← All posts
4 min

Why we don't bid on every request we get

Written by The Adbustr team

Not every bid request that reaches our endpoint reaches a DSP. A meaningful share is stopped at the door, before we ask anyone to price it. We get questions about this from suppliers and partners. The short answer is that we think volume without quality is worse than no volume at all. The longer answer follows.

What we stop at the door, and what we call it

  • no_ip_ua — the request arrives without device.ip or device.ua. We cannot resolve geo, we cannot classify the device, and no DSP should be asked to bid blind.
  • browser_not_allowed — the user agent does not resolve to anything on our browser allow-list.
  • non_mainstream — the inventory declares IAB25, IAB25-3 or IAB26 on an endpoint running the mainstream filter.
  • no_imp — the request carries an empty imp array. There is nothing to auction.
  • no_feeds — the endpoint has no live demand feed that matches the request after geo, device and zone list rules are applied.
  • unknown_endpoint — the uid is not one of ours, or the endpoint is switched off.

A request can fail more than one of these. We log it once, with the reason that stopped it first, and the row carries the verdict — bid or nobid — alongside it. That funnel is the same one we read internally; there is no second, friendlier version of it.

A specific cleanup story

In late November we onboarded a publisher with a CTV inventory mix that looked good on paper. Within two days of going live, request volume from that endpoint ran far past anything their audience could plausibly produce, and nearly all of it resolved to a small block of rotating IPs inside a single ASN.

We throttled the endpoint from the config snapshot — one flip in Redis, no restart, no traffic loss on the way down — and called them. We walked the request log together and found a CDN configuration bug on their side that was rebroadcasting impression callbacks from cached responses. Once they fixed it, the next batch was clean and we opened the throttle back up. Their fill rate went from looking great-and-fake to looking modest-and-real, and that is exactly what we wanted.

We could have just bid through the noise. Some of those impressions would have rendered to real viewers. Some would not. We chose not to find out the hard way, on someone else's budget.

The position

We would rather show a DSP partner a real fill rate than a flattering one. Partners can ask for the reason breakdown on their own traffic and get it straight out of the per-request funnel, not out of a slide. If we filter your inventory and you disagree, we want to hear about it.

Programmatic ad tech has spent fifteen years optimizing for the wrong number. Adbustr is for people who want both sides of the auction clean. We hope that is most of you.

  • quality
  • compliance
  • policy