The platform tax and the true cost of an install

The store commission is the largest single cost in most mobile game businesses and the one least likely to appear in a UA model. Finance books it as a cost of revenue, deducts it somewhere between gross bookings and net revenue on the spreadsheet, and it never sees the inside of a bid. That is a categorisation error with real consequences: the platform's cut behaves like an acquisition cost, and a UA team that prices it as one will bid differently and, in most cases, more accurately.

Practitioners will push back that any competent shop already calculates ROAS on net revenue.

Sometimes true. But netting revenue and pricing the commission into acquisition are not the same operation, and the difference shows up precisely where the money is: in the bid ceiling; in channel comparisons; in how a studio values the users it can route around the store.

Why a revenue-side cost behaves like an acquisition cost

The mechanism is simple enough. The platform levies its commission on every unit of value the acquired user creates, where a CPI comes due once and never again, which makes the commission proportional to lifetime value rather than to volume, and that is exactly the property determining what a marketer can afford to bid.

Take an illustrative user with 10.00 in expected gross in-app spend. On standard store terms in most markets, the studio keeps something in the region of 7.00 to 8.50 depending on the commission tier it qualifies for and how much of the revenue sits in the first year of a subscription. Suppose the team's model says "pay up to 4.00 for this user" on the basis of gross LTV and a 40 percent margin target. The true margin after commission is far thinner than it appears. Say the model instead reads "this user is worth 7.50 to us, and 2.50 of what they generate goes to the platform before we see it": now the 2.50 sits alongside the CPI as a cost of acquiring the revenue, and the affordable bid falls to a number the studio can actually defend.

The two framings produce the same net profit when every input is right. They diverge when the model is approximate, which is always. Gross-LTV models drift upwards as the team celebrates strong cohorts, while the commission never drifts. Parked beside the CPI, it stays visible.

The second-order effect: it changes channel comparisons

Where this starts to matter operationally is when acquired users do not all pay the same platform tax.

First, regional variation. Since 1 January, Apple's EU terms under the Digital Markets Act replace the per-install Core Technology Fee with a 5 percent Core Technology Commission on digital goods for developers on standard terms, a structure that shifts the effective rate for EU cohorts relative to elsewhere. A user acquired in Germany and a user acquired in Texas can carry different commission burdens for the same gross spend, and a UA model that nets revenue at a single blended rate will misprice one of them.

Then web stores and external purchase flows. A studio that moves a share of a cohort's spending to a direct web store keeps a larger fraction of each purchase. The obvious effect is a better margin. The less obvious effect is that the affordable bid for users likely to convert on the web sits above the bid for users who will only ever pay through the store, and the campaign machinery has no way of knowing which is which unless the studio tells it.

Then the platforms' own channels, where Apple Search Ads and Google's app campaigns are acquisition channels owned by the party collecting the commission.

For any user who converts inside the store, the platform earns both the CPI and the take. That is not an argument against using those channels, which are often excellent, but it is a reason to compare them with third-party networks on a fully loaded basis rather than on media CPI alone.

Where the accounting breaks down

Most coverage misses the trade-off. Pricing the commission into acquisition is easy for revenue the attribution stack can see, and web store revenue is exactly the revenue it cannot see well. A player acquired through a paid campaign, who installs from the store and later buys a bundle on the studio's website, generates revenue that reaches the studio at a lower commission but often never gets attributed back to the campaign that acquired them. Deterministic joins require the player to log in on both surfaces; probabilistic joins are the thing privacy frameworks exist to prevent.

The result is a systematic bias.

Cohorts that look least profitable in the attribution tool are often the most profitable on a net basis, because their highest-margin revenue is missing. Bids fall on precisely the users the studio should be paying more for. A studio that has invested in a web store and has not fixed this join is, in effect, still bidding as if every purchase paid full commission, and under-buying its own best customers.

A working definition and a decision rule

The true cost of an install, for the purposes of a bid, is the media cost plus the expected commission on that install's projected revenue, adjusted for the share of revenue the studio expects to route around the store. Written out for the illustrative user above: a 3.00 CPI on a user expected to generate 10.00 gross, of which 80 percent flows through the store at 30 percent and 20 percent through the web at a payment-processing cost of around 5 percent, carries a true cost near 5.50 and leaves 4.50 of contribution before other costs. Change the web share to zero and the contribution falls to 4.00. Change it to half and the contribution rises to nearer 5.10.

The decision rule is to maintain the commission burden as a segment attribute, not a global constant. Region moves the number, and so does platform; so does a cohort's observed or predicted propensity to buy off-store. Where the campaign platform allows value-based bidding, the values passed back should already be net of the platform's expected take, so that the algorithm optimises for the studio's money rather than gross bookings.

One practical starting point: run the studio's existing payback model twice, once on gross revenue, then once on net-of-commission revenue with regional rates applied, and compare the ranking of channels and geographies. Any channel or region that moves more than a couple of places between the two runs is one where the current bidding is wrong in a direction the team can now name.

Related archive reading

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

• What is mobile game user acquisition? A 2026 guide
• Retention is a UA metric