3. Reporting Discrepancies
“There are three sides to every story: your side, my side, and the truth.”
“Ad serving discrepancy refers to a difference between the number of ad impressions counted by a publisher or ad network adserver and the one counted by the agency or advertiser ad server. Ad serving discrepancy can be the source of billing issues.”
It’s unthinkable your bank would “lose” 10% of a transaction between accounts but bizarrely in advertising technology it’s “normal” to have conflicting counts of impressions & inventory between ad-servers. The truth is ad-serving isn’t easy and the challenge of this fast-moving technology is most acutely felt in the often painful reality of reporting discrepancies.
“Mobile discrepancies can range from 5-50% depending on a number of factors.”
A serving discrepancy is the difference of impression & request counts between a third-party reporting network and an ad-server counts. Or put another way, the difference of ad-server measured transactions vs the network (buyer) transactions. In our experience mobile advertising has an accepted rate of discrepancy (usually 10-15% depending on the medium) and given the massive numbers of impressions, complexities of reporting, attribution, technology and fraud-prevention techniques in place, it’s not surprising this is an accepted shortcoming in the market.
How are they caused?
As you may imagine the list of why the ad-server vs network counts are off is virtually endless but there are a common list of problems, here are the ones we most commonly see:
- Incomplete information: Obviously if you are serving a portion of traffic outside of your ad-server (via another exchange or ad-server, a high discrepancy will exist between your network and ad-server.
- Fraud Prevention / Difference of the definition of “impression”: Fraud is a touchy subject with most networks but commonly we see an increasing number of discrepancies arise from a network “attempting” to buy an impression but not counting the impression due to the impression not being “delivered.” For instance if your ad-server sends an impression to a network, the network buys said impression but the ad only stays on screen for 2 seconds before the user moves to a different screen, the network may not count that impression against their total, where MoPub will. This can be true with occluded, or incompletely rendered impressions.
- Time zone differences: Where possible we unify time-zones but some networks don’t allow reporting to be called in UTC (standardized ad-server timezone), therefore daily impressions may spill into different days.
- Failed impression passbacks: If a network attempts to pass an impression back to your ad server as “unfilled” but your ad server doesn’t receive said impression, the ad server will consider this impressions as “served” but the network won’t.
- Broken integrations: similar to missing passbacks, if an impression is incorrectly handed to a network (via SDK or tag) the network may not recognize the impression and it could be lost.
- Time-outs: If a network missed the ad server timeout in handing back an unsold impression, the ad-server will mark as sold, the network will not.
Why this should concern you:
While discrepancies may part of ad serving technology you should be aware of the impact on your bottom line. They can be an indication of lost impressions, the thus lost revenue. For instance, if you optimize using Network CPM and a network is losing/not-counting 50% of incoming impressions, the actual CPM could be 50% lower than the network’s measurement, so you might be better off selling to a network with a lower CPM and lower discrepancy.
In cases of large discrepancies we advise testing optimizing using MoPub-calculated CPMs which unify measurement of all networks and keep a common method of counting. This doesn’t always work however, as sometimes the mis-measurement can be tied back to MoPub counts.