Seasonal bidding for Shopping
How to plan bidding around known peaks and troughs, when to use seasonality adjustments, why over-reacting to normal variance costs you, and how to set targets ahead of a season instead of chasing it.
Shopping demand isn't flat, it spikes at known peaks like Black Friday and gift seasons, and dips predictably in between. Smart Bidding already learns gradual seasonality on its own, so the job isn't to react daily. It's to plan targets around peaks you can see coming, use the seasonality adjustment tool only for sharp short events, and resist over-tuning during normal week-to-week noise.
Seasonality isn't noise: but most week-to-week movement is
Two different things get called "seasonality" and they need opposite responses.
- Real seasonality is the predictable rhythm of your category: gift peaks, summer ranges, the January lull. It repeats, it's visible in your history, and Smart Bidding largely learns it on its own.
- Normal variance is the day-to-day and week-to-week wobble every account has. It looks like a trend for three days and then reverts. Reacting to it (nudging targets up and down chasing noise) is how accounts stay in a permanent re-learning loop and never let the algorithm settle.
The skill is telling them apart. If a movement is on your calendar (a known peak) or in your history (last year's dip), it's seasonality and you can plan for it. If it's a three-day blip with no cause you can name, it's almost certainly noise: leave it alone.
Plan targets around peaks you can see coming
The best seasonal bidding happens weeks before the season, not during it. Because every target change triggers a re-learning period, you want any adjustment settled before demand arrives: not kicking in mid-peak when the algorithm is still adjusting.
A workable rhythm for a known peak:
- Two to three weeks out, decide your peak stance. Higher-converting periods often justify holding your target (winning more at the same efficiency) or easing it slightly to capture surge volume. Move in small 10–20% relative steps so learning doesn't reset.
- Make sure budget can keep up. A great target is worthless if the campaign is budget-limited and stops serving by mid-afternoon on your biggest day. Pace the budget for the peak, not for a normal week.
- Hold through the peak. Resist tuning on the day. You set the stance in advance precisely so you don't have to make panic changes while the numbers are moving fast.
- Revert deliberately afterward, again in small steps, back to your baseline target.
Pre-labelling seasonal ranges with custom labels (q4-gift, summer, evergreen) means you can swing a whole segment's budget or target in one move when the season turns, instead of hunting products one at a time.
Deep dive The seasonality adjustment tool vs planning ahead: and not over-reacting
Google's seasonality adjustment tool lets you tell Smart Bidding "expect conversion rate to be X% higher/lower for this short window". It's a genuinely useful instrument, but it's the most over-used lever in Shopping bidding, so be precise about when it earns its place.
Use it for: short, sharp events (typically one to seven days) where you have a real reason to expect conversion rate to move hard and the algorithm has no way to know: a flash sale, a one-day promo, a product launch with pre-registered demand. The tool tells the algorithm to bid through the spike instead of waiting to observe it after the fact.
Don't use it for: long or gradual seasons. Smart Bidding already learns your recurring annual and weekly patterns from history; stacking a manual adjustment on top double-counts the same signal and distorts bidding. And don't use it as a substitute for a target change: if you want a different efficiency, change the target (in small steps, ahead of time), not the conversion-rate expectation.
The deeper discipline is not over-reacting at all. The failure pattern: demand wobbles for a few days → you apply a seasonality adjustment or yank the target → the wobble reverts → you undo it → learning resets twice for nothing. Peaks that are genuinely predictable should be planned, calmly, weeks ahead. The seasonality tool is for the rare sharp event you can't plan into the target: not for the everyday nerves of watching a live account. When you do want the steps chosen for you, a system that watches per-product pacing daily can propose seasonal target moves in advance and stage them for your approval, rather than leaving you to react in the moment.
What to watch during a peak
- Pacing, not just efficiency. In a surge, the risk is running out of budget before the day ends. Watch for budget-limited signals (see budget pacing) and top up the budget rather than tightening the target.
- Feed freshness. Peaks are exactly when price and availability errors cost the most. A product that goes out of stock or mis-prices during Black Friday is a wasted peak; keep feed health clean going in.
- Reversion discipline. The most common post-peak mistake is leaving a peak-time target or adjustment in place into the lull, then wondering why efficiency collapsed. Plan the revert with the same care as the ramp-up.
Frequently asked questions
Does Smart Bidding handle seasonality by itself?
For gradual, recurring seasonality, yes, largely. It learns your normal weekly and annual rhythm from history and prices auctions accordingly. What it can't anticipate is a sharp, short spike it hasn't seen before, which is exactly what the seasonality adjustment tool is for.
When should I use a seasonality adjustment?
Only for brief, sharp events where you expect conversion rate to jump or drop hard for a few days, a flash sale, a one-off promo. Don't use it for long or gradual seasons; the algorithm already handles those, and a manual adjustment on top double-counts.
Should I raise my tROAS for Black Friday?
Usually the opposite. Peak periods convert better, so the same target often lets you win more profitably. If anything, a slightly lower target during a genuine demand surge captures volume you'd otherwise leave on the table, but plan the move in advance, don't scramble on the day.