Target CPA changes: what size jump counts as a reset in Smart Bidding?

How big does a tCPA change have to be before you treat it as a new strategy? Say a store has a tCPA of 40 and the client is fine with fewer but more expensive orders, so you want to move to 55. My instinct is that a 10 to 15 percent move usually gets absorbed within a week or two. But plenty of people call anything over 20 percent a reset and pause for two weeks, which costs a lot of spend. I'm not sure anyone has clean data on where that line sits for low-volume accounts. What threshold do you use, and do you tell the client before or after you make the change?

Honestly it depends more on conversion volume than on the percentage. At around 15 conversions a month, going from 40 to 55 can act like a reset, because there's so little data to re-learn from. At 100 a month the same move barely registers. So I'd measure it in conversions per window, not percent.

Volume is the right lens, but I'd turn it into a rule you can actually use. Take the conversions the strategy got over the last 30 days and divide by 30. If that's under about one a day, treat any tCPA move above roughly 20 percent as a reset and stage it. Going from 40 to 55 is a 37 percent jump, so split it into steps a week apart: 40 to 46, then 50, then 55. Each step is small enough for the algorithm to adjust without throwing out what it has learned.

I'd stage rather than pause, because a pause loses both the learning and the spend you'd have gotten at the old target. Staging keeps delivery running and gives you a clean read after each step. The cost is that you reach the new CPA more slowly, which is usually fine when the client has already accepted fewer, pricier orders.

Staging makes sense, but I'd set the gap between steps by conversion lag, not a fixed week. If orders take five to seven days to show up in reporting, a week between steps means you're judging each step before its results have landed.

Agreed, and the lag throws off Priya's volume count too. If orders land a week late, the last 30 days look thinner than they really are, so a store doing a real conversion a day can read as 0.7 and get staged when one step would have been fine. Google Ads has a conversion lag report under Tools, Measurement, Conversions. It's worth checking how much of your volume lands after day 3 or 4 before you trust that threshold. The spacing rule holds up, but the threshold needs the same correction.

The bidder doesn't wait for your reporting. Smart Bidding picks up a conversion the moment it's recorded, so the algorithm often has fresher data than the Conversions column shows. That makes the 30-day count a proxy for what the strategy has learned, not an exact match. For the threshold I'd count only matured days in the numerator, and if volume is thin, switch to a 60-day denominator. The number gets noisier, but it's less likely to trigger a reset by accident.

Are the store's orders tracked on-site or imported offline? Offline imports can lag by weeks, and that changes the whole count.

Kai, I don't think the matured-days idea holds up. Dropping the last few days from the numerator just lowers the count, which pushes you toward a reset, the opposite of what you want. The lag doesn't change what the bidder has already learned, so the cleaner move is to count everything the strategy got, by the date of the click, and keep the denominator fixed.

You can check this in Google Ads by segmenting conversions by conversion time versus click time. For an offline import, the click-date view is usually the one that matches what Smart Bidding was trained on. If it still shows a big gap from the conversion-date view, that gap is your real lag, and it's a better reason to slow the step-up than any percentage rule.