How to tune your bidder's timeout budget without dropping winnable bids

How to tune your bidder's timeout budget without dropping winnable bids

Your bidder has a fixed time budget to respond to each bid request; spend it wrong and you either time out or skip valuable decisioning. Tune it in this order.

1. Map your end-to-end response time into its stages: network ingress, request parsing, model scoring, bid construction, network egress. You can't optimize a stage you haven't measured.

2. Get the SSP's hard timeout per integration. Your p95 response must sit safely under it with margin for network jitter.

3. Find the stage eating the most time. If model scoring dominates, a heavier model that scores 5% better but pushes p95 over the timeout loses 100% of those auctions — a catastrophic trade.

4. Set a decision deadline inside your bidder: if scoring isn't done by the budget, fall back to a cached or default bid rather than timing out silently.

5. Tier integrations — give distant or strict-timeout SSPs a lighter, faster decision path.

6. Re-measure p95 after every model or infrastructure change; latency regressions are invisible until win rate drops.

Why it matters: There's a hard ceiling where a smarter bid that arrives late is worth nothing, and teams chasing model accuracy routinely cross it without noticing. Treating the timeout as a fixed budget — and spending it deliberately across stages — protects the win rate that all the cleverness depends on.