The CPI Trap
Cost Per Install is easy to measure and easy to optimise. It is also, in isolation, the most dangerous metric in mobile marketing.
Cost Per Install emerged as the primary metric for mobile user acquisition because it is simple, comparable across campaigns, and responds predictably to bid adjustments. If you want to lower CPI, you broaden your audience targeting, reduce your bids, or shift spend toward lower-cost channels. In the short term, it works. CPI goes down. Volume goes up. The dashboard looks good.
What CPI optimisation actually produces
The users acquired through CPI-optimised campaigns are, by definition, the cheapest users to acquire — which means they are the users least motivated to engage with your product. Broad targeting captures users who install apps impulsively. Lower bids win auctions where your competitors have already decided the user is not valuable. Lower-cost channels reach audiences with lower commercial intent. You can acquire a million users at $0.80 CPI and generate less revenue than 50,000 users acquired at $12.00 CPI, because the latter group actually uses your product.
The metric that should govern acquisition
The correct north star for mobile user acquisition is Revenue Per Cohort — the cumulative revenue generated by a group of users acquired in a specific period, tracked over their lifetime in the product. This metric is harder to compute and slower to materialise than CPI, but it is the only metric that tells you whether your acquisition programme is generating value or consuming it. Every campaign optimisation decision should be traceable to its effect on Revenue Per Cohort.
Building the measurement infrastructure
Shifting from CPI to cohort-based measurement requires a mature mobile measurement partner implementation, clean event tracking across the user lifecycle, and the patience to let cohorts develop before making optimisation decisions. Most organisations are reluctant to invest in this infrastructure. The ones that do find that their acquisition costs fall naturally — not because they optimised for CPI, but because they stopped acquiring users who do not convert, and the algorithm learned what a valuable user looks like.