The Last Day to Start a Holdout Test Before Black Friday

The last day to start the standard holdout test and still read it before Black Friday 2026 is Friday 2 October. The design needs eight weeks live, Black Friday is 27 November, and a test that finishes inside Cyber Week has no baseline left to compare against.

This is an appointment, not a suggestion. Octalysis calls the mechanic Appointment Dynamics: a fixed time at which an action is available and after which it is not. Here the appointment is set by the calendar and by the statistics of your own revenue, which is a more reliable source of urgency than any vendor.

Where the eight weeks come from

The Price of Being Found gives the geo-test arithmetic in a table of detectable lift by revenue wobble (ν, the coefficient of variation of weekly regional revenue) and by design length. The row the book marks as a typical DTC brand, ν of 0.08, reads:

Four weeks live buys you 11%. Eight weeks buys you 8.3%. The book notes a reviewer once read the 0.06 row instead and got 6.2%, which flatters the design by a third, so check your own row before you trust anyone's.

Counting back from 27 November

Eight weeks before Friday 27 November is Friday 2 October. Six weeks is 16 October. Four weeks is 30 October. A test started on 16 October with four weeks live ends on 13 November, two weeks before the peak, and can only see an 11% lift. A test started after 30 October is either too short to read or ends inside the peak, and Cyber Week is not a baseline for anything.

The pre-period matters as much as the live window, and it has to be fixed before you look at anything. Twenty-six weeks before 2 October is 3 April 2026. If your weekly revenue by region exists back to then, the history is already there. If it does not, the history cannot be manufactured, and the 4-week column is what you have.

The rule that decides whether to bother at all

Before booking the start, multiply two numbers you already have: the channel's spend as a share of revenue, and what you honestly believe a euro in that channel returns. A channel at 5% of revenue with an honest incremental ROAS of 2 would move total revenue by about 10% if switched off. Ten is above 8.3, so the test can answer. A channel at 2% with the same return moves revenue by 4%, which is below the floor, and the book's verdict is blunt: that test cannot work, and a null result will be read as an answer anyway. Can a brand your size measure Black Friday lift at all? has the full table.

Why the deadline is also a torture break

Octalysis has a second name for what happens after 2 October: the Torture Break, the enforced pause after which the reward is unavailable. Miss the start and the honest options are to run a shorter test that can only see a bigger effect, or to plan Q4 on judgement and say so. Both are defensible. Running the standard test anyway and reporting whatever comes out is not, because the answer was decided before the campaign started.

What to do this week

• If you have to defend the number: compute ν from last year's weekly revenue by region (standard deviation divided by mean, averaged across regions), read your row, and put the start date in front of the owner before 2 October with the MDE written on the same slide.
• If you own the budget: decide which single channel gets the test. The book's advice is one holdout, on the channel with the largest margin over its floor, registered in writing before it runs.

The Black Friday 2026 calendar has the other three dates.

Dates as of 9 September 2026. The detectable-lift table and the A times iROAS rule are from The Price of Being Found (Edition 2.10), Chapter 15, and carry the book's caveats.