CPI vs CPA vs CPM: Choosing a Pricing Model for Mobile Campaigns
Every mobile campaign starts with a commercial question disguised as a technical one: what exactly are you paying for? CPM, CPI and CPA are not interchangeable labels for the same thing — each moves risk, margin and control between you and the traffic source. This guide walks through what each model really buys, where each one breaks, and how we match models to offers in practice.
The three models in one minute
CPM — cost per mille: you pay for every thousand ad impressions, install or no install. CPI — cost per install: you pay when a user installs your app. CPA — cost per action: you pay when the user completes a defined post-install event — a registration, first deposit, purchase or subscription. Its close cousin CPL (cost per lead) pays for lead-form completions rather than in-app events.
The pattern is a staircase: the further down the funnel the billable event sits, the more risk the traffic source carries, and the more each unit costs. Nobody sells cheap CPA at scale for a difficult event — if they do, you are about to learn about fraud the expensive way.
CPM: buying attention
CPM is the rawest model: you buy eyeballs and own 100% of the performance risk. It shines when your creative and funnel are already proven — then impressions convert at a known rate and CPM is usually the cheapest way to buy volume. It is also the model of brand-style pushes and of retargeting pools.
CPM breaks when the funnel is unproven: you can burn a test budget on impressions that never had a chance. Our rule of thumb for advertisers entering a new GEO or vertical — for example a first crypto acquisition test — is to avoid pure CPM until click-to-install and install-to-event rates are measured on paid models first.
CPI: buying installs
CPI moves the billable event one step down: the network eats impression risk, you eat everything after the install. It is the workhorse of mobile user acquisition, especially for mobile games and utility apps where install volume is the primary currency.
The trap is that an install is not a user. Cheap CPI inventory can be stuffed with incentivized installs, emulators or click-injected attributions — installs that exist only on the attribution report. CPI only works with two guardrails: MMP validation on every install, and source-level scoring on post-install behavior so low-quality placements get cut, not averaged into the CPI you tolerate.
CPA and CPL: buying outcomes
CPA moves risk furthest onto the source: payment only when the defined action happens — a funded account, a deposit, a subscription. For forex brokers or casino operators, buying on first-time-deposit CPA aligns incentives perfectly: the network only earns when your finance team sees money.
The costs are real, though. CPA payouts are the highest per unit, volume is capped by what sources can profitably deliver, and approval logic matters — who validates the event, on what data, with what delay. Define the billable event with brutal precision before launch: "registration" that includes abandoned half-filled forms and "registration" that requires verified contact details are two different products at two different prices.
Hybrid and revenue share
Real partnerships rarely live at the extremes. A hybrid deal combines a reduced fixed CPA with a share of the revenue the acquired users generate — the source takes less cash upfront and participates in the upside. Revenue share removes the fixed part entirely: the source earns a percentage of player or subscriber revenue over time.
These models fit subscription and long-LTV products where value arrives over months. They demand trust and infrastructure: the source needs read access to revenue events, and both sides need agreed attribution windows. We run hybrid structures in several of our covered verticals once a campaign has proven stable economics on a simpler model — hybrid is a graduation, not a starting point.
How to choose: a working decision frame
Strip the debate to three questions. What event actually pays you? If it is a deposit, negotiate toward CPA or hybrid and treat CPI as a testing tool. How proven is the funnel? Unproven funnels belong on outcome models or small CPI tests; proven funnels unlock CPM scale. Who can measure better? If your MMP setup tracks post-install events reliably and the source cannot, you have an information edge on CPI; if the source sees quality signals you cannot, outcome models protect you.
Whatever you choose, write the billable event, validation source and approval window into the insertion order. And if you are scaling spend, pair the pricing decision with a structured scaling process — the right model with chaotic scaling still loses money. Ready to price your offer? Brief our team and we will come back with a model recommendation and target economics.