A campaign’s performance today is judged by a combination of multiple Ad Sets and Ad Creative. Creative performance has always been one of the most questionable aspect of Facebook marketing. The reason behind this is Facebook’s campaign structure which tends to skew budgets on certain ad over the others within a specific Ad Set. This limits the advertisers learning since the creative performance might not have had the room to grow due to the said skew.
Facebook’s release of A/B test (or split test) was their way to combat issues like these.
A Split test is one of the most common experiment to test an elements performance over another. This test helps marketers like you analyse which element will deliver the best results in comparison to its counterpart. This test also involves factors such as a gestation period of the campaign duration, Ad dollar spend etc. However, the confidence on creative split test has not always been able to justify the real worth of the test.
Split test is a more long-term solution but not a feasible option for regular campaigns that are operationally heavy.
In an example below you would notice that Facebook’s split test treats one of the ads as “the winner” with 95% confidence even though the difference is lesser than the currency base of the account.
To be able to guide through this problem, our algorithm splits all your ads into three major buckets:
- Clear winners
- Similar communication
- Insignificant data
For the same test, the DMC optimisation engine would classify the two ads as “Similar Communication” due to one of the factors such as a Significance test that defines how similar or different are these creatives from one another. In the below example you would also see an instance where an ad with a higher CPA is getting a lot more exposure.
Along with Significance test, DMC’s creative optimization engine looks at Cost Per Action, Volatility, Frequency etc. to help campaign managers take an informed decision.
For any questions, reach out to us on denstumarketingcloud@dentsuaegis.com
