What is CPI (cost per install)?

What CPI does and does not guarantee

CPI guarantees an install. It guarantees nothing about the user attached to it.

That distinction is the whole discipline of app marketing. Install quality varies enormously between sources, and the correlation between CPI and quality runs the wrong way: the cheapest installs reliably come from the least engaged users, incentivised traffic, or outright fraud.

A source at half the CPI with a third of the retention rate is a considerably worse buy, and no install-stage metric will show it.

What to measure instead

• Cost per activated user, where activation is the point the product becomes useful.
• Day 7 and day 30 [retention](/resources/glossary/retention-rate) by source, cohorted.
• [LTV](/resources/glossary/lifetime-value-ltv) to CAC by source, which is the only ratio that decides whether a channel should scale.
• [IPM](/resources/glossary/installs-per-mille-ipm) as a creative diagnostic rather than a buying metric.

What privacy changes did to CPI buying

CPI as a price still works — an install is still countable. What broke is the link between the install and everything after it.

On iOS, SKAdNetwork reports installs in aggregate with a small, delayed conversion value as the only post-install signal. Judging source quality now depends on encoding the right thing into that value in advance, and on keeping campaign structures consolidated enough to clear anonymity thresholds.

The practical effect is that CPI buying has become less granular and slower to optimise, and the teams that adapted best moved decision-making up a level — fewer, larger campaigns judged on cohort outcomes rather than many small ones tuned daily.