of impressions you have to sell in the first place.
Measuring it on your own inventory
- 1 Pull the bid response time distribution per bidder from your wrapper analytics. You want percentiles, not averages — the 50th tells you almost nothing useful here, the 90th tells you who you are cutting off.
- 2 Find where the curve flattens. Typically a large majority of bids arrive well inside a second and the tail after that is thin. The flattening point is your candidate value.
- 3 Cross-check what the tail is worth. If bids arriving between 1,500 and 2,500 ms are also your highest bids, the extra second may pay for itself. Often they are not, and you are paying a full second for scraps.
- 4 A/B test it properly. Split traffic, run for at least a week to cover the weekly cycle, and compare revenue per session — not CPM. CPM will happily rise while total revenue falls, because you sold fewer impressions at a better price.
- 5 Watch LCP and TBT in field data throughout, not just in the lab. Lab tests do not have real users on real networks.
Why one fixed number is the wrong shape
A user on office fibre and a user on rural 3G do not have the same optimal timeout. For the first, 1,500 ms is more patience than the auction needs. For the second, 1,500 ms may not be enough to get any bids at all, and you have made the page slower and earned nothing.
An adaptive timeout reads the connection and adjusts. In the ssm.codes v7 engine the shared deadline is shortened by up to 20% when the Network Information API reports a fast connection, extended by 20% on 3G, and bidding is skipped entirely on 2G or on devices reporting 1 GB of memory or less. On the very worst devices the correct auction is no auction: the ad server is called directly, the page stays usable, and you keep a user who would otherwise have left.
Per-placement timeouts
Below-the-fold placements are not on the critical path. If a unit is lazy-loaded and only requested as the user approaches it, it can afford a longer timeout than your header banner, because nothing is waiting on it. Splitting your timeout by position is usually a bigger win than tuning one global value.
Related questions
Is 1000ms too short for header bidding?
1,000 ms is on the aggressive side but defensible, particularly on mobile. You will lose some bids from slower bidders and your clearing price may dip slightly. Whether that trade is worth it depends on whether those late bids are actually high — measure before assuming. Many mobile-heavy publishers run 1,000 ms deliberately.
Does a longer timeout always mean higher CPM?
It usually raises CPM and can still lower total revenue. A longer wait means more impressions never render or never become viewable, so you sell fewer of them at a higher price. Always judge timeout changes on revenue per session or per pageview, never on CPM alone.
Should mobile and desktop have different timeouts?
Yes. Mobile users are on slower, higher-latency connections and are more likely to abandon, so the cost side of the trade-off is higher, while responses also arrive later. Reasonable setups run mobile somewhat shorter than desktop, or use an adaptive timeout that reads the actual connection instead of guessing from device type.
What happens to bids that arrive after the timeout?
They are discarded for that auction. The ad server request has already fired with whatever targeting existed at the deadline, so a late bid cannot win. Some wrappers cache late bids for a subsequent refresh; none can retroactively change a request that has already gone out.
Try it on your own site
Free up to 1 million ad impressions a month. You keep your Ad Manager account and your SSP contracts.
Start free