In one line
In environments like SKAdNetwork you cannot see who installed at the user level, so CPI cannot be counted directly. eCPI (Estimated CPI) is the figure a network models back from spend and aggregated conversion data instead.
How it differs from regular CPI
Regular CPI is a straightforward calculation: cost ÷ actual installs counted. eCPI is an estimate built from a network's own modeling (SKAN aggregate conversion values, probabilistic matching, etc.) — and that modeling error varies by network and by period.
Why it has to be an estimate
Without ATT consent on iOS the advertising identifier is unavailable, so no one can link a specific click to a specific install. SKAdNetwork sends postbacks aggregated at campaign level instead — group information, not people.
On top of that, SKAdNetwork withholds detailed conversion values when a campaign falls below its crowd anonymity threshold. The lower the volume, the more blanks a platform has to fill in, and the wider eCPI's error grows. When platform-reported installs and your own numbers disagree for the same campaign and period, this is usually why.
Why to be careful
eCPI is an estimate, not a count — it can diverge from your own tracked install numbers. Don't judge a channel as "cheap" on eCPI alone; cross-check against your own server-side conversion data where possible.
In practice
Optimizing iOS campaigns purely on eCPI can push budget in the wrong direction. Look at SKAN aggregate data alongside your own first-party conversion signals to confirm direction.