Smart Bidding after a conversion tracking fix: wait it out or change targets now?

Say you swap a lead import from form submits to qualified calls, or fix a purchase tag that was double firing. Smart Bidding has been learning from the old signal for months, and the conversion counts shift overnight. Some people say wait a full learning cycle, roughly two weeks of data, before touching targets. Others argue the model is already optimizing toward a broken number, so every extra day is wasted spend and waiting just pays for the mistake.

I lean toward the second view when the old signal was flat-out wrong, and the first when it was only incomplete. Where do you draw that line, and do you hold the target steady during the transition or adjust it right away?

Double-firing and swapped lead definitions break the target itself, not just the training data. If your tCPA was set while the tag was double-counting, it's too tight for your real cost per conversion, so bids end up too conservative. Compare old and new counts over the same days and rescale the target rather than waiting it out.

Rescaling works when the error is a multiplier. A double fire inflates every conversion by roughly the same amount, so the old and new counts give you a ratio you can apply to the target. A definition swap is a different animal. Qualified calls aren't a fixed fraction of form submits, and that ratio can swing a lot by campaign, device or keyword. So what matters is what kind of error you had. For a clean multiplier, Kai's approach holds. For a swap, I'd set the target from the new action's own recent cost per conversion and accept a rougher learning period.

Agreed on the swap case. A lot of people miss that the old action often stays set to "include in conversions" after the switch. Smart Bidding then optimizes to a blend of both signals, so your new tCPA ends up aimed at neither. Move the old action to secondary or turn it off before you set the target, and make sure the new one is the only primary action for that campaign. Otherwise the numbers you're comparing in reports aren't the ones bidding is actually using.

Agree on the blend problem, and there's a simple rule that falls out of it. If the new action is just a stricter version of the old one, like the same form fill with junk leads filtered out, treat it as a multiplier and rescale, like Kai suggested. If it's a different event, say a booked call replacing a form fill, treat it as a swap and set the target from the new action's own cost per conversion.

The reason is that a stricter version keeps the ratio fairly stable across campaigns, while a different event doesn't. Run both actions side by side for a week, split by campaign, and you'll see which one you're dealing with.

If both actions are set as primary, bidding is already optimizing on a blend of the two, so your side-by-side week only measures that blend. Keep the old action as secondary for the week instead. Secondary conversions still show up in reporting without feeding the bid, so you get a clean comparison by campaign.