The first week of every November brings the same messages: "Black Friday is coming — should we restructure the account? Change target ROAS? Launch new campaigns?" The intent is always good: entering the big day at full strength. The timing is usually disastrous, because smart bidding systems have one hard truth: big changes renew learning, and performance during learning is volatile. You can absorb that volatility in the year's cheapest week or its most expensive one — your choice.
Why the learning phase bites exactly that week
Bid-strategy changes, large target adjustments, campaign restructures — all of these partially invalidate the system's performance predictions and start a recalibration period. During it, the system explores: overbidding some auctions, underbidding others; CPAs swing. In a normal week that's a tolerable cost. In Black Friday week two effects stack: CPCs are already inflated by competition, and user behaviour (discount-waiting, comparison-shopping, cart-parking) deviates from historical patterns. You've pushed the system into its blindest period on the most expensive, most atypical traffic there is. Behind most "the ads went crazy on Black Friday" complaints sits exactly this: the calendar went crazy, not the ads.
The right calendar: planning backwards
My seasonal plan is built backwards from the big day. 6–8 weeks out: all structural work — campaign architecture changes, bid-strategy migrations, trials of new campaign types, feed/Merchant Center health locked down. The system must enter the big day stable and trained. 3–4 weeks out: creative and copy variants uploaded and tested; discount-message variants ready but paused. Budgets begin climbing gradually — a ramp, not a jump, so the system acclimatises to rising volume. The final week: only pre-planned, "safe" changes: opening budget caps (budget increases don't reset learning), activating the prepared discount creatives, switching on promotion assets and Merchant Center promotions, and using tools platforms built precisely for this — like seasonality adjustments for the expected conversion-rate spike. Nothing structural gets touched.
The day itself and after: refusing panic
The most common in-day mistake is looking at pre-noon performance and changing targets in a panic. Conversion lag is real: the morning clicker thinks at noon and buys in the evening; midday ROAS always looks worse than the day's final number. My in-day intervention list is short: budgets must not run dry (watch), the site must stay up (watch), sold-out products must drop from ads (feed automation) — don't touch bid targets. The post-season demands symmetrical discipline: budgets and targets return to normal gradually, because a sharp pullback produces its own turbulence. The classic reason accounts can't recover in January is the hard brake in late December.
One exception: when it's genuinely broken
All of this is "don't touch a working system" discipline — not "protect a broken one." If conversion tracking is broken, the feed is suspended, the budget sits on the wrong campaign — you fix those even mid-season, because every hour on corrupted data is a loss. The distinction: infrastructure faults are fixed immediately; strategy changes are kept out of the season. Not confusing the two may be seasonal management's most critical judgment call.
In short: Black Friday performance is won in September and October, not November. The boldest move before the big day is usually making no bold moves at all. Give the systems the time they need to learn; you don't switch textbooks in exam week.