Journey Orchestration After the Third-Party Cookie

Orchestrating on first-party data changes what a system can observe and join. It does not change what the system can prove. A platform that decides which customer meets which experience, and then reports on how those customers behaved, has the same structural problem with or without third-party identifiers. The identifiers were never the reason observed journeys could not establish cause.

What actually changes

Three things get harder without third-party identifiers. Cross-site observation, meaning behaviour on properties you do not own, largely disappears. Identity stitching across devices weakens unless the customer logs in. And view-through measurement, always the weakest link, becomes mostly unavailable.

Three things do not change. Your own site behaviour, your own email and SMS engagement, and your own order history are all first-party data and remain available. So orchestration logic keeps working. What degrades is reach and joinability, not decision-making.

What does not change at all

The measurement problem. An orchestration system routes the customer it predicts will respond, then counts the responses. The routed group was more likely to convert before any routing happened, because that is what good targeting means. That inflation existed when third-party cookies were plentiful and it exists now, unchanged, because it comes from selection rather than from identifiers.

Losing cookies makes the visible set smaller. It does not make the visible set causal.

The number that moves, and by how much

Coverage moves. That is attributed conversions divided by the orders your store actually shipped, and it falls as identifiers thin out, as consent declines rise and as more traffic arrives through in-app browsers and assistants with no resolvable referrer. The Price of Being Found reports its publisher's own census at 48.6% of sessions with no resolvable source over 17 October 2025 to 2 September 2026, which is one company's data rather than a population figure.

Because reported cost per order is the real figure divided by coverage, a falling coverage rate inflates every cost figure you quote without anything happening in the market. Brands routinely read that inflation as rising acquisition costs. The number nobody checks has the query for your own property.

What survives the transition

Designs that never depended on following individuals in the first place.

A randomised holdout works on aggregates. You withhold an experience from a randomly chosen slice, fix the window in advance, and compare revenue per customer. No cross-site identity is required at any point. Incrementality testing for ecommerce is the playbook.

Counterfactual estimation on aggregated exports also survives. Causality Engine uses causal inference on the GA4 history you already have to estimate what each channel caused.

The trap in the transition

Vendors are selling identity graphs and modelled conversions as the replacement. Both fill the gap in the data, which is a different thing from measuring the effect. A modelled conversion is an estimate of an observation, not an estimate of a cause, and stacking a model on top of a coverage gap makes the reporting look more complete while leaving the causal question exactly where it was. Ask what the system would report if the channel had no effect at all.

What to do this week

• If you own the budget: track coverage monthly rather than annually, so a falling denominator does not get read as a rising acquisition cost.
• If you have to defend the number: state which of your metrics need cross-site identity. Anything that does is on a shrinking foundation and should be planned out of the reporting now.

Product facts as stated on causalityengine.ai on 9 September 2026. The coverage census is from The Price of Being Found (Edition 2.10), Chapter 8, rated Supported rather than Established by the book's own appendix.