Cost per install (CPI) is an advertising metric and pricing model that expresses how much an advertiser pays, on average, for each new installation of a mobile app driven by a paid campaign. In performance marketing it is the entry-level efficiency measure for app user acquisition, calculated as campaign spend divided by attributed installs.
CPI can describe either a result (what installs actually cost under CPM or CPC buying) or a billing model in which the advertiser is charged only when an install occurs. In both cases, an install is usually counted at the first app open, not at the moment the download starts, because the attribution software inside the app must run to record it.
How cost per install is calculated and attributed
The formula is simple: CPI equals total ad spend divided by attributed installs. A campaign that spends 6,000 in a given currency and generates 2,400 attributed installs has a CPI of 2.50. The difficult part is not the arithmetic but deciding which installs belong to which campaign.
Attribution is handled by mobile measurement partners (MMPs) and by the ad platforms themselves. On Android, attribution typically relies on the advertising identifier and the install referrer. On iOS, since the App Tracking Transparency framework required user permission for cross-app tracking, much measurement runs through Apple’s privacy-preserving frameworks, SKAdNetwork and its successor AdAttributionKit, which report aggregated and delayed results.
| Input | Typical setting | Effect on reported CPI |
|---|---|---|
| Attribution window | Often 7 days after click, 1 day after view | Longer windows attribute more installs |
| Install definition | First app open | Downloads never opened are excluded |
| Organic baseline | Installs without paid touch | Mixing in organics understates CPI |
| Platform | iOS or Android | iOS reporting is often aggregated and delayed |
| Re-installs | Counted or excluded | Counting them lowers apparent CPI |
| Currency and taxes | Net or gross media cost | Fee treatment changes comparisons |
Because these settings differ between an ad network’s dashboard and an MMP, the same campaign can show different CPIs in different tools. Teams should agree on one source of truth before comparing channels.
What drives CPI up or down
CPI levels vary widely, from well under one unit of currency for simple games in lower-cost markets to several units or more for finance, subscription and productivity apps in high-income countries. The variation follows predictable factors rather than chance.
| Factor | Tends to raise CPI | Tends to lower CPI |
|---|---|---|
| Country tier | High-income markets with strong competition | Emerging markets with cheaper inventory |
| Operating system | iOS audiences, often higher-spending | Android audiences in many markets |
| App category | Finance, subscriptions, strategy games | Casual and hyper-casual games |
| Targeting depth | Narrow audiences, value-based bidding | Broad audiences, install-volume bidding |
| Creative quality | Weak or fatigued creatives | Strong hooks and clear app previews |
| Store conversion rate | Poor store listing and ratings | Optimized screenshots and reviews |
| Seasonality | Holiday periods with high demand | Quieter periods after peaks |
Two levers are under direct control and often underestimated: creative testing and app store optimization. If the store page converts poorly, every paid click is wasted on a visitor who leaves, so improving store conversion lowers CPI on every channel at once.
CPI compared with related performance metrics
CPI is a top-of-funnel metric. It says how cheaply users arrive, not whether they stay or pay. Mature app marketers therefore read it together with downstream metrics and optimize toward value rather than volume.
| Metric | Formula | Question it answers |
|---|---|---|
| CPI | Spend divided by installs | How cheaply do users arrive |
| CPA or CPE | Spend divided by in-app events | What does a registration, trial or level cost |
| CAC | All acquisition costs divided by paying customers | What does a real customer cost |
| ROAS | Attributed revenue divided by spend | How much revenue does each unit of spend return |
| CLV | Expected revenue or margin per user over time | How much can be paid for a user |
The decisive relationship is between CPI and lifetime value. A CPI of 4.00 is excellent if users generate 12.00 in margin and unsustainable if they generate 1.50. This is why bid strategies have moved from install volume toward in-app events and value-based bidding.
Bidding and optimizing for installs
Major app campaign platforms let advertisers set a target CPI, an install volume goal or a target cost per in-app action. Automated bidding then adjusts bids per auction, which means that a target set too low restricts delivery and one set too high wastes budget on low-quality users.
- Start with install optimization to collect data, then move to event or value optimization once enough conversions accumulate.
- Change targets gradually, because large jumps reset the learning phase of automated bidding.
- Refresh creatives regularly; fatigue raises CPI even when targeting is unchanged.
- Evaluate by cohort, comparing retention on day 1, day 7 and day 30 for users acquired in the same week.
- Separate incentivized traffic, where users are rewarded for installing, because its retention is usually much weaker.
Pitfalls and install fraud
A low CPI is not automatically good news. Install fraud specifically targets CPI-priced campaigns because fraudsters are paid per attributed install. Unusually cheap installs with near-zero retention are a classic warning sign.
| Fraud type | Mechanism | Typical signal |
|---|---|---|
| Click spamming | Mass fake clicks claim organic installs | Low click-to-install rates, long click-to-install times |
| Click injection | Malware fires a click just before an install completes | Extremely short click-to-install times |
| SDK spoofing | Fake install signals sent without real devices | Installs with no in-app activity |
| Device farms | Real devices installed repeatedly by hand or script | Clustered device models and locations |
Fraud protection features in MMPs, post-install event validation and payment terms tied to retained users rather than raw installs are the standard defenses.
A non-technical view
Cost per install is the price of getting one new person to install and open an app through advertising. If a company spends 1,000 on ads and 500 people install the app, each install cost 2.
It is like a shop paying for flyers and then counting how many people walked in because of them. A cheap visit is nice. What matters more is whether those visitors come back and buy something.
A concrete example
A language-learning app runs install campaigns on two networks in Germany and the United Kingdom. Network A delivers installs at a CPI of 1.20, while Network B costs 3.10 per install. At first glance, Network A looks almost three times more efficient.
The growth team compares cohorts in its MMP. Only about 6% of Network A users are still active on day 7, and almost none start a paid trial. Network B users show day-7 retention near 25% and a trial start rate several times higher. Fraud analysis also flags many Network A installs with suspiciously short click-to-install times.
The team pauses most Network A spend, switches Network B campaigns to optimize for trial starts instead of installs and tests new video creatives. Blended CPI rises, but the cost per paying subscriber falls substantially, and the marketing budget is now judged by subscriber value rather than install volume.