CPA vs CPI: Conversion Cost or Install Cost
In short: Both are cost-per-action metrics, but they sit at very different depths of a mobile funnel. CPI prices the install, the cheapest and least meaningful event a mobile campaign can track, attributed by the app store or a measurement SDK. CPA prices whatever conversion the account is actually configured to count, which in a mobile context is usually a post-install event - a purchase, a subscription, a level completed - that requires real engagement past the download. A campaign can show an attractive CPI while its post-install CPA is terrible, because installing costs nothing to the user beyond a tap. Watch CPI for raw funnel volume; watch CPA, tied to a real post-install event, for whether the spend is actually working.
By the AdFlint research team · Fact-checked against current Google and Meta platform behavior · Last reviewed July 2026
CPA
Ad spend divided by the number of conversions, showing the average cost of each conversion the platform counted in that reporting window.
It answers what a conversion costs, but only for conversions the platform saw and claimed. Attribution windows, view-through counting, and duplicate events all move it while nothing in the account changes. The most common misreading is comparing the figure across platforms as though each measured the same thing; Google and Meta credit conversions under different rules and different lookback settings.
Full definitionCPI
Ad spend divided by app installs, the average cost of one install attributed to your advertising by the store or measurement partner.
It prices the download and nothing after it. Installs are the cheapest event in a mobile funnel, so campaigns optimized toward this metric reliably find people who install once and never open the app again. Attribution is also contested between networks and SDKs. The common misreading is judging campaigns on it when retention and post-install revenue say the opposite.
Full definitionSide by side.
The differences that actually change what happens in your account.
| CPA | CPI | |
|---|---|---|
| What it counts | Whatever event is configured as the conversion action - often a post-install purchase, subscription, or key action. | App installs, attributed by the store or a measurement SDK, nothing past the download. |
| Depth in the funnel | Typically deeper - requires the app to be opened and used, not just downloaded. | The very top of the mobile funnel - before the app has even been opened. |
| Dependent on attribution | Yes, on whatever conversion tracking setup the account uses. | Heavily - install credit is assigned by the store or SDK, and networks can disagree on who gets credit under last-click rules. |
| What a good number requires | The app is actually being used and the tracked event reflects real engagement. | Only that someone taps install, which requires no ongoing commitment at all. |
| Optimization risk when bid to it | Lower, since the bidder is chasing a meaningful downstream action. | Higher - optimizing purely to installs reliably finds people who install once and never open the app again. |
| Comparability across networks | Depends on whether the conversion event and attribution setup match across networks. | Weak - different attribution windows and last-click rules credit installs inconsistently between networks. |
| Failure mode | Comparing it across platforms as though each credits the conversion event under identical rules. | Judging a campaign as successful when retention and post-install revenue say the opposite. |
What actually separates them.
CPI is credited by an app store or SDK before your ad account even sees it, while CPA is calculated from whatever conversion event your own tracking setup defines, giving you more control over its meaning.
Optimizing a bid strategy toward CPI trains the algorithm to find the cheapest possible install, which tends to select for low-intent users, while optimizing toward a post-install CPA event trains it to find people who actually use the app.
CPI cannot tell you anything about what happens after the download, while a post-install CPA is specifically built to measure engagement or revenue past that point.
Cross-network CPI comparisons are unreliable due to differing attribution windows and credit rules, while CPA comparisons are only as reliable as whether the same conversion event and attribution setup are used on both sides.
A cheap CPI alongside an expensive post-install CPA is one of the more common patterns in mobile advertising and usually means the campaign is optimizing for the wrong funnel stage.
Which one should you use?
Use CPA when
- You have post-install tracking set up and need to know whether installs are turning into real usage or revenue.
- You are deciding whether to scale a mobile campaign and need a number tied to actual business value.
- You have enough post-install conversion volume for the figure to be statistically reliable.
- You are comparing mobile spend against non-mobile channels on the same downstream outcome.
Use CPI when
- You are early in a launch and need install volume before enough post-install data exists to bid on deeper events.
- You are comparing install campaigns within a single network where attribution rules are consistent.
- You are diagnosing whether a funnel problem sits before or after the install itself.
- You are budgeting a raw awareness-to-install push and will layer in post-install optimization later.
Common questions.
Should I bid to CPI or a post-install CPA event?
Bid to CPI only in the earliest stage of a launch when you lack enough post-install conversion volume for the algorithm to optimize against anything deeper. Move to a post-install CPA event as soon as volume supports it, since bidding to CPI alone reliably attracts low-intent installs that never open the app again.
Why is my CPI great but my post-install CPA terrible?
This is a common pattern when a campaign is optimized purely for installs. Installing costs the user almost nothing, so cheap CPI campaigns often find people with no real intent to use the app, and that shows up as a weak or nonexistent post-install conversion rate.
Can I compare CPI across two different ad networks?
Only cautiously. Each network's attribution SDK and lookback window credit installs by different rules, and last-click attribution can hand credit to whichever network's ad the person interacted with most recently, regardless of which one actually drove the install. Confirm both networks use comparable attribution settings before comparing the figures directly.
Or stop choosing between them.
AdFlint picks the setting, writes the ads, and keeps optimizing inside the Google and Meta accounts you already own.