How it works
As a metric, CPI is calculated per campaign, network, country or creative: spend for the period divided by the installs attributed to it. Spend comes from the ad networks, by API or file import, and installs come from the measurement platform, so that every channel is counted by the same rules. When campaigns are bought on another basis, such as per impression or per click, the result is often called effective CPI (eCPI). As a pricing model, CPI means the network is paid only for installs that the agreed measurement platform attributes to it.
Why it matters for an app team
CPI is the first number that tells you what growth costs and whether a channel is getting more or less efficient. It is also the reference point for payback: a user is worth acquiring only if they return more than their CPI.
In practice
Example with made-up numbers: a campaign spends $1,500 in a week and is attributed 1,200 installs, so its CPI is $1.25. A second campaign spends $1,500 for 600 installs, a CPI of $2.50. If its users generate three times the revenue, it is still the better buy.
Common confusion
A low CPI is not a goal in itself; cheap installs that never return are expensive. CPI calculated from a network's own install count differs from CPI based on attributed installs. Blended CPI, which divides spend by all installs including organic ones, is a different and always lower number.