A log-level audit found 8% of 'won' impressions were never rendered

A log-level audit found 8% of 'won' impressions were never rendered

Log-level data (the row-per-event export of every bid, win, and impression a DSP processes). One buyer reconciled their win logs against actual rendered impressions and found a leak.

1. They joined the DSP's win-notification log to the publisher's impression-tracker log on the auction ID.
2. They counted wins with no matching render event within the attribution window.
— 8.2% of recorded wins had no impression render.
3. They segmented the gap and found it concentrated on two SSPs, where it ran above 15%.

The cause was win-notification timing: those SSPs fired the win signal at auction close, before the creative actually had a chance to render. Bid caching and downstream client drop-off meant a meaningful share never served, but the buyer was billed on the win.

Evidence: shifting spend away from the two high-discrepancy SSPs raised the win-to-render ratio from 92% to 99% and cut wasted spend by roughly 7% of the line's budget.

Why it matters: a win is not an impression. Reconciling win logs against render logs on the auction ID is the only way to find the gap, and the gap is rarely evenly distributed across your supply.