Affiliate marketing · Glossary

CPI Cost per install

In plain words

CPI means you are paid every time someone installs an app through your link and opens it. The payout is small, often well under a dollar, but installs are the easiest paid action in mobile, so volume does the work. The catch is that advertisers watch what users do after the install, and traffic that installs but never uses the app gets cut.

Definition

Cost per install (CPI) is the mobile-app version of CPA: the advertiser pays a fixed amount for each new install attributed to the affiliate. “Install” in practice means the app was downloaded from the store and opened for the first time, because that first open is the event an attribution SDK can record. Many offers add a condition, such as reaching a tutorial step, registering or staying active for a day, and the fine print, not the headline, says which one triggers the payout.

Attribution on mobile runs through measurement partners (MMPs) such as AppsFlyer, Adjust or Branch. The affiliate link passes a click ID to the MMP, the app reports the first open through the SDK, the MMP matches the two and sends a postback to the network. Since Apple’s App Tracking Transparency arrived with iOS 14.5 on April 26, 2021, matching on iOS is largely probabilistic or runs through SKAdNetwork, which delays and aggregates results; Android still relies on the Google Play Install Referrer and advertising ID where the user allows it. The practical effect is that iOS CPI campaigns report slower and less precisely than Android ones.

CPI offers split into incentive and non-incentive. Incentive traffic (offerwalls, rewarded placements, cash-back apps) pays users to install, so it converts extremely well and retains badly; payouts are a fraction of non-incentive rates and only some advertisers accept it. Non-incentive traffic (social, in-app display, ad networks) costs more per install but produces users who chose the app. The offer tag tells you which you may send, and mixing the two is the fastest way to a rejected invoice.

Behind every CPI campaign sits a retention metric the advertiser cares about more than the install: day-1 and day-7 retention, registrations, purchases. Gaming advertisers in particular publish target KPIs with the offer and pause sources that miss them. For an affiliate, CPI is therefore a quality deal disguised as a volume deal: cheap installs that do not retain disappear from the payout within a reporting cycle.

In practice

Case from the industry

What ATT did to iOS install attribution

When iOS 14.5 shipped on April 26, 2021, every app had to ask permission before tracking a user across other companies’ apps and websites. Flurry Analytics reported that in the first weeks only about 4% of US iPhone users, and roughly 12% worldwide, allowed tracking. The device identifier that attribution had relied on effectively disappeared for most of the iOS audience.

For CPI campaigns this changed three things at once. Attribution on iOS moved to Apple’s SKAdNetwork, which reports installs in aggregate and with delays rather than per click, and to probabilistic matching by the MMPs where allowed. Reporting that had updated within minutes now took a day or more to settle, and campaign-level optimisation lost the granular data it was built on. And because Android attribution kept working as before, payouts and allowed traffic began to diverge by platform; many offers today list separate CPI rates and KPIs for iOS and Android. An affiliate who treats the two as one offer is reading the wrong number for one of them.

Common mistakes

  • Assuming “install” is the payable event. Read whether the offer requires an open, a registration or a day-1 return before paying.
  • Sending incentivised traffic to a non-incentive offer. The install numbers look great until the advertiser compares retention and declines the batch.
  • Ignoring platform differences. iOS attribution after ATT is slower and more aggregated than Android; judge iOS campaigns on settled data, not day-one dashboards.
  • Optimising for installs instead of the advertiser’s KPI. If the offer states a day-7 retention target, that is the number that keeps your source alive.
  • Forgetting the store step. Users who click but abandon on the store page are counted as clicks, not installs; a store listing with poor ratings or a heavy app size lowers your conversion rate without any fault in the ad.

Go deeper

FAQ

How is an install tracked?

Through a mobile measurement partner (MMP) SDK inside the app. Your link records a click with a click ID, the SDK reports the first app open, the MMP matches the two and fires a postback to the network. On iOS the match is often probabilistic or aggregated through SKAdNetwork.

What is the difference between incentive and non-incentive CPI?

Incentive CPI rewards the user for installing (points, cash, in-game currency); non-incentive relies on the user wanting the app. Incentive installs convert far more often, retain far worse and pay much less, and most advertisers allow only one of the two.

Why was my CPI campaign paused despite good install numbers?

Almost always because of post-install KPIs. Advertisers compare retention, registrations or purchases per source and pause the ones that miss their targets. Ask for the KPI before launch and track it per placement.

Which GEOs pay the most on CPI?

Tier-1 markets such as the US, UK, Canada, Australia and Western Europe, where users monetise best. Payouts in lower tiers can be a tenth of that, but so are the traffic costs, so EPC and margin matter more than the absolute rate.

Sources

  1. What is CPI (cost per install)? — AppsFlyer glossary (appsflyer.com)
  2. User privacy and data use (App Tracking Transparency) — Apple Developer (developer.apple.com)
  3. iOS 14.5 opt-in rate: daily updates since launch — Flurry Analytics blog (flurry.com, 2021)
  4. Apple’s App Tracking Transparency feature has arrived: here’s what you need to know — TechCrunch (techcrunch.com, 2021)
  5. Google Play Install Referrer API — Android Developers (developer.android.com)

References are listed as plain text on purpose; look them up by title and publisher. Updated: 2026-10-05.

Terms are the easy part.

Run them inside a network with real tracking, roughly 48-hour payouts and a dedicated manager.

Join the network