Research we carried out for a specific advertiser revealed that two critical elements of campaign insight, attribution and performance data only matched half of the time.
- On a campaign level, we generally see that DSP conversions are approximately 60% higher than Ad Server conversions, but this figure can easily vary from -20% to 140%
- When we dove into the log level data for an individual advertiser then accuracy is just as unpredictable.
- 51% of the time DSP and Ad Server agree that the same impression was responsible for a conversion.
- On average conversions on any given site are 54% lower on Ad Server compared with DSP
- 66% of the time, the number of Ad Server conversions on a site was over 50% less than the DSP.
- 39% of sites that the DSP has recorded a conversion against had 0 corresponding Ad Server conversions
- 15% of sites that the DSP had recorded at least 3 conversions against had 0 corresponding Ad Server conversions
- On average, DSP conversions on any given site are between 12% and 79% higher than what Ad Server reports (taking outliers out)
- On average, DSP conversions on any given exchange are between 20% and 65% higher than what Ad Server reports (taking outliers out)
- DSP was recording more desktop conversions, whereas Ad Server tracked the best performance on smartphones.
- On average, DSP conversions on any given site are between 22% and 138% higher than what Ad Server reports. (taking no outliers out)
- On average, DSP conversions on any given exchange are between 23% and 72% higher than what Ad Server reports. (taking no outliers out)
Whilst we have headline stats above from site-to-site and exchange-to-exchange we saw lots of variability, and not even in a particular direction. Sometimes the DSP conversion count was much greater than the Ad Server count, and sometime the other way around. The ranges of difference between DSP and Ad Server was anywhere between 20% and 140%, a huge margin of inaccuracy. This means you can’t just assume an overall level of error and make an adjustment for it (for instance always reducing DSP conversion by 50%) – you need to use data on a case-by-case basis to inform your optimisations and adjust your setup.