Cookies, consent and iOS: what measurement still works

Over the past few years, the ability to follow individual people across websites and apps has steadily shrunk. Consent banners mean many visitors decline tracking. Browsers limit how long some cookies last. Apps must ask permission before tracking users across other apps, and many people say no.

The result is a common complaint: our measurement is broken. Some of it is. Not all of it.

What depends on following individuals

Attribution works by stitching together each person's journey (this ad, that click, this purchase) and assigning credit along the path. Every gap in tracking is a gap in the path. Platforms fill some gaps with modelled conversions, which are estimates, not observations.

So attribution has become both less complete and less verifiable, and it was already measuring credit rather than cause.

What doesn't

• Mix models use aggregate data: total weekly spend by channel, total weekly sales, plus price, promotions and seasonality. No individual tracking is needed. Consent rates don't change how much you spent or how much you sold.
• Geo experiments compare regions, not people. They need regional sales and controlled spend, both of which you already have.

These methods were designed for a world without user-level data. That world has partly returned.

What to do now

1. Shift budget decisions onto aggregate methods. Use mix models and geo tests for the questions that move money between channels.
2. Keep attribution for in-channel tactics, where its gaps matter less and speed matters more.
3. Fix what you can in tracking. Good consent implementation, cleaner event design and first-party data collection still matter for operations and for the data your models use.
4. Treat modelled conversions as estimates. Useful, but not evidence of cause.

The reframe

Privacy changes didn't break marketing measurement. They broke the part that depended on following individuals. Teams that move their big decisions onto methods built on aggregate evidence often end up with better answers than they had before, not worse.