Ask five agencies how they price a campaign and you’ll get five different acronyms back — CPI, CPA, CPE, CPS, ROAS — often without much explanation of what’s actually being optimized underneath. That’s a problem, because the pricing model isn’t just a billing detail. It determines what the buying algorithm is trying to maximize, which means it quietly shapes the quality of every user you acquire.
Here’s what each model actually means, and where it fits.
CPI — Cost Per Install
The oldest and simplest model: you pay a fixed price for each install, full stop. It’s easy to forecast and easy to compare across sources, which is exactly why it became the default years ago.
Its weakness is also its simplicity — CPI says nothing about what happens after the install. A source optimizing purely for CPI has every incentive to find the cheapest install it can, and “cheapest” doesn’t always mean “most likely to become a customer.” This is why CPI campaigns are the ones most exposed to low-intent or fraudulent installs unless fraud filtering is doing real work underneath the pricing model, not layered on as an afterthought.
Best for: early-stage volume tests, top-of-funnel awareness, or apps where install itself is close to the whole conversion event (some utility and free-to-play titles).
CPA — Cost Per Action
You pay when a user completes a specific in-app action — a signup, a level completed, a form submitted. This pushes optimization one step past the install, toward a signal that at least correlates with genuine engagement.
CPA is a meaningful upgrade over CPI for most verticals because it forces the buying side to care about what happens after the click, not just the click itself. The tradeoff is that CPA campaigns need a clean, unambiguous event to optimize toward — a vague or infrequent action makes it hard for the algorithm (or the humans running it) to learn what “good” looks like.
Best for: apps with a clear early-funnel action — account creation, first booking, first deposit — that reliably precedes real usage.
CPE — Cost Per Engagement
A variant of CPA scoped specifically to engagement milestones rather than a single defined action: session depth, time in app, feature usage. It’s common in gaming and entertainment, where “did they play” matters more than any single discrete event.
Best for: gaming, entertainment, and content apps where retention and session behavior are the real product signal.
CPS — Cost Per Sale
You pay only when a transaction closes — a purchase, a subscription, a completed booking. This is the tightest link between spend and revenue you can price on, and it shifts nearly all of the acquisition risk onto whoever is buying the traffic.
That’s a real advantage for the advertiser, but it comes with a practical constraint: CPS only works cleanly when the purchase event is well-instrumented and happens close enough to install that attribution stays reliable. E-commerce, subscriptions, and travel bookings fit this well. A game with a 45-day median time-to-first-purchase is a harder fit.
Best for: e-commerce, subscriptions, travel, and any funnel where the sale happens within a trackable window of the install.
ROAS — Return on Ad Spend
Rather than pricing a single event, ROAS campaigns are managed against a target return — revenue generated per dollar spent, tracked over a defined window. This is less a pricing unit than an optimization target that spending decisions get held to, campaign by campaign, source by source.
ROAS is the model most aligned with how a finance team actually thinks about acquisition spend, which is why it’s increasingly the default for mature apps with enough transaction volume to make the math statistically meaningful. It needs good data to work: reliable revenue events, a sensible attribution window, and enough scale that the numbers aren’t noise.
Best for: apps with steady transaction volume and clean revenue tracking, where the real question isn’t “how many installs” but “did this spend pay for itself.”
Choosing between them
The honest answer is that the right model depends on how far down your funnel you can reliably measure, not on which acronym sounds most sophisticated.
- If you can’t yet measure past install, start with CPI — but insist on fraud filtering that’s built in, not billed as an add-on.
- If you have one clear early action, CPA will align spend with intent better than CPI ever will.
- If retention and depth of use are the real product signal, CPE fits gaming and content apps better than a single event.
- If purchases happen fast and are well-tracked, CPS puts the acquisition risk where it belongs.
- If you have transaction volume and clean revenue data, ROAS is the model that actually answers the question your finance team is asking.
It’s also worth noting that the pricing model shouldn’t be fixed for the life of an account. A campaign often starts on CPI to find volume, graduates to CPA once an early signal is identified, and matures into ROAS once there’s enough transaction data to optimize against a return target directly. What matters is that whoever is buying your traffic can move with you through that progression — and that the KPI they’re optimizing for is the one you actually care about, not the one that’s easiest for them to hit.
At RVM Ads, the pricing model is set by the outcome you name, not the other way around — CPI, CPA, CPE, CPS, or a ROAS target, chosen for where your funnel actually is today and revisited as it matures.
