Apple has quietly introduced another significant restriction on identity-based advertising. With iOS 27, WebKit began blocking domains associated with Unified ID 2.0, LiveRamp, ID5, Permutive, Audigent and others. Apple initially also blocked The Trade Desk’s ad-serving domain, a move it has partially reversed in the iOS 27.2 beta. The restrictions on identity-focused domains remain.
Much of the reaction has focused on what this means for ad tech companies and whether advertisers will move more money toward vertically integrated platforms that can still offer identity signals and rapid optimization.
I think that frames the problem backwards.
The Tail Is Wagging the Dog
Advertising technology lets marketers attribute conversions to individual exposures, monitor campaigns continuously, adjust bids, shift audiences and move money among tactics almost instantaneously.
But are we measuring the total effect of advertising on sales, or just the portion the technology can track? Are we measuring what matters to the business, or what is easiest for the measurement stack to observe?
User-level measurement necessarily gives more weight to exposures and outcomes that can be connected through tracking technology. Untracked impressions, offline sales, unmatched transactions, cross-device behavior and other unobservable activity can disappear from the measurement entirely. Models can estimate what happens outside that observable universe, but estimates don't turn a partial view into ground truth.
In our work running media ROI experiments, we often find that a trackable online channel shows little or no incremental advertising effect, while the advertising has a substantial impact on sales through an offline channel that the advertiser isn't measuring.
This is endemic to the ecosystem. It is a structural limitation of user-level advertising measurement.
Feedback Loopholes
The industry likes to talk about “closed-loop measurement.” But it isn't really a closed circuit. There are feedback loopholes everywhere: exposures and outcomes the technology cannot observe or connect.
This creates a fundamental problem with optimization. A system can become extremely efficient at optimizing toward the outcomes it observes while missing much of the actual business impact of the advertising.
That's Goodhart's Law applied to advertising: when a measure becomes a target, it ceases to be a good measure. Optimize hard enough against attributed conversions, clicks or another observable proxy and the system will get increasingly good at producing that metric. It doesn't establish that advertising is causing incremental sales.
Randomized experiments are much harder to game. The question isn't what happened among the people a platform can identify. It's what actually changed because advertising was present.
The Biggest Levers Give the Most Lift
Marketers have become obsessed with pulling increasingly granular levers: bids, audiences, placements, frequency, timing, creative variants and countless other variables.
They are starting at the wrong end of the problem.
The biggest lever is allocation across channels. Before worrying about whether one audience or bid strategy performs better than another, marketers should know whether they are putting the right amount of money into CTV, search, social, retail media and other channels in the first place.
I would prioritize the questions in this order:
Are we measuring the total sales impact of advertising, or being guided by what the tech stack can observe?
Are we measuring independently, or relying on media platforms and intermediaries whose incentives may differ from the advertiser’s?
Are we allocating the media budget correctly across channels?
Are we allocating correctly among vendors within each channel?
Are the tactics within each vendor actually effective, not merely efficient against the platform's own metrics?
Are the offer, creative, media pricing and other tactical elements contributing to incremental business results?
This is also where experiments and Marketing Mix Models should work together. MMM can help make allocation decisions across the portfolio. Randomized experiments can establish causal ground truth and calibrate the models. Get the biggest allocation decisions right first, then work down the hierarchy.
The Fast and the Frivolous
The case for user-level feedback loops often comes back to speed. Advertisers can get signals immediately and continuously optimize against them.
But what is actually at stake if the answer takes three weeks instead of three hours?
Just because I can weigh myself five times a day doesn't mean I'll lose weight faster.
Advertising budgets are managed against monthly, quarterly and annual business objectives. Well-constructed geo experiments can deliver high-confidence causal evidence in a matter of weeks. Decisions affecting millions of dollars in advertising investment and revenue are worth waiting three weeks for rather than day-trading on partial, modeled signals.
Marketing investment isn't day trading. The objective isn't to move money as frequently as possible. It's to allocate money correctly.
Is Apple Doing Us a Favor?
Every new restriction on identity infrastructure forces the industry to reconsider how much machinery is actually necessary to answer the fundamental business question.
Device graphs, clean rooms and identity systems have legitimate uses. They can help advertisers manage reach and frequency, coordinate targeting across devices, combine data held by different organizations and analyze observable consumer behavior. But the consumers whose exposures and purchases can be matched are not necessarily representative of the broader audience. Even a large matched sample can produce misleading results if it systematically underrepresents certain groups, such as iOS users. And if we can measure incremental advertising ROI without identifying and matching individual consumers, perhaps the answer isn't always to rebuild what Apple just broke.
Randomized geo experiments offer a more future-proof measurement architecture. They don't require an identity graph or individual exposures to be matched to individual transactions. They can measure online and offline sales, tracked and untracked customers, and the total business outcome the advertiser actually cares about.
And the method is universal across media. The same experimental framework can move from channels to vendors, tactics, audiences, creative, offers and other marketing decisions. So long as media delivery can be varied geographically, the method survives changes to identifiers, browsers, operating systems and the measurement rules of individual platforms.
Platform data still has a role. It can tell advertisers how different creatives, offers, audiences and delivery tactics perform among the users and outcomes a platform can observe. That's useful operational information.
But it isn't the same as measuring the incremental effect of advertising on total sales.
The first question should remain the simplest and most important: Did the advertising cause additional sales?
Get that right first. Then start pulling the levers.