The auction, not the audience, sets your CPI

Most UA teams talk about CPI as if it belonged to the audience. Puzzle players cost this much, mid-core players cost that much, Tier 1 is expensive, LatAm is cheap. The language is convenient and it is mostly wrong. CPI is a clearing price. It is set by how many other buyers want the same impression at the same moment, and by how much they are willing to pay for it. The audience matters only insofar as it changes who else is in the room.

That distinction sounds academic until January, when it becomes the whole job. The first fortnight of the year is the clearest natural experiment the industry gets. Nothing about players changes on New Year's Day. Retail and e-commerce budgets, which spent December bidding for the same rewarded slots and social feeds that games rely on, disappear almost overnight. Impressions that cleared at a premium in the last week of December clear for much less in the first week of January. Same users, same creative, different room.

What an install actually costs

It helps to write the price out. CPI is CPM divided by the product of click-through rate and install rate, with the units tidied up. Two of those three terms belong to you. Click-through rate is creative and placement. Install rate is store page, creative honesty and the friction between tap and open. The CPM term belongs to the auction.

When a buyer says their audience is expensive, they usually mean CPM is high. High CPM does not tell you the users are valuable. It tells you that somebody else has decided they are worth paying for, and that somebody may be a mattress brand, a betting app or a casual publisher with a lower payback threshold than yours. You inherit their valuation whether or not it matches your economics.

The reverse is also true, and less comfortable. A cheap audience is cheap because the marginal bidder for it has gone away or never arrived. Sometimes that is an opportunity. Often it is information: the buyers with the best data on that inventory have decided it does not pay back, and the price you are celebrating is what is left when informed money leaves.

Narrow targeting raises the price of the room

The second-order effect most coverage misses is that targeting is a bid on competition density, not just on user quality. Every refinement you add to an audience definition also describes the audience other buyers in your category are refining towards. Lookalikes built from payers, interest clusters around specific genres, high-end device filters: these do not just find your best users, they find the same users everybody else's model found.

The result is a crowding effect. The narrower and more obviously valuable the segment, the more bidders it attracts and the more the clearing price rises relative to the underlying user value. Broad targeting hands the selection problem to the platform's model, which is imperfect, but it also spreads your bids across impressions where the room is emptier. Whether that trade is worth making depends on how good the platform's prediction is for your event, not on how good your audience definition sounds in a planning deck.

There is a further consequence for portfolio decisions. Two games from the same studio with similar players are not competing for the same users in any meaningful sense. They are competing for the same impressions, and the studio's own bids are part of the density. Running them in the same auction with independent bid targets means the studio pays a premium to itself.

A decision rule for reading a CPI move

When CPI moves, the question is whether the auction moved or you did. A useful discipline is to refuse to discuss CPI until the decomposition is on the table.

• If CPM moved and click and install rates held, the room changed. Creative and store work will not fix it and may waste a sprint.
• If CPM held and click rate fell, the creative is tired or the placement mix shifted. This is a creative problem wearing a buying costume.
• If CPM and click rate held and install rate fell, look at the store page, the creative-to-store mismatch or a new app version.
• If everything moved at once, somebody changed the campaign structure and the rest is noise until the learning phase settles.

An illustrative example makes the point. Suppose a casual title runs at a CPM of 12, a click rate of 1.5 percent and an install rate of 20 percent, for a CPI of 4. In the first week of January the CPM falls to 9 with no other change, and the CPI drops to 3. A team that attributes this to its new creative batch will double down on the wrong lesson and be confused in February when the CPM recovers. A team that reads it as an auction move will use the window to buy volume it could not afford in December, and will set the February forecast on the old CPM, not the new one.

Buying the room rather than the user

Once CPI is understood as a price, the practical levers change. The audience is fixed by the game. The room is not.

Time of day and day of week are room decisions. Placement type is a room decision; rewarded video and interstitial inventory clear differently because different categories of buyer prioritise them. Geography is partly a user decision and partly a room decision: a market where a large publisher has just launched a competitor is expensive regardless of player quality, and a market they have just pulled out of is briefly cheap.

Bid strategy is the biggest room decision of all, and the one buyers have least visibility into now that most spend runs through automated bidding. A target-cost or target-return setting is not a bid; it is an instruction to a platform that then bids on your behalf against everyone else's instructions. The clearing price you see is the aggregate of those instructions and the platform's own margin.

This is why January matters beyond the cheap installs. It is the one month where the auction term moves so much, and so predictably, that a team can measure how much of its CPI is room and how much is craft. Run the decomposition weekly through the month. The proportion you get will be the most honest baseline you have for the rest of the year, and it will tell you where a creative sprint can actually earn its cost and where it is fighting the auction.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• Alternative app stores will not lower your CPI
• The CPI obsession is a category error