Does CPI matter in a ROAS-centric strategy? | Mobile Dev Memo

Does CPI matter in a ROAS-centric strategy?

With Google’s UAC and Facebook’s AEO and VO bid strategies becoming so prominent in mobile marketing strategy, the relevance of the CPI metric has diminished.

Last week I wrote about how I believe the LTV metric should be retired by mobile advertisers: the metric is anachronistic and doesn’t reflect the modern reality of mobile advertising, which increasingly is optimized for ROAS by black-box algorithms as on Facebook and Google. One common refrain I’ve heard in this context is that, when campaigns are operated against ROAS targets, CPA metrics like CPI don’t matter: the underlying cost of an event or install is irrelevant if a ROAS goal is hit.

This line of thought makes sense, to a point. One could extrapolate this logic to an extreme to showcase why marketers, to some degree, must care about CPI even if ROAS targets are being hit. Which of these cohorts would a company prefer?:

  1. 1,000 users in its app that were acquired at an average CPI of $1 (campaign cost: $1,000) and will completely recoup that cost in seven days;
  2. 1 user in its app that was acquired at an average CPI of $1,000 (campaign cost: $1,000) and will completely recoup that cost in seven days?

It’s fairly obvious that the company would prefer 1,000 users. But why?

For one, virality effects accrue at the absolute size of the DAU-base, so even if the k-factor is the same for both of these groups of users — at, say, 2 — then the group of 1,000 users generates an additional 2,000 users whereas the lone user in the second group recruits exactly two new users. One could quibble about the value of these virally-acquired users (is the $1,000-spending user recruiting other $1,000-spending users?), but if organics tend to look the same, then the organic value of the 1,000-user cohort is far higher than the value of the one-user cohort.

Additionally, one simply can’t know anything about a cohort of one: the law of large numbers helps marketing teams derive insights around user behavior and to build predictive models of churn and monetization, but those insights need to be applied to groups of users. A retention profile or recoup curve can’t be applied to a cohort of one: that user is an unpredictable liability. Diversifying the user base with a large number of users is a protection against knowable average behavior (like the retention profile) but especially against unknowable changes to behavior (eg. the market changes and the retention curve shifts downward).

One might look at the example above and call it fatuous, but a $1,000 CPI is not completely unrealistic: CPIs can easily creep into the mid-$100s with VO campaigns on Facebook for niche audiences. And the higher those CPI numbers climb, the more confidence an advertiser needs to have in the immutable shape of its ROAS curve: will ROAS progression really remain unchanged as CPIs increase? At CPIs in the hundreds of dollars, the cost of historical ROAS curves changing for new cohorts is substantial.

And what’s more, one has to recognize that the competition for (cost of) users that monetize to a high degree is disproportionate to their monetization: the competition for those users scales non-linearly as the best-monetizing apps fight to recruit them. So a high-value user might install the app and monetize along the schedule of the company’s observed ROAS curve, but the survival function for those users is different: they are more highly-sought than other users and thus they are seeing more relevant, precisely-targeted ads for other apps than other users are. In other words, the $1,000-spending user from the above example is under greater threat for being poached by a competing app than any of the $1-spending users are.

At the margins, a marketing team should be more focused on ROAS targets than CPI. But common sense needs to prevail in running ROAS-centric campaigns as cost creeps up: will new users really monetize in the same way that older ones have as campaigns target more and more precisely on the basis of expected revenue contribution, as the VO bid strategy does? It makes sense to always be conscious of CPI, even if the CPI number isn’t central to the campaign goal.

Photo by NeONBRAND on Unsplash