Blog

CPI vs CPA vs CPM for App Advertisers

Lakshith Dinesh

Lakshith Dinesh

Head of Growth, Linkrunner

Last updated on 7 min read

CPI vs CPA vs CPM for App Advertisers

Short answer: CPM charges per thousand impressions, CPC per click, CPI per install and CPA per in-app action such as a sign-up or purchase. On Meta you pay per impression and on Apple Ads Advanced per tap, and on Meta and Google you set a target CPI, CPA or ROAS that the algorithm aims for on average; true pay-per-install or pay-per-action prices usually come from affiliates, ad networks and Apple Ads Basic. The further down the funnel you pay, the more risk shifts to the seller, and the more you need independent attribution to check what you were charged for.

Most app teams quote "our CPI" as if it were a price. On Meta, Google and most of Apple Ads it is not. It is an outcome: spend divided by installs, worked out after the auction has already charged you for impressions or taps. That decides how you compare networks, negotiate with affiliates and spot fraud.

What do CPM, CPC, CPI and CPA mean for an app?

Each model names the unit you pay for: an impression (CPM), a click or tap (CPC), an install (CPI) or a post-install action (CPA). The further down the funnel the unit sits, the more conversion risk sits with the seller rather than with you.

  • CPM (cost per mille): the price of 1,000 impressions. If nobody clicks, you still pay.
  • CPC or CPT: you pay when someone clicks or taps. Apple calls it cost per tap.
  • CPI (cost per install): you pay for an install, usually counted as the first app open after attribution, not the store download.
  • CPA (cost per action): you pay for an event after install: registration, KYC complete, first deposit, first purchase, trial start.
  • Value bidding (tROAS): not a pricing unit. You tell the network the return on ad spend (ROAS) you want, and it chases users likely to spend.

Bid target vs billing unit: the bid target is what the algorithm optimises towards (installs, actions, value). The billing unit is what appears on the invoice (impressions, taps, installs). On the big platforms they are usually different.

How do Meta, Google and Apple Ads actually charge you?

The large self-serve platforms let you optimise for installs, actions or value, but the number you set is a target average, not a fixed price. Meta bills on impressions and Apple Ads Advanced on taps.

Meta

  • Meta's developer guide to app ads optimisation says app event optimisation is "charged on an impression basis, similar to how you are charged when you optimize for installs". Value optimisation is billed on impressions too.
  • You pick what to optimise for (installs, one app event, or value) and a bid strategy such as highest volume, a cost per result goal, a ROAS goal or a bid cap.
  • So on Meta, CPI and CPA are results you steer towards; impressions are what you pay for.

Google App campaigns

  • Google's App campaign bid strategy page lists target cost per install (tCPI) and target CPA for install campaigns, target ROAS for in-app action value, and maximise conversions options, some still in beta.
  • Google's bidding guide calls the bid "the average amount you'd like to spend each time someone installs your app", and notes that auction swings can push a single day above or below target.
  • Google advises choosing one in-app action for tCPA. Which event you choose matters more than the number, as covered in why smart bidding optimises the wrong users.

Apple Ads

  • Apple Ads Basic is a genuine CPI product. Apple's Basic page says "Pay only for installs at a price you set", with ads in search results.
  • Apple Ads Advanced charges per tap. Apple's Advanced page has you set "the maximum amount you're willing to pay for a tap" across the Today tab, Search tab, search results and product pages, and the invoice guide shows a CPT line for each.

Affiliates and ad networks

  • This is where fixed CPI and CPA deals live. The partner agrees a rupee price per install or action and is paid on the conversions your MMP reports back.
  • It looks safe because you only pay for outcomes, but the seller now has a reason to manufacture them.

How do you convert between CPM, CPC, CPI and CPA?

Each model links to the next through one conversion rate: CPC = CPM ÷ (1,000 × CTR), CPI = CPC ÷ click-to-install rate, and CPA = CPI ÷ install-to-action rate.

An illustrative funnel. The inputs are invented to show the maths; they are not benchmarks.

StepIllustrative inputCost per unit
ImpressionsCPM of ₹150₹150 per 1,000
Clicks1% CTR, so 10 clicks per 1,000₹15 per click
Installs30% of clicks install, so 3₹50 per install
Sign-ups40% of installs sign up₹125 per sign-up
First purchase5% of installs buy₹1,000 per purchase

What the maths shows:

  • A cheaper CPM does not guarantee a cheaper CPI. Halve the CTR and the same ₹150 CPM gives a ₹100 CPI.
  • CPI hides activation. Two campaigns at ₹50 CPI with 40% and 20% sign-up rates cost ₹125 and ₹250 per sign-up. That is why activation beats installs.
  • Work offers backwards. A ₹40 CPI affiliate that converts 25% of installs to sign-up really costs ₹160 per sign-up. Compare that with your cost per sign-up on Meta or Google, not with their CPI.

For market ranges, see our India CPI benchmarks by app category, built on Linkrunner data, and the free CPI benchmark explorer, which compiles third-party figures by region, platform and category. Treat both as reference ranges, not targets.

Which pricing model suits which situation?

