Customer Journey Attribution and What It Cannot Tell You

Customer journey attribution divides credit for a sale across the touchpoints a system observed on the way to it. The division follows a rule that somebody configured: first touch, last touch, linear, time decay, or a model the vendor calls data-driven. Change the rule and every channel's number changes while not one customer behaves differently. That is the whole problem. A rule for sharing credit is bookkeeping, and bookkeeping cannot tell you what caused anything.

What the report actually contains

A journey report is a list of sequences the system could resolve, with a share of revenue assigned to each step. It contains two implicit claims and you should separate them. The first is descriptive and usually reliable: these touchpoints were observed, in this order, before this order was placed. The second is causal and is never established by the first: this touchpoint is worth this much of the revenue.

Between those two claims sits an allocation rule. The rule is a policy decision. It is not a finding.

Why the rule changes the answer

Take a customer who sees a paid social ad, searches your brand name a week later, opens an email, and buys. Under first touch, paid social takes the revenue. Under last touch, email does. Under linear, each of three touchpoints takes a third. Under time decay, email and search take most of it. Four defensible rules, four different budget conclusions, one unchanged customer.

The attribution model you picked is therefore doing more work in your reporting than the customer did. Causal, rule-based and data-driven attribution in GA4 covers how each family behaves on the same export.

The journeys that are not in the report

Every allocation happens inside the visible set. GA4 resolves the sessions it can and files the rest under Direct or Unassigned. 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, one company's data rather than a population estimate.

Coverage is the fraction of real orders your data could see, and every figure in a journey report is a share of that fraction rather than a share of your business. A channel that looks like 20% of revenue in the report is 20% of the visible part. The number nobody checks has the query for your own property.

The question the report is not built to answer

What would have happened without the channel. That is a counterfactual, and it cannot be recovered by rearranging observed sequences, because the sequences in which the channel was absent are precisely the ones the report does not contain.

The evidence on this is not subtle. At eBay, brand-keyword search ads scored as a top channel under observed-journey logic until the ads were switched off, and 99.5% of the forgone paid clicks returned through natural search. The journey data had been describing which door a purchase walked through, not whether the door caused the walk.

What replaces the rule

Not a better rule. A different design. A causal read estimates what revenue would have been without each channel.

For the channel that carries the most budget, a holdout is still the anchor.

What to do this week

• If you have to defend the number: put the allocation rule's name on the slide next to the channel table. A number whose rule is unnamed cannot be discussed, only believed.
• If you own the budget: compute coverage before you read another journey report, so you know what share of your business the percentages are shares of.

The coverage census and the eBay experiment are from The Price of Being Found (Edition 2.10), Chapters 8 and 12, with the book's stated caveats.*