Doubling a Target CPA campaign's budget: step it up or jump in one go?

A client's Search campaign on Target CPA has been steady at around 40 conversions a month, and they want to double the budget ahead of a seasonal push. The usual advice is to raise it in 20% steps every few days so the bidding doesn't wobble. A big jump does get you to the new volume faster, though, and you only pay the learning cost once. My worry is that CPA usually drifts for a week or two afterward, and if the client sees that on the dashboard, you end up explaining a bad number instead of a bidding change. So I'd stage it, unless the business can handle a rough fortnight. Where do you land on stepping versus jumping when the reporting is client-facing?

I'd push back on the idea that a jump gets the client to the new volume faster. On a Target CPA campaign, budget often isn't what's limiting volume, the CPA target is. If lost impression share to budget is already near zero, doubling the daily cap may add little, because the bids still only win the auctions that clear at the same tCPA. Before picking a ramp, check Search Lost IS (budget) for the last month. If it's low, the real question is whether the client can accept a looser target, not how fast to raise the budget.

On learning cost, I'm not sure a budget change alone resets learning the way a tCPA change does, so I wouldn't assume a jump pays it once while staging pays it repeatedly. The drift usually comes from the extra spend reaching weaker queries, and that happens either way. A jump just shows it sooner and in one lump, which is the dashboard problem you described. Staging smooths the line, but it doesn't guarantee a clean one.

Elena's Lost IS (budget) point is where I'd start too, because it tells you whether the campaign is actually budget-capped. If it's losing a lot of impression share to budget, doubling the cap makes sense and the question of how to step it is a real one. If it's already near zero, a big jump mostly just pushes more spend into the same auctions, so a staged raise is safer since you can watch where the extra conversions come from.

Lost IS (budget) only tells you the cap isn't clipping you. On a tight tCPA, the auctions you're missing are often lost to rank, so look at Lost IS (rank) too. If that's the big number, raising budget won't add any volume.

Agreed, and on a tight tCPA the rank number is often what decides things. The bid is capped by the target, so bids sit low and you lose the auctions where competitors have better expected CTR or ad relevance. More budget doesn't change that. You either loosen the target, which the client has to accept, or improve quality on the same keywords with tighter ad groups and landing pages that match the query. Only the second one keeps CPA where it is.

I'd also check whether the rank losses cluster in a few ad groups. If they do, fixing those groups matters a lot more than the budget decision.

Did Lost IS (rank) change over the last couple of months, or has it stayed flat?

Quality work isn't the only way to hold CPA, and it has a cost people skip over. Tighter ad groups and better landing pages usually trigger a learning period and can cut volume for a couple of weeks, which is the same dashboard problem Iris raised about budget changes. Before rebuilding anything, look at whether the rank losses sit on queries that were never converting well. If they do, losing those auctions may be the cheapest way to hold CPA, and the client never sees the dip.

Marco, "the client never sees the dip" doesn't hold up. Giving up rank on weak queries still costs conversions, and the dashboard shows that just as plainly as a learning wobble does. The difference is that the drop is permanent and nobody agreed to it.

The bigger problem is the data. At around 40 conversions a month, a query-level verdict is mostly noise. A term with 25 clicks and no conversions may just be unlucky, and one lucky conversion can make a weak term look fine. So before calling anything "never converting well," pull 90 days and group by ad group or intent instead of by individual query. The losses worth accepting are the ones that stay weak across that wider window.

The version to bring to the client is then simple: these are the auctions we're deliberately not chasing, and this is what it costs in conversions.

Theo's right about sample size. At 40 conversions a month, 90 days gives you roughly 120 conversions, which is enough to judge by ad group but still thin for individual queries. Before you cut anything, look at whether the weak terms are weak on conversion rate or on cost per click, because the fixes are different.

I'm not sure "enough to judge by ad group" holds. Spread 120 conversions across five or six ad groups and each one has maybe 20 to 30. A group sitting at 20 can swing by a third on a handful of clicks, so ad group is only slightly less noisy than query level. I'd read those numbers as a list of where to look first, not a verdict.

The 90-day window is a problem too. If it runs into the seasonal period the client is planning around, you're blending two different demand levels, so check the dates before trusting the averages.