Every platform privacy change is a pricing change

UA teams file privacy changes under compliance. The legal team reads the developer terms, the SDK gets updated, the consent dialogue gets a new line, and someone logs the work as done. That filing is the mistake. A privacy change is a pricing change, and the person who owns the media budget should read it before anyone else does.

The argument is simple. Signal doesn't leave the market when a platform restricts it; it moves. The device owner keeps seeing what the open buyer no longer can, so do the SDK owner and the walled garden, and an auction with uneven information always prices that unevenness. Five years after ATT, that is still the underrated part of the story.

What an auction does with missing information

A real-time bid is a prediction dressed as a price. The bidder estimates the value of the user in front of it and subtracts its margin, then bids. Take away the identifier that made the prediction possible and the honest bidder widens its uncertainty and bids lower. The bidder that still has a good view of the user does not.

The result is not a cheaper market. It is a market where the informed bidder wins more auctions at a price set by the uninformed bidders' caution, and where the uninformed bidders drift toward the informed bidder's own products to get their signal back.

None of this requires bad faith. A platform that can see more will, quite reasonably, build products around what it sees and charge for them. The point is that the buyer's cost of restoring signal is a real cost of media, and it never appears on the invoice as one.

Adjust's March report on gaming app sessions put ATT opt-in at 39 percent in the first quarter of 2026, against 38 percent a year earlier. That number is often read as a stable nuisance. It is better read as a stable tax rate. Roughly three in five iOS users are visible only to whoever has first-party context, and that visibility is now the product being sold.

How the toll booth is priced

The mechanism deserves more precision than "platforms win". The pricing of the replacement product is where the money actually moves.

Before ATT, the signal a buyer needed to value an iOS user was free at the point of use. The identifier came with the impression, and the buyer's own models or its MMP did the valuation. The marginal cost of knowing who converted was close to zero. After ATT, the same knowledge exists, but it exists inside the platform's own modelled conversion product and on-platform attribution, or inside aggregated event measurement that only the platform can compute because only the platform sees both the impression and the first-party context on the other side.

The platform doesn't price that product as a line item. It prices it as a share of performance. The platform's bidder uses the modelled signal to bid more accurately and win the auction, then keeps the difference between what it predicted the user was worth and what the uninformed bidders would pay. The buyer pays for the restored signal through a higher clearing price, through the platform's margin on the modelled prediction it cannot audit, and through the reporting it must now accept on trust. Something that used to be free and inspectable is now bundled and opaque, and it's charged as a percentage of media.

There is a second toll. Buyers who want to keep an independent view pay for it again: probabilistic modelling in the MMP and incrementality testing, plus web-to-app flows that recover consented signal. Those are real budget lines, and they exist only because the free signal went away. Add both tolls together and the effective CPM on iOS is materially higher than the one on the invoice, before anyone has tested a single creative.

Two changes, one pattern

Look at the changes that landed over the last year through this lens and the pattern is hard to miss.

Google's Privacy Sandbox retirement for Chrome, which began with the Chrome 144 deprecation, was a privacy change that ran in reverse. Cookies stayed. The pricing effect still happened, because the buyer investment in Sandbox-shaped measurement was written off, and the large platforms that never needed the Sandbox lost nothing. The signal did not move that time; the cost of preparing for it to move did, and it moved onto the open buyer alone.

Unity's ironSource network shutdown on April 30 was a business decision, but the stated rationale was to pour everything into Vector, an AI model that works best where Unity's own runtime sees the most. That is signal consolidation by another name. It means one fewer place to buy where the seller does not also own the observation layer, and one more place where the buyer's only view of outcomes is the seller's model.

The second-order effect nobody budgets

The obvious cost of a privacy change is measurement noise. The cost that shows up a quarter later is concentration.

Each time signal narrows, mid-tier networks lose the ability to price users accurately; their margins compress and they either exit or consolidate, or they become resellers of the big platforms' modelled audiences. Fewer sellers means fewer competing bids on the supply side and less price discipline on the demand side. Sensor Tower's State of Mobile 2026 described a strategic shift from new-user volume toward lifetime-value expansion; part of what drives that shift is that the marginal install has quietly become more expensive to buy well.

There is a creative consequence too. When the algorithm cannot see the user, the ad has to do more of the sorting, which is why creative volume has become a proxy for targeting. This site argued in March that privacy loss changed the job of the creative strategist. The same shift changed the job of the buyer: creative production is now a line in the media cost, not a separate budget.

A ledger to run on every platform notice

When the next developer bulletin arrives, ask three questions before anyone touches the SDK. Who loses signal? Usually the open buyer and the MMP. Who keeps it? Usually the device owner, the network with the SDK in the game, and the walled garden. What does the price of the kept signal do? It rises, and the rise appears as a higher effective CPM, a wider spread between modelled and observed ROAS, or a new paid product that restores what was free.

Then translate the answers into a budget move rather than a compliance ticket. As an illustrative example only: if a change looks likely to reduce attributable installs on a channel by a fifth, and the channel's automated bidder holds its ROAS target, the bidder will cut spend rather than accept a worse number. The buyer who doesn't pre-emptively adjust targets will read that as "the channel stopped scaling" when what actually happened is that the channel's price went up and the algorithm declined to pay it.

The decision rule that follows: treat any platform privacy notice as a trigger to re-underwrite channel ROAS targets and the modelled-versus-observed gap within the same sprint, not after the next monthly review. The platform has already repriced. The only question is whether the buyer notices before the automated bidder does, and the automated bidder is not going to wait.

Related archive reading

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

• Every SDK you install is a data-sharing agreement
• The platform tax and the true cost of an install