ModelYou pay forConversion risk sits withSuits an app whenMain risk to you
CPMImpressionsYouMeta or Google at scale, launches, creative testsPaying for views that never install
CPC / CPTClicks or tapsMostly youApple Ads Advanced, search intent, web-to-appBot or accidental clicks, weak store page
CPIInstallsSeller, for the installApple Ads Basic, affiliate and network dealsFake or incentivised installs that never activate
CPAA post-install eventSeller, for the eventAffiliates on a known event value; tCPA on Google, cost goals on MetaFaked events, events unlinked to revenue
tROASUsually impressions, optimised for valueSharedSteady purchase volume reaching the networkSmall samples, late or dirty revenue data

A common progression: optimise for installs while event data is thin, move to tCPA or an app event once sign-ups or purchases are steady, and test value bidding when revenue reaches the networks cleanly. With affiliates, prefer CPA on an event that is hard to fake (KYC complete, first deposit, first purchase).

What risk does each model shift onto you?

Pay-per-outcome models move cost risk to the seller but create an incentive to fake the outcome. CPM removes that incentive but leaves all the performance risk with you.

  • CPM and CPC: you pay whether or not anything happens; the defence is creative and conversion-rate work.
  • CPI: the classic target for affiliate fraud: click spamming to claim organic installs, click injection, device farms, or paid users who install and leave.
  • CPA: harder to fake, but SDK spoofing can fake events, and cheap events (an email sign-up) are easy to inflate.
  • tROAS: data quality is the bigger risk. If refunds or test purchases reach the network, it optimises for the wrong users.

A field example: a CPI that was not what it looked like

On one fintech app's affiliate traffic between August and November 2026, about three quarters of installs from one affiliate opened within 10 seconds of the click, far faster than a real store download, with heavy IP repetition. Around 16k of roughly 200k installs were flagged by a time-to-install rule.

The twist: about half the flagged installs still went on to sign up. That looks like paid-to-install users rather than pure fake installs, so the traffic remains flagged and under review, not confirmed fraud. On a CPI deal every one of those installs costs the same. On a CPA deal priced on a deeper event, the low-intent ones mostly fall out of what you pay for. How mobile install fraud is detected covers the timing and IP signals.

Why do you need attribution to compare networks fairly?

You can only compare a ₹50 CPI on Meta, a ₹15 tap on Apple Ads and a ₹120 CPA affiliate deal if one referee counts the installs and events behind each figure, with the same windows and definitions.

  • Each network grades its own homework. Meta, Google and Apple report their own installs with their own windows. Added together, they usually exceed your real installs.
  • Affiliates bill on what you report. On CPI and CPA deals, your MMP's postbacks are effectively the invoice, so they must be deduplicated and fraud-checked first.
  • Windows move the numbers. Linkrunner's default click window is 7 days for most networks and 30 days for Meta and Google, adjustable from 1 to 30 days per channel. Align windows before comparing, as why Meta ROAS and MMP ROAS don't match explains.
  • Networks optimise on what you send. tCPA and value bidding need sign-ups or revenue sent back, not just installs. See the postback setup guide.

How to check this in your MMP

  1. Pull spend by network and campaign for one date range.
  2. Pull attributed installs and your key event from the MMP, not from each network.
  3. Calculate CPI, cost per event and cohort ROAS for every channel on that basis.
  4. Re-price affiliate deals on cost per real event, after flagged installs are removed.
  5. Confirm CPI and CPA partners are paid only on installs your fraud rules did not flag.

A mobile attribution platform like Linkrunner shows this in one dashboard, from click to install, sign-up, revenue and ROAS, across Meta, Google, TikTok, Apple Search Ads and affiliates. Fraud Control has 8 rule templates (click injection, install hijacking, SDK spoofing, SDK spoofing with IP mismatch, click flooding by IP, click flooding by device, installs from fraudulent clicks, reinstalls from fraudulent clicks), and each can tag installs or tag them and withhold postbacks, so a CPI partner is not paid on a flagged install. Postbacks send revenue signals, not only installs, to Meta, Google and TikTok. The first 25k attributed installs are free, one time, then ₹1.20 down to ₹0.80 per attributed install; most teams integrate the same day and are fully live within a week.

FAQ

Is CPI the same as CPA?

CPI is a CPA where the action is the install. In practice "CPA" usually means an event after install, such as a sign-up or purchase, so it runs higher than CPI for the same campaign.

Does Meta charge per install?

Meta's developer documentation says app event optimisation is charged on impressions, the same as optimising for installs. You can set a cost per result goal, but your CPI is spend divided by installs.

Does Google charge a fixed price per install?

No. Google describes target CPI as the average you would like to spend per install. Single days can land above or below it while the campaign aims for the target over time.

Should affiliate deals be CPI or CPA?

CPA on a meaningful event is usually safer, because it is harder to fake and closer to revenue. Either way, pay only on installs and events your MMP counted after fraud checks.

The takeaway

CPI, CPA and CPM are not rival prices for the same thing. They are points on one funnel, linked by conversion rates you can measure. Buy on whatever unit each network sells, judge every channel on cost per real event and cohort ROAS, and let one neutral source count the installs and events. To see that view on your own data, request a demo from Linkrunner, then run the five-step check above on last month's spend.

Start measuring the installs your team cares about

Bring attribution, deep links, SKAN, cohorts, and campaign intelligence into one workflow your growth team can trust.