What Is Cost per Install?

Cost per install is ad spend divided by attributed app installs in the same period. The result depends on which installs count and which system assigned them.

Cost per install (CPI) is advertising spend divided by the number of app installs you attribute to that spend in the same period. If the period, the spend, and the install count come from different systems, the ratio is not the campaign's CPI. It is a mix of definitions.

CPI is an acquisition cost. It does not include what those installs earn later.

How do you calculate it?

Pick three boundaries and write them down.

  1. Spend. The amount the ad platform charged, in one currency, for the campaigns in the report. Fees, credits, and taxes belong in the definition. Spend lives in the ad account. An analytics pipeline that never imports it cannot compute CPI on its own.
  2. Installs. The count you accept as the denominator. That might be the ad platform's self-attributed installs, the mobile measurement partner's installs, or SKAdNetwork postbacks. These counts are not interchangeable. Web ad, app install: who gets the credit? is the short version of why Meta, Google, and TikTok can each claim the same install.
  3. The period. Use the same dates and time zone for spend and installs. A click on the last day of the month can produce an install the next month. Your window decides which month keeps it.

CPI equals spend divided by installs. A campaign with spend and zero installs has no CPI. Report it as undefined, not as zero. Zero would read as a free install.

Reinstalls and redownloads need an explicit rule. Apple's postback field redownload marks an Apple Account that had the app before. It does not name the person. If you exclude redownloads from the denominator, say so, or CPI falls whenever redownloads are common.

A concrete example

A campaign report has three install columns for one week:

  • The ad network reports installs it claims under its own click and view windows.
  • The MMP reports installs it awarded after its attribution rules.
  • SKAN postbacks with postback-sequence-index 0 are a third count, delayed and missing some campaigns at low crowd anonymity tiers.

Spend is one number, from the ad account, in currency USD. Dividing that spend by each install column produces three CPI figures. They will not match. Publishing one of them as "the" CPI without the column name hides the gap.

Do not fill a missing SKAN install with the network's install to force the columns to agree. The gap is the measurement difference you are trying to see. Attribution is the general version of this problem. Install attribution is the mobile case.

What are the limits?

CPI ignores revenue. A lower CPI can coincide with worse predicted lifetime value and worse payback. Read it next to a value metric, not instead of one.

View-through installs and click-through installs do not cost the same kind of claim. If the denominator includes view-through conversions, CPI drops without a matching drop in what you pay for clicks. Say whether the count includes them.

Organic installs do not belong in a paid CPI denominator. They had no spend in this ratio. Blending them in makes paid CPI look lower.

iOS totals that depend on SKAN are late. Computing CPI on the day of spend, before the postback delay, undercounts installs and overstates CPI. Wait for the window described in the SKAdNetwork guide.

Server-side delivery of an install event does not bypass App Tracking Transparency, and it does not force two networks to share one winner.

How Hardal handles it

Hardal can record the install and first-open events your app or measurement setup sends, and the purchase events that follow. It does not replace the ad account as the source of spend, and it does not pick a single winning network when several claim the install.

Build CPI only after you name the spend export and the install column. Keep the network count, the MMP count, and the SKAN count separate. Hardal for Gaming and the revenue signals note are the surrounding context: a cheap install is not the goal if the value events never leave your stack.

Related terms: Attribution, SKAdNetwork, App Tracking Transparency, Web-to-app

Frequently asked questions

How is CPI different from CPA?

Cost per install divides spend by installs. Cost per acquisition divides spend by whatever action you defined as the acquisition, such as a registration or a paying user. The same campaign has one CPI and a different CPA.

Why do Meta, Google, and an MMP show different install counts?

Each one can claim the install under its own click window, view window, and matching rules. A self-attributing network does not have to agree with the mobile measurement partner. Compare them as separate columns.

Do reinstalls count in CPI?

Only if your definition says so. SKAdNetwork can flag a redownload. An MMP may count a reinstall as a separate install or ignore it. State the rule before you compare CPI across partners.

Does a lower CPI mean the campaign is more profitable?

No. CPI is the cost of the install, not the revenue that follows. A cheap install that never pays can lose money against a more expensive install that purchases.

Hardal is the first-party data and agentic analytics platform.

We help enterprises send accurate signals to their ad platforms and grow revenue, using server-side data from their websites, mobile apps and AI agents.