Smart Bidding in a slow season: hold the target CPA or drop to manual?

Every January someone asks whether to pause or loosen Smart Bidding during a dead stretch and restart it later. My instinct is to hold, because pausing throws away the learned signal and the restart often costs more than the spend you saved. But the counter is fair: if conversion volume drops to a trickle for three weeks, the model is mostly bidding on noise and you're paying for clicks that won't convert. Take a furniture shop after its Black Friday spike. Do you move to manual CPC with a tight cap for the quiet weeks, or keep the target and accept a higher CPA until volume returns? I lean toward holding with a daily budget ceiling, but I'd like to hear how people explain the higher CPA to a client without it looking like the account broke.

Holding makes sense, and I'd add that the spike distorts the target itself. Black Friday conversions came in cheap, so the CPA the model learned from looks better than the shop's normal economics. When volume drops, the bids are still tuned to that peak. Before the quiet weeks, check what the CPA was in the same stretch last year and set the target to that level, not the peak-season number. And with the client, putting last January's CPA next to this one tends to land better than arguing about the account.

Agreed on the target, and the peak-season CPA really is the trap. Google has a seasonality adjustment for Smart Bidding that you can set up ahead of a short event, which tells the model conversion rates will shift for a set window. It's built for days, not a three-week slump, so it won't fix the quiet weeks on its own. But if you know the spike is temporary, flagging it keeps the model from carrying peak conversion rates into January. Check that the option is actually in your account first, since it's been rolling out in stages.

Agreed, and timing is the part people miss. As far as I know, the seasonality adjustment only affects bidding from the moment it goes live, so it won't undo what the model already learned from the Black Friday spike. Set it before the event starts, not after the CPA has already drifted. Put the end date on the calendar too, because an adjustment left running into January is the same distortion you were trying to avoid.

Here's the rule I'd give a client: count conversions over the last 30 days. If you're well under the 30 or so Google generally cites for stable tCPA bidding, the model is mostly guessing, so loosen the target to a CPA the shop can live with until volume recovers. Switching to manual doesn't give you better information, it just gives you less automated information. Put a date on the change so you remember to tighten it back.

Loosening makes sense, but count conversions per campaign, not per account. A shop can clear 60 across the account while one campaign sits at 4, and that's the one that's guessing. Loosen only that campaign's target and leave the others alone.

The campaign-level count can mislead in two ways. First, if the campaign sits on a shared portfolio bid strategy, Google pools conversions across every campaign on that strategy, so a campaign with 4 sales may look fine to the model. Loosening its target would then move the whole group, so check the strategy's campaign list before you edit anything, and detach the campaign onto its own tCPA if you only want to change that one.

Second is which conversion actions feed bidding. Furniture shops often count add-to-cart or lead events alongside purchases, so a campaign can show 40 "conversions" that are really 4 sales plus soft events. Count only the action that brings in revenue, or the thin campaign will look healthier than it is.