Black Friday 2026 is November 27 and Cyber Monday is November 30, which means the useful planning window is open right now and closes faster than it feels. The rule that survives contact with a Q4 auction: enter the quarter with a tested asset library, not with three hero ads and a plan to iterate. CPMs rise through Q4 and peak in Black Friday week, so testing late means paying peak prices to learn what you could have learned in September at a discount.
The calendar, backwards from November 27
| When | What must be done |
|---|---|
| Early September | Offers defined, creative direction agreed, testing budget released |
| Mid September to early October | Volume testing at low CPM. This is the cheap-learning window |
| Early October | Creator and talent commitments locked, products seeded |
| Mid to late October | Winners identified, tracking validated, budgets set |
| Early November | Library locked, launch sequencing agreed |
| November 27 to 30 | Execute. Do not test |
The single most expensive Q4 mistake is running the discovery phase in November. Auction prices climb through the quarter and spike in Black Friday week, so every learning bought then costs a multiple of what it cost in September.
How many assets to enter with
Meta's own advertiser guidance for small and mid businesses has settled around entering the quarter with a diverse, tested set in the range of ten to twenty assets rather than a handful of heroes. The reasoning is mechanical rather than aesthetic: a small set fatigues quickly at Q4 frequency, and a fatigued creative in a peak-CPM auction is the most expensive thing in the account.
Treat that as a floor for the tested library, not the total you produce. Producing thirty to reach a tested fifteen is normal, and the ones that die in testing did their job.
We are not going to quote a percentage-lift figure for asset count. Several circulate and we could not source any of them to a primary document, which by our own rule means we do not print them. The directional guidance above is what the platform actually publishes.
Where AI changes the plan, and where it does not
It changes the volume math. Producing fifteen tested variants used to mean a shoot day plus an edit week, which is why teams shipped three heroes and hoped. Using canon 8frame prices, a credit being $0.01 in a top-up pack: an 8-second Veo 3.1 Lite clip with audio is 54 credits, a 5-second Kling v3 Standard clip with audio is 86, and a still on Seedream 5 is 6. Fifteen video variants drafted on Lite is 810 credits, roughly $8, inside a single $19 Starter month.
It changes when you can start. You do not need the product photography finished to test hooks. Generate the context, test the openings, and swap the real product in when it arrives.
It does not change the offer. No amount of creative volume rescues a weak offer, and Q4 is the quarter where that becomes obvious fastest.
It does not change the tracking work. Validate first-party tracking in October, not in the first week of a peak-spend campaign.
A production plan that fits the window
September: test hooks, not ads. The opening two seconds carry most of the performance variance, so generate one body and five to eight openings against it rather than eight complete ads. See UGC hook formulas for patterns worth testing and how to version creative for a campaign for the one-axis-at-a-time discipline that keeps results readable.
Early October: build the winners out. Take the two or three openings that survived and produce full variants around them, in every format the placements need. This is the moment to spend on the premium models: a Veo 3.1 Standard 8-second clip with audio is 448 credits, and it is worth it on the shot that carries the ad.
Mid October: lock formats and captions. Every asset needs its vertical, square, and silent versions. Q4 feeds are muted feeds, so the caption layer is not a nice-to-have.
November: refresh, do not redesign. Keep two or three unreleased variants in reserve for the fatigue that arrives mid-flight. Refreshing with a held-back asset is cheap; designing a new campaign in peak week is not.
What to prepare that is not creative
The offer, written as one sentence. If it takes a paragraph, it will not survive a feed.
Tracking, validated on a live purchase. Not configured, validated.
A kill rule. Written in October: what spend and what metric retires a variant. Decisions made in advance survive the pressure of a peak-CPM week; decisions made during it do not.
The reserve library. Two or three assets nobody has seen, for week two.
FAQ
When is Black Friday 2026? November 27, with Cyber Monday on November 30.
When should Q4 creative be locked? Testing through September and early October, winners built out by mid-October, library locked in early November. Anything still being decided after that is being decided at peak CPM.
How many creative assets do I need for Q4? Enter with ten to twenty tested assets as a floor, plus a small reserve for mid-flight fatigue. Produce more than that to get there.
Can I test with AI-generated creative and run the real thing later? Yes, and hooks are the best candidate since the opening rarely depends on final product photography. Keep the tested variable identical when you swap, or you have retested nothing.
The cheap-learning window is September, and it is open. Draft the variants, test the openings, and spend the premium credits in October. The 8frame canvas is free and unlimited, and generation is paid from $19/month.