Changing target ROAS: how much and how often

Target ROAS is a commercial lever, not a maintenance dial

If you nudge target ROAS every week, in the end you are only measuring your own interventions. A target change is justified only when the actual ROAS lies outside the campaign's fluctuation band, on the wrong side of the value the business needs, and has done so for longer than two conversion cycles. Anything else is noise.

Which value the business needs is covered in the article Setting target ROAS and in the calculation of break-even ROAS. This article is about changing it while the campaign is running: when, by how much and how you check the effect.

How often: the rhythm follows the conversion cycle

Google defines the conversion cycle as the typical time it takes for a click to convert. According to Google, optimisation towards a new target starts within minutes, but because of conversion delay it takes one to two conversion cycles to reach it. After that, Google recommends evaluating performance over two conversion cycles or a two-week window, whichever is longer. Google advises against several changes within one cycle, because the bidder then has to process different versions of the target at the same time.

That gives a rhythm you can work out for your own campaign:

The last column is my own calculation from Google's guidance, not a statement from Google. For a typical shop where buyers decide within a week, that means realistically one evaluable change per month. If you change more often, you can no longer say which change produced which result.

When a change is needed: the gap must be larger than the noise

Before you touch the target, determine your campaign's fluctuation band. This is the real test, and I consider it more important than any rule of thumb on step size.

1. Take six to eight weeks without target changes, promotions or seasonal peaks.
2. Cut off the most recent days, in which conversions are still coming in. Why the latest days always look too weak is explained in the article on conversion delay.
3. Note the ROAS for each week. The range between the weakest and the strongest week is your noise band.

That this fluctuation is real is shown by an analysis from the agency smec of around 600 PMax campaigns: even campaigns without any target change delivered very different actual ROAS values. The target you set is therefore a steering value, not a result the campaign hits week after week.

A worked example with invented, round numbers. Suppose a shop needs a ROAS of 3.5 to cover its contribution margin, and the conversion cycle is seven days. Over eight weeks without changes, the weekly ROAS ranged between 3.6 and 4.6.

Case D is often overlooked. A ROAS far above the required value is no success if it means profitable sales are missed. For budget-limited campaigns, something has been added since August 2026: according to Google, they now follow a set target ROAS more closely, campaigns above target move towards it, and Google does not adjust targets automatically. A target set too high now costs volume more directly there. Details are in the article on budget-limited campaigns.

How much: step size according to the swing you can bear

Google names no percentage limit. The example in its help is a cut from 500 to 400 percent, so by a fifth, and according to Google a new target is used for bidding immediately. At the same time, Google points out that a major change in targets can have a similarly sized impact on spend or volume. The often-quoted limit of 30 percent is a rule of thumb, not a Google requirement.

So I would not derive step size from a rule of thumb, but from two questions:

• Is the step big enough to get out of the noise band? A step of 5 percent produces an effect that you can never tell apart from chance in a campaign with a 25 percent range. Then you have waited a month and know nothing.
• Can you bear the swing for a month? If you lower the target by 20 percent, expect spend to rise noticeably. If that does not fit your budget and stock, make the step smaller.

The smec analysis gives two pointers here: target increases reacted more volatilely than decreases, and campaigns with more than 60 conversions a month reacted more predictably. In practice, in my view, that means smaller steps when raising a target and more caution overall with few conversions; see also how many conversions target ROAS needs.

What happens with big jumps

According to Google, a big jump does not start a new learning phase. Changes to targets do not delete historical data and do not trigger the "Learning" status. Dr. Web has analysed this revised version of the help page. Guides that warn of a new learning phase with every target change are therefore out of date.

The risk of big jumps lies elsewhere. Bids react immediately, but the conversions that go with them arrive only with a delay. In the first week after a clear cut you will therefore see higher costs with apparently weak revenue, and that is exactly what tempts people to turn it back. Anyone who then intervenes has two targets in the same cycle and a campaign that follows neither of them cleanly.

How to tell whether the change worked

Evaluation is the part where most changes fail, because the measurement is taken too early or over the wrong period. These checks are enough:

1. Cut the period correctly. Start only after the adjustment period and, according to Google, exclude the most recent period affected by conversion delay. For the bid strategy report, Google recommends at least two full conversion cycles, ideally a month or 50 conversions.
2. Compare with the average target. The "Average target ROAS" metric averages your target changes over the period. If the actual ROAS in the evaluation window is close to that value, the campaign followed the target.
3. Hold it against the noise band. If the change in actual ROAS lies within the old range, the effect is not proven, even if the direction is right.
4. Check contribution margin, not ROAS alone. A cut has worked if absolute contribution margin after ad spend has risen, not if ROAS stays above a number.

If you want to keep the uncertainty small, Google offers to save a target change as an experiment instead of applying it immediately. Part of the traffic then runs with the old target and part with the new one, and the noise affects both sides equally. This is worthwhile above all for big steps in high-revenue campaigns.

When the rule does not apply

The monthly rhythm is not dogma. Google itself names a true business need as the exception. I count a clearance sale, a price change that shifts the required ROAS immediately, or a promotion with a different margin. In those cases you change the target because the economics have changed, not because the numbers fluctuate. How margin differences across the range lead to separate targets is covered in the article on target ROAS by margin group.

If ROAS suddenly drops sharply, check the measurement before you raise the target. Broken tracking or double-counted conversions is not solved by a target; that is a job for the reconciliation with the shop system. And with very few conversions a month, the noise band is often so wide that hardly any change can be evaluated cleanly. Then the better question is whether the campaign structure pools enough data, not which number sits in the target field.