Zovda
Campaign reporting is built on a join between an impression and an action. The join only exists for people who clicked, which is a small and unusual subset of the audience.
Everything else is modeled, and the model is usually invisible in the report.
Three populations, three methods
- Clicked and converted: measurable, and the least interesting
- Saw and converted later: needs a match or a holdout
- Saw and did nothing: the majority, and the one that decides brand impact
Reports that collapse all three into one number describe the first population with the language of the third.
Holdouts are the honest method
A small share of the audience that is deliberately not exposed, compared against a matched share that is. The difference is the effect, without a click in the middle.
It costs reach and it buys a number you can defend.
Match on what, exactly
Matching by time window and geography produces an effect that includes everything else happening in that window. Matching by prior behavior produces a smaller, more credible number.
Report which match was used, because the two are not comparable.
Keep the unmatched rows
Rows that could not be joined are not missing data. They are the population the report is silent about, and their count belongs in the footnote.
Zovda assembles these collections through Datuva, where the exposure and the outcome stay in separate, joinable records. The data research collection shows how they are carried.
Tell us what you need to prove. We reply within one business day.