Every January we run the same post-mortem across accounts, and the pattern never changes: the retailers who had a great Black Friday made their important decisions in summer. The ones who scrambled made them in November, when Meta CPMs across the retail accounts we track were running about 45% above October.
July: decide what winning means
Pull last year’s numbers apart before touching anything else. Which categories carried the volume? What margin was left per category after the discount, returns included? Which offers made money and which only made revenue? A 30% site-wide discount that pulled December sales forward at zero margin is a common finding, and it’s worth discovering now rather than next January.
Out of that analysis come two numbers per category: the revenue target and the discount floor. Then fix the measurement while things are quiet. If tracking is broken in Q4, you’ll be flying blind at 400 km/h.
August: feed and creative
Two production jobs. Both are slow, and both are cheap in August.
The feed first: titles, GTINs, categories and availability sync cleaned up now, so every improvement is live and stable long before you stop touching things. Shopping placements carry most retail spend in Q4, and the feed is their foundation.
Then the creative bank. Black Friday needs three to four times your normal creative volume: teasers, offer variants per category, urgency formats, retargeting versions, plus spares for whatever fatigues mid-week. Brief it all in August. Studio time costs normal rates, nobody is panicking, and September stays free for testing.
September: test with real money
Everything in the bank gets tested at moderate spend: hooks, formats, offer framings. By mid-October you want a ranked list, not opinions. Build audiences now too; engagement audiences and site-visitor pools need weeks of volume before they’re usable at scale. And if a new channel is on the Q4 wishlist, launch it in September or push it to next year.
October: freeze and lock
From mid-October, freeze the feed. No title-pattern rewrites, no category remapping, no restructures within three or four weeks of the peak. Price and stock updates keep flowing; structural change stops.
The same logic applies to bidding. Smart bidding needs two to three weeks to relearn after a strategy switch or a hard target change. Change bid strategy on 10 November and the algorithm does its learning during the most expensive auctions of the year. Lock strategies by the end of October, then move targets gently.
Last job of the month: write the pacing plan. Daily budgets from 20 November through 1 December, with triggers agreed in advance. If ROAS holds above X at 14:00 on Friday, budget scales by Y. Decisions made calmly in October beat decisions made at 23:40 on Black Friday.
November: hands off the structure
The work now is watching: pacing against the plan, hourly during peak days, and stock against spend, because nothing burns budget like advertising a sold-out bestseller. Wire stock levels into the review directly; a bestseller that sells out on Friday morning will happily keep spending through Sunday if nobody connects those two systems. Keep a short pre-approved list of changes anyone can make instantly. Everything else waits until December.
Accounts that follow this rhythm scale into Black Friday with tested creative, stable learning and a plan for every likely scenario. Accounts that start in October choose from what’s left: placements booked, learning windows too short, creative rushed. The gap between those two positions is built right now, in July.