Google turned your ROAS target into a ceiling

Over the last few weeks we've been seeing some odd behaviour across Google Ads accounts. Efficiency sliding on campaigns nobody had touched. Conversion volume down while spend sat flat. Brand campaigns quietly losing reach for no reason anyone could point at.

It traces back to a change Google made on 17 August, to how target-based bid strategies behave on budget-constrained campaigns. The rollout finished on the 27th. Because bidding changes take weeks to fully express, the damage mostly surfaced around 10 September — three weeks downstream of the cause, which is why a lot of teams are still looking in the wrong place.

You may already have spotted it. You may have seen the numbers dip and reasonably put it down to September being September. Either way it's worth twenty minutes to check, because the fix is straightforward and the run-up to Black Friday is not the time to find out the hard way.

Here's what changed, how to check your own account, and what to do before peak.

What changed

Before 17 August, a campaign flagged "Limited by budget" and running Target ROAS or Target CPA treated its target as a soft goal. It bid into the most efficient demand it could find, ran out of budget, and stopped. Which is why targets were routinely smashed. A brand campaign set at 4x tROAS could return 18–24x quite happily, because there was never enough budget left to go and buy the marginal traffic that would have dragged the average down to 4x.

Now Google optimises toward the number you typed in, whether the budget constrains the campaign or not. Google's own example: if your Target CPA is $10 and your actual CPA is $5, the campaign "now delivers more closely to a $10 actual CPA."

Read that again, because it's the whole thing. The target is enforced in both directions now. If you set it as a floor, it's a ceiling.

Drawn out, it looks like this. The shape is illustrative rather than one specific account, but it's the pattern that keeps repeating.

In scope: Target CPA and Target ROAS on Search, Shopping, Performance Max, Demand Gen and Travel, plus Target CPC for Demand Gen, across Google Ads, Search Ads 360 and Display & Video 360. Out of scope: App, Video reach, Video view. Display and Hotel already worked this way. If a campaign was never budget-constrained, none of this touches it.

Google changed nobody's targets or budgets. It shipped the behaviour to live accounts and left the correcting to advertisers.

Why it took three weeks to show up

Three things conspired here, and they're worth understanding because they're the reason this gets misdiagnosed.

The rollout ran from 17 to 27 August, so the cliff edge sits on a different day in every account, which kills the obvious "what changed on the 17th?" instinct. Smart Bidding then has to relearn, and a change this size takes around 30 days to fully express, so what you see is a slope rather than a step. And it compounds: lower bids buy fewer impressions, which produce fewer clicks and fewer conversions, which the model reads as a smaller opportunity, so it bids lower still.

Put those together and a change made in mid-August does its clearest damage in the second week of September, looking for all the world like seasonality, creative fatigue, or something you broke.

What it looks like in a real account

Here's the case of a health and beauty brand. All 14 live campaigns, nothing excluded, 20 July – 16 August against 17 August – 13 September.

Spend was flat: +0.7%. Conversion value fell 40.5%. Account ROAS went from 7.53x to 4.45x.

Same money in, 40% less value out. That's the sentence that makes the phone ring.

One caveat on those figures and the portfolio numbers further down: post-period conversