CPI vs CPA: how to choose the right pricing model for app campaigns

CPI vs CPA is one of the first decisions in any app campaign: do you pay for installs, or for users who do something useful after installing? The answer changes who carries the risk, how fast you can scale and how you should compare channels.

Below we explain both models, work through an example with simple numbers, and share the rules we use to decide.

What CPI and CPA mean

CPI (cost per install) is your spend divided by the number of attributed installs. If you spend $6,000 and get 10,000 installs, your CPI is $0.60.

CPA (cost per action) is your spend divided by the number of users who complete a chosen in-app event, such as registration, KYC, a first order or a first purchase. The event is agreed in advance and tracked by your mobile measurement partner.

You will also see CPL (cost per lead), CPC (cost per click) and CPM (cost per thousand impressions). They follow the same idea: the name tells you which unit you pay for.

CPI vs CPA: the real difference is who carries the risk

Under CPI, the advertiser carries the quality risk. If installs never open the app, you still pay for them. Under CPA, the publisher or ad network carries that risk, because they are only paid when users complete the event.

That is why CPA payouts are always higher than CPI payouts for the same traffic. You are paying the partner to take on uncertainty.

A worked example (illustrative numbers)

Two sources each deliver 10,000 installs at a $0.60 CPI, so each costs $6,000.

• Source A: 20% of installers register. That is 2,000 registrations, or $3.00 per registration.
• Source B: 6% of installers register. That is 600 registrations, or $10.00 per registration.

On a CPI report the two sources look identical. On cost per registration, Source B is more than three times as expensive. This is the core of the CPI vs CPA debate: installs are easy to buy, users are not.

When CPI makes sense

• Launch and ranking pushes, when you need volume quickly and have time to measure quality afterwards.
• Sources with proven retention, where months of data show installs turn into active users.
• Placements where events take time, such as some OEM preloads, where the first session may happen days after the install.

Even then, we add a retention floor so that CPI traffic below an agreed day-7 rate can be rejected.

When CPA makes sense

• New partners whose traffic quality you have not yet measured.
• Affiliate networks with long-tail publishers, where you cannot vet every source. Our guide to affiliate marketing for mobile apps covers this in detail.
• Apps with a clear value moment, such as KYC in fintech, a first order in commerce or tutorial completion in gaming.

Choosing the right event for a CPA deal

A good CPA event is:

• Close to revenue, so that paying for it is paying for value.
• Frequent enough to optimise, because ad platforms need a steady flow of conversions each day to learn.
• Completed within the attribution window, so it is credited to the right source.
• Tracked reliably, ideally confirmed server-side rather than only in the app.

If your true goal is a purchase but purchases are rare, pay for an earlier event that predicts purchase, such as adding a payment method.

Write the event definition into the insertion order in plain words, including how it is tracked and which attribution window applies. Vague definitions are the most common cause of payout disputes between advertisers and partners.

Hybrid models worth considering

• CPI with a retention floor: pay per install, with the right to reject sources below an agreed retention rate.
• CPI plus a CPA bonus: a lower install payout and a bonus for each user who completes the key event.
• Tiered payouts: higher rates unlocked when a partner beats your quality targets.

How to compare CPI vs CPA across channels

Put every channel on the same scorecard, whatever model you pay on:

• Cost per key event, for example cost per registration or per first order.
• Day-7 and day-30 retention for each source.
• Return on ad spend for each monthly cohort, once revenue data is available.

This stops cheap installs from looking like a win and lets you move budget to the sources that build your business. Fraudulent installs distort all of these numbers, so read our guide to spotting affiliate fraud as well.

Key takeaways

• In CPI vs CPA, the main difference is who carries the quality risk.
• Equal CPIs can hide very different costs per real user.
• Use CPA for new or unvetted sources, and CPI with retention floors for proven ones.
• Judge every channel on cost per key event and retention, not installs.

Want help setting payouts and targets for your app? Send Filly Media a short brief and we will reply with a channel and pricing plan.