The Black Friday Number Nobody Checks: Unassigned Traffic

Every cost figure in your Black Friday plan is the real figure divided by the share of sales your tracking can see. The Price of Being Found reports one company's census, its own publisher's, at 48.6% of sessions with no resolvable traffic source over 17 October 2025 to 2 September 2026. That is one company and not a population estimate, and the authors say so. The number for your company is one query away, and almost nobody has run it.

The identity

Spend 1,000 on ads and make a hundred sales: cost per sale is 10. If your tracking sees eighty of the hundred, the dashboard divides 1,000 by 80 and reports 12.50. Cost per sale rose 25% and nothing happened. The book calls the share your measurement catches α, coverage, and states the identity plainly: reported cost per sale is the real one divided by α. If α fell from 0.8 to 0.6 over a year, your measured cost per acquisition rose 33% from the decay of your own visibility alone, before any change in media price or conversion rate.

Three things produce a rising cost line and look identical on the chart: media got more expensive, the site converts worse, or tracking decayed. Only the first two are real. Almost nobody separates the third one out.

Where the missing half lives in GA4

In the census the unresolvable sessions were 5,009 recorded as direct with no medium and 849 as (not set) on both source and medium, out of 12,054. In GA4 the first group surfaces as the Direct channel group, source (direct) and medium (none), and the second as (not set) on both, which lands in Unassigned. They are not misattributed. They are unattributed, arrivals from a place the system cannot name: consent declined, app traffic, cross-device journeys, stripped referrers, pasted links.

Peak week has its own version. More first-time visitors, more mobile, more social in-app browsers, more people arriving from a message on a phone and buying on a laptop. The α you measure in a quiet week is not the α of Cyber Week, and the plan is built on the former while the money is spent in the latter.

The query

1. Orders from the store for a period, the ones finance reconciles. Call it N_true.
2. Conversions your analytics attributes to any source for the same period, deduplicated. Call it N_seen.
3. α is N_seen divided by N_true.
4. Repeat for the same period twelve months earlier.

Run it for 24 to 30 November 2025 and for a quiet week in February 2026. The gap between the two is the peak penalty on your visibility, and it is the first thing to put beside any Cyber Week cost comparison. The one-hour claim ratio audit computes the companion number, the claim ratio, from the same pull.

Why this is the number to check before any of the others

The book's Monday protocol starts here, in week one, before cost, before tests. A level of 0.6 means every reported cost per acquisition is 67% above reality. A slope means part of every cost trend reported for a year was your own visibility decaying, and no amount of creative testing or bid tuning recovers that part, because there is nothing there to recover. It is also the first of the four properties of a defensible number: revenue of X, of which the systems can attribute Y, coverage Y over X, reported rather than hidden.

The book's publisher could only compute it over eleven months because that is how much warehouse history existed, which the authors treat as a finding about how long anyone's measurement memory is. Your Black Friday 2025 data may already be gone is the GA4 version of the same limit.

What to do this week

• If you own the budget: run the four steps for Cyber Week 2025. Write the number on the plan.
• If you have to defend the number: present every cost trend in decomposed form from now on, media price, conversion efficiency, and visibility, and be the only person in the room who can tell the three apart.

The calendar has the dates.

As of 9 September 2026. The 48.6% census (5,858 of 12,054 sessions), the α identity and the four-step method are from The Price of Being Found (Edition 2.10), Chapter 8, rated Supported rather than Established by the book's own appendix because the underlying data is not independently reviewable.