Timeouts and Bidder Pruning: A Header Bidding Tune-Up for Gaming Sites

Gaming sites have an audience that ad tech tends to underserve. Readers of game guides, wikis, patch notes and esports coverage often arrive with a specific task, scroll quickly, open several tabs and move between pages in rapid succession. Many are on capable desktops, but a large share also read on phones while playing. That behavior is unforgiving of slow auctions. An ad that shows up two seconds after the reader has scrolled past it is an impression nobody saw and a buyer who will bid less next time.

Two settings do more than anything else to decide how fast your header bidding auction completes: the timeout, and the number of bidders you invite. Most sites set both once and never revisit them. Here is how to tune them properly.

How the timeout actually works

In Prebid, the bidder timeout is the maximum time the auction waits for bids before sending what it has to the ad server. Bids that arrive late are not used for that auction. The Prebid documentation on timeouts describes several layers: the auction timeout itself, a separate Prebid Server timeout that should be set lower than the auction timeout so server-side bids return in time, and a failsafe timeout outside Prebid that makes sure the ad server is called even if something breaks. It also points out that consent platforms and user ID modules can add delay before the auction even starts.

The trade-off is simple to state and hard to get right. A longer timeout collects more bids but delays every ad. A shorter timeout shows ads sooner but drops bids from slower partners.

Step 1: Measure before you change anything

You cannot tune what you cannot see. Make sure you have a Prebid analytics adapter reporting auction-level data. For each bidder, you want:

• Bid request count and bid rate (how often it responds with a bid)
• Response time distribution, not just the average, especially the 90th and 95th percentiles
• Timeout rate (how often it responds too late)
• Win rate and revenue contribution

Split all of it by device type and by country. A bidder that responds quickly on desktop in the US may be slow on mobile in Southeast Asia, and gaming audiences are often global.

Step 2: Set the timeout from data

Look at where your valuable bids actually arrive. For most setups, a large majority of winning bids come in well before the timeout. The question is how much revenue sits in the tail.

• If your top revenue bidders complete most responses comfortably inside your current timeout, and very little revenue arrives near the limit, you can likely shorten the timeout without losing much.
• If a meaningful share of winning bids arrive close to the limit, shortening it will cost money. Look at why those bidders are slow before cutting.

Consider different timeouts by device. Mobile connections are slower and less predictable, but mobile readers also scroll faster and have less patience. The right answer is not always "longer on mobile."

Step 3: Prune bidders that do not earn their place

Every bidder adds a network request, script weight and processing. On a busy gaming wiki page with several ad units, that cost multiplies. For each bidder, ask:

1. Does it win? A bidder with a high bid rate but almost no wins is adding latency for competition that rarely matters.
2. Does it raise prices when it loses? Some bidders rarely win but push up clearing prices. Check whether removing them in a test lowers winning bids.
3. Is it unique demand? If a bidder mostly resells the same demand available through other partners, it is adding duplicate requests without adding buyers.
4. Is it consistently slow? A bidder with a high timeout rate is paying for its seat with your users' time.

Removing a bidder is not a permanent judgment. It is a test with a measurable outcome.

Step 4: Move some demand server-side, carefully

Prebid Server lets you run some bidders from a server instead of the browser, which reduces the number of requests the page makes. That can speed up pages significantly, especially on mobile. The trade-off is that server-side bidders may see less user data and sync fewer cookies, so they can bid lower. Move bidders server-side one at a time and compare revenue per session, not just speed.

Step 5: Load and refresh with intent

Timeouts interact with how and when you start auctions.

• Lazy loading. Start auctions for below-the-fold slots shortly before they scroll into view, not at page load. For fast-scrolling readers, trigger them a little earlier than you would on a news article.
• Refresh. Long guide sessions invite refresh. Refresh only viewable slots in active tabs, keep reasonable intervals, and declare refresh to demand partners accurately.
• Single-page navigation. Many gaming sites and wikis load new content without full page reloads. Make sure old slots are destroyed and new auctions are triggered cleanly on each view.

Step 6: Prove it with a controlled test

Run changes against a control group. Split traffic randomly, apply the new timeout or bidder set to the test group, and run the test long enough to cover weekdays and weekends, since gaming traffic often peaks differently than news. Compare:

• Revenue per thousand sessions
• Viewability
• Page load and layout stability metrics
• Pages per session

The last one matters. Faster pages often lead to more pages per session, and more pages means more impressions. A change that looks slightly negative on CPM can be clearly positive on revenue per session.

A tune-up schedule

Review timeouts and bidder performance at least quarterly, and after any major game launch or seasonal traffic change. Bidder performance drifts, partners change their infrastructure, and your audience mix shifts. A setup tuned a year ago is almost certainly leaving money or speed behind.

The takeaway

For gaming publishers, auction speed is not a technical nicety. It decides whether ads load while readers are still looking. Measure bidder behavior, set timeouts from real response data, prune partners that do not add value, and prove every change with a test. It is routine work, but it compounds, and it is the kind of ongoing tuning a managed header bidding setup should handle for you.