Discrepancy

Buyer and seller rarely count exactly the same thing. The publisher's ad server might record 1,000,000 impressions, the advertiser's advertiser ad server 930,000, and the verification vendor 910,000. The gap is a discrepancy.

Common causes include timing (one system counts when the ad is requested, another when it renders), users leaving before an ad loads, latency and timeouts, ad blockers, different invalid traffic filtering, time-zone and reporting-window differences, and tracking tags that were not installed correctly. In video and CTV, SSAI and different counting points in the VAST response add more room for difference.

Contracts usually agree which system is the billing source and what level of discrepancy is tolerated; around 10% is a common rule of thumb. Larger or sudden gaps deserve investigation, because they can signal broken tags, but also fraud, such as impressions claimed by a seller that never reached a real browser. Consistent discrepancy analysis is one form of data anomaly detection.

Think of it like this

A discrepancy is like two people counting the same crowd from different gates: they will be close, but rarely identical.

An example

A publisher in Italy bills 2.0 million impressions; the advertiser's ad server shows 1.62 million, a 19% discrepancy. Investigation finds a mobile placement where the advertiser's tag failed to load in one browser.