Meta Wants 50 Events in 7 Days. Cyber Week Is Four

Meta's own campaign guidance states that learning happens at the ad set level and that "the learning period stabilizes with a minimum of 50 events over a 7-day period", and that "making changes to your ad set can trigger a new learning period" (developers.meta.com, retrieved 9 September 2026). Black Friday to Cyber Monday 2026 runs 27 to 30 November, which is four days. A new ad set launched into that window cannot finish learning inside it. For seven days of learning to sit behind you before the peak, the ad set has to be live and untouched from Friday 20 November, which is 72 days from today.

That is the entire argument. The rest is the arithmetic, the trap that catches people on the Saturday, and the separate reason a single day of data cannot grade a channel even when the campaign is stable.

Fifty events, per ad set, in seven days

The unit matters more than the number. Fifty is per ad set, so a budget split across six ad sets needs three hundred optimisation events in a week before any of them is out of the phase. Meta's same page warns that too many ad sets with overlapping targets can cannibalise data collection, which is the polite description of a structure that guarantees none of them learns.

The arithmetic is worth doing with your own figures rather than a benchmark. At a cost per purchase of fifty euros, fifty purchases costs two thousand five hundred euros per ad set per week, before you have a stable delivery pattern to read. At a hundred euros it is five thousand. Neither number is a claim about your business. Both are multiplication you can redo in a spreadsheet in a minute, and the result tells you how many ad sets your Black Friday budget can actually support. For most brands under the fit floor the honest answer is one or two, not the twelve in the plan.

Why four days is the wrong denominator

Cyber Week is short, and the peak inside it is shorter. A campaign launched on the Wednesday before Black Friday has three days of data before the biggest spending day of the year, which is not a learning window, it is a coin toss with a budget attached. The 20 November date is not a marketing deadline invented to make you hurry. It is 27 November minus the seven days Meta names in its own documentation, and you can check both halves.

If you cannot be ready by then, the honest plan is to run the peak on ad sets that already have history rather than new ones, and to accept that anything launched inside the window is being bought rather than measured.

The edit trap

The instinct on a disappointing Friday evening is to change something: raise the budget, swap the creative, narrow the audience. Meta documents that changes to an ad set can trigger a new learning period. So the reaction that feels like taking control is the one that guarantees the rest of the weekend is unreadable, and it arrives precisely when the numbers look worst and the temptation is highest.

Decide the intervention rules before 20 November, write them down, and treat the weekend as a period where you execute rather than tune. The sunk cost prison covers the related trap, which is refusing to stop spending on something because of what you already spent on it.

The second problem, which no launch date fixes

Suppose the campaign is stable and out of the learning phase. One day of data still cannot tell you what a channel caused. The smallest effect a design can detect depends on how much data it has and how much the underlying revenue wobbles, and a single day of a highly promotional week is the noisiest data you will collect all year. The interval around a one-day read is wide enough to contain both "this worked" and "this did nothing", which is why the platform number looks decisive and the honest number does not.

That is a floor, not a preference. Whether your channels are measurable at all has the arithmetic, and incremental ROAS per channel from a GA4 export covers what an observational read can and cannot say between experiments.

What the honest read looks like instead

Wait for the attribution window to close, then compute two numbers before you rank anything: coverage, attributed conversions over orders, and the claim ratio, summed platform claims over orders. If the platforms collectively claimed more orders than your store shipped, the channel table is arguing with itself and the ranking is not a ranking. The one-hour claim ratio audit is the procedure, and the post-mortem that waits for the window to close is the calendar version.

What to do this week

• If you have to defend the number: put 20 November in the campaign calendar as the freeze date, and write the intervention rules now, while nothing is on fire.
• If you own the budget: count how many ad sets your peak budget can carry to fifty events a week at your own cost per purchase, and cut the plan to that number.

Two other dates in this series are already inside the next seven weeks. Four measurement deadlines come first lists them, and the last day to start a holdout test is 2 October.

Learning phase figures and the wording about edits are from Meta's campaign optimisation guidance at developers.meta.com, retrieved 9 September 2026; Meta may change both. Black Friday 2026 falls on 27 November and Cyber Monday on 30 November, and the 20 November and 72-day figures are arithmetic from those dates as of 9 September 2026. Measurability and interval material is from The Price of Being Found (Edition 2.10), Chapters 15 and 19, with the book's caveats.