Prebid docs - browse
Prebid/Settings/Prebid bidder timeout
SETTING

Prebid bidder timeout

The bidder timeout is how long Prebid waits for adapters before it sends the bids it has to the ad server.

Plain English

Prebid distinguishes two clocks. The auction timeout is inside Prebid: when it expires, the auction uses the responses in hand. The failsafe timeout is the publisher's own timer, outside Prebid, that calls the ad server if Prebid never calls back. Prebid's FAQ gives starting points of 1,000 ms or less for the auction and 3,000 ms or less for the failsafe. The setConfig page also shows an example of bidderTimeout: 3000. Neither figure is a rule for every site.

A response after the auction timeout does not get targeting. Prebid also notes that a busy page can run the timeout callback late, because setTimeout is not a precise clock, and that it closes the auction more firmly once the overrun passes 200 ms.

One slow adapter does not have to own the whole wait. setBidderConfig is Prebid's overlay of setConfig for the bidders you name. The examples on that page are custom arguments and ortb2, not a catalog of timeouts. When a bidtune test sets one bidder's timeout, the script puts bidderTimeout in that overlay for the named bidder. See `setBidderConfig`.

How it shows up in pbjs

Set the auction timeout before requestBids. The snippet is the example from the setConfig page. The FAQ gives a stricter starting point for the same field: 1,000 ms or less. The failsafe is not this field. It is a setTimeout in the page, shown in the basic example. requestBids can also take a timeout for that one call. The two clocks are drawn on the timeout reference.

pbjs.setConfig({ bidderTimeout: 3000 });

When it helps, when it hurts

Helps

A shorter auction renders sooner when the bids that actually win have already arrived. Giving a bidder who rarely answers less time spends less of the wait on silence.

Hurts

Shortening the auction past the arrival of winning bids throws those bids away. Lengthening it delays every impression, including the ones that were never going to get a late bid. A timeout copied from a faster audience, or from a partner's default, is not evidence. Fill can rise while revenue per session does not.

What bidtune tests here

Autopilot can test another auction timeout on optimized traffic. A single-bidder timeout, when the test has one, is applied with setBidderConfig for that bidder. The page script treats 2,500 ms as the untouched default: it leaves the publisher's timeout alone unless the test selects a different value. The holdback is never given the test timeout.
A timeout test is judged on revenue per session, with CPM, fill and latency as guardrails. A faster ad that loses the winning bid is not a keep. The metric and the guardrails are the same ones on how we measure.
The 10% holdback keeps your current configuration. A test runs on 30% of optimized traffic, for at least 72 hours and 2,000 sessions in each arm, and no longer than 14 days. bidtune keeps the change only after the capped session read is better at two daily checks.
How bidtune tests itTHE RULES
A test takes 30% of optimized traffic
Judged against its own control. The 10% holdback never changes.
10% holdback60% control30% test
sessions per arm2,000minimum runtime72hdecides onrev / session