The dramatic predictions about life after third-party cookies have been overtaken by a more mundane reality. Two years past the major deprecation milestones, attribution still works, just differently than it did before, and with an honestly acknowledged loss of precision that the industry has mostly come to terms with.

The current state of practice for most mid-sized advertisers in 2026 looks like this: a primary identity layer based on first-party email and authenticated user signals, supplemented by Google’s Privacy Sandbox attribution APIs where they integrate with the buying tool, supplemented by media mix modelling for the unaddressable layer, supplemented by strategic incrementality testing on key channels.

What works well: media mix modelling has had a renaissance and the tooling is genuinely better than it was. Marketers who built MMM capability over the past two years are getting useful answers about channel-level effectiveness. The output is less granular than the click-based attribution it partially replaces, but it’s more honest about the actual range of uncertainty.

What works poorly: cross-device attribution for users who haven’t authenticated, view-through attribution beyond very short windows, and any attempt to measure the long-tail of mid-funnel touchpoints with the precision the old MTA models claimed.

The data clean room ecosystem has continued to grow. Several Australian retailers and media owners are now running clean-room collaborations that allow attribution and audience analysis without underlying data exchange. The technology works. The economics still favour the largest data holders, which entrenches their position.

Identity solutions outside Google’s ecosystem (UID2, RampID, and similar) have had mixed adoption. They work technically. The publisher willingness to deploy them has been uneven. The browser-side resistance from Apple in particular has remained a structural ceiling.

For marketers thinking about their 2026 measurement stack, the honest priorities are clearer than they were. Invest in MMM. Invest in clean-room infrastructure if your spend justifies it. Run incrementality tests on big channels. Don’t over-invest in micro-attribution that the data no longer supports. Accept that some loss of precision is real and not a vendor problem to be solved.

The political and regulatory backdrop continues to push in the direction of less granular tracking, not more. Anyone planning their 2027 measurement stack on the assumption of a return to 2020-style cross-domain attribution is planning for a world that isn’t coming back.