March 29, 2017
About the author : Joni holds a PhD in marketing. He is currently working as a postdoctoral researcher at Qatar Computing Research Institute and Turku School of Economics. Contact: joolsa (at) utu.fi
This is a follow-up post on my earlier post about “fake” conversions — the post is in Finnish but, briefly, it’s about the problem of irreversibility of conversions in the ad platforms’ reporting. In reality, some conversions are cancelled (e.g., product returns), but the current platforms don’t track that.
So, my point was to include a ‘churn coefficient’ which would correct for the CPA calculation. In other words, it adjusts the CPA reported by the ad platform (e.g., AdWords) in regards to churn from “conversion” to conversion (as per the previous explanation).
The churn coefficient can be calculated like this:
in which churn is the churn from the reported conversion to the lasting, real conversion.
However, I got to think about this and concluded this — since we consider the churn taking place due to real world circumstances as a lift to the reported CPA, we should also consider the mitigating factor of customer-to-customer references (i.e., word-of-mouth).
Consider it like this – on average, converted customers recommend your company to their friends, out of which some convert. that effect would not be correctly attributed to the referring customers under normal circumstances, but by attributing it uniformly to the average CPAs we can at least consider it in aggregate.
So, hence the ‘wom coefficient’:
1-(Cn / Cm), in which
Cn: conversions from new customers non-affiliated with any marketing channel
Cm: conversions from all marketing channels
The idea is that the new visitors who convert can be attributed to wom while conversions from marketing channels create the base of customers who are producing the recommendations. Both pieces of information can be retrieved in GA (for Cn, use an advanced segment).
So, the more accurate formula for “true” CPA calculation would be:
1-(Cn / Cm) * 1/(1-churn) * CPA
In reality, you could of course track at least a part of the recommendations through referral codes (cf. Dropbox). In this case you could have a more accurate wom coefficient.
Consider that in period t, not all Cn are created by Cm. Hence, it would be more realistic to assume a delay, e.g. compare to period t-1 (reference effect does not show instantly).
The formula does not consider cases where the referred customers come through existing marketing channels (this effect could be eased by not including branded search campaigns in Cm which is a good idea anyway if you want to find out the true performance of the channel in new customer acquisition).
Finally, not all customers from non-marketing channels may not originate from wom (especially if the company is using a lot of non-traceable offline marketing). Thus, the wom efficient could have a parameter that would consider this effect.
Dr. Joni Salminen holds a PhD in marketing from the Turku School of Economics. His research interests relate to startups, platforms, and digital marketing.
Contact email: [email protected]