Feeding loss-notification data back into bids lifted win rate 7 points

Feeding loss-notification data back into bids lifted win rate 7 points

Most buyers analyze wins. One DSP buyer instrumented their losses and used them to recalibrate bids.

Loss notification (a signal some exchanges return explaining why a bid lost: below floor, lost the auction, or filtered out).

1. They captured loss codes per auction in log-level data and tabulated them by placement.
— On a key placement, 70% of losses were 'lost auction' (outbid), only 5% 'below floor'.
2. 'Lost auction' at this density meant their bid was close but consistently short, not floor-blocked.
3. They raised the bid on that placement by 12% and watched the loss-code mix shift toward 'won'.

Evidence: win rate on the placement rose from 21% to 28%, and because the added impressions converted in line with existing ones, cost per acquisition held flat despite the higher bid. The loss codes told them exactly which placements were near-misses worth pushing versus floor-blocked ones not worth chasing.

Why it matters: losses carry the signal wins cannot. A 'lost auction' code at high density marks a near-miss where a small bid increase pays off; a 'below floor' code marks a wall. Tabulating loss reasons turns bid tuning from guesswork into targeted adjustment.