Almost a year and a half has passed after the GDPR roll-out, but many questions remain unanswered and many publishers have a lot of interrogations about the impact they should be seeing on their websites and specifically on their advertising revenues.
Some publishers think that the only thing to do to keep things rolling is to implement a Consent Management Platform. While this is true from a compliance point of view, it is far from being optimal for your advertising revenues.
Relying on CMP reports solely is not a good choice! Basically, these reports show you the proportion of consent acceptation and refusal on your website. But, knowing that a user gave his consent DOES NOT mean that all the advertising auctions of this user have had his consent.
Moreover, a CMP can report up to 95% of accepted consent, but the CMP itself may be badly implemented on a part of the website, meaning that in its report, it won’t show the percentage of unavailable consent which can be pretty high as the figures below show.
For the purpose of this survey, our Data Science team has conducted a survey on more than 100 billion ad auctions to get a better understanding of the GDPR user consent impact on publishers’ advertising revenues.
1 – Survey scope and context definition
The study was conducted on our network to estimate the impact of GDPR user consent on RPM. As said previously, the study is based on the analysis of more than 10Bn auctions and more than 1000 websites, with a strong focus on some websites for A/B testing purposes and to get deeper insights.
IMPORTANT FACT: we led our study on the probed auctions and bid requests themselves, instead of the “user session” which is much more relevant in order to make the bridge with advertising performance.
When it comes to consent, there are many possible states depending on the vendors and purpose options. But we have decided to simplify the model by classifying “Auctions” and “Bid requests” in 3 different states:
– Consent accepted in the auction, which means that at least one purpose or one vendor was accepted by the user,
– Consent refused in the auction, which means all purposes and all vendors were refused by the user.
– Consent not available, which means that the content string was found EMPTY. This could mean that there is no CMP implemented or it can be a sign of bad CMP implementation.
2 – What is the consent acceptance rate?
For starters and in order to set the context, below are some figures showing the occurence for each of the 3 states:
As you can see, “Refused consent” is really low and it is usually the case, as all current CMPs are pretty pushy to get the “accepted consent” (which won’t be the case anymore with TCF V.2).
The “Unvailable” state is high, and this occurs when there is no CMP implemented, it can also be a sign of bad CMP implementation.
3- What is the impact on RPM and advertising performance?
To get deeper insights, we ran an advanced study on 100 websites where all three states where represented (accepted, not available, refused).
We can clearly identify a MASSIVE gap in advertising performance between all the three states.
|Metrics||Accepted consent||Consent Unavailable||Refused consent|
|RPM Auction WON||2€||1.4€||0.8€|
|SUM of Auctions||108 098 (58%)||38 513 (21%)||39 626 (21%)|
To sum up:
Below is a table showing the corresponding bidder behavior. Individually, each bidder is impacted, with different levels of amplitude and the impact is HUGE.
|Bidder RPM (x1000)
|Bidder RPM (x1000)
|Bidder RPM (x1000) Refused|
4 – Conclusion:
Figures speak for themselves; user consent is definitely a metric publishers should care about and must start monitoring as it is greatly correlated to advertising performance.
While an even deeper study should be conducted to confirm these results. The consistency between all figures is enough to conclude that there is a 40% gap in revenue between the “not available” state and the “accepted” state.
AdX is not included in this study as the access to its bid stream is more complicated, we will dive into this specific case in another survey.