Every campaign is priced on a model, and the model decides one thing: at what moment do you owe money? Move that moment later and you buy something more valuable; move it later than you can measure and you cannot run the campaign at all.
The models
CPI — cost per install. You pay when the app is installed and opened. Simple, fast to attribute, and the standard for mobile user acquisition. The risk is entirely yours: an install is not a player.
CPA — cost per action. You pay when a defined action happens, and you define it: a registration, a completed profile, a first purchase. The most common model on a rewarded wall, and the one worth defaulting to.
CPE — cost per engagement. You pay when the user does something that proves they engaged — reached level 5, played for twenty minutes, completed a tutorial. The most expensive per unit and usually the cheapest per retained user, because the requirement filters out everyone who was never going to stay.
CPL — cost per lead. You pay for a qualified contact. Works when your funnel continues off-platform.
CPS — cost per sale. You pay a fixed amount, or a share, when a purchase happens. The lowest risk and the hardest to fill: publishers earn least often on it.
SOI and DOI — single and double opt-in. A sign-up, or a sign-up with a confirmed email. DOI costs more and removes most of the junk.
CPC and CPM. Cost per click and per thousand impressions. Rarely right on a rewarded wall, where a click is cheap to manufacture and the whole point is the completion.
Choosing
Work backwards from the first moment you can measure something you actually want.
If you know a user who reaches level 5 has a decent chance of still being there in a month, buy level 5 and pay properly for it. If you have no idea what your retention looks like yet, buy installs, measure, and move the goalpost once you know where it should be.
The trap is buying an early event because it is cheap. Ten thousand installs at a low price that never open the app again is more expensive than a thousand engagements, and it takes a month to find out.
Ladders
You do not have to pick one moment. A multi-step campaign pays a small amount for the install, more for the tutorial, more again for level 5 — so the publisher earns something early, which keeps the offer visible on walls, while most of your budget still goes to the users who did the thing you cared about.
It is the best structure for anything with a funnel, and it is what the goal ladder in a campaign is for.
What rewarded traffic changes
One thing, and it is important: the user was paid to do this. They wanted the currency, and your app was the route to it.
That is not fraud and it is not worthless — see the separate piece on that — but it does mean early events convert far better than they do on other channels, and later events convert worse. A CPI on rewarded traffic looks spectacular and predicts nothing. A CPE at day seven tells you the truth.
Price accordingly, and measure at the depth you actually care about.
The number that decides whether it runs
Whatever model you pick, the payout has to clear the whole chain: the publisher's cut has to be worth showing your offer, and the reward has to be worth a user's time. Price an install at four cents and it will technically be live and nobody will ever see it, because every wall it sits on has better things to show.
If a campaign is not delivering, the payout is the first thing to look at and usually the answer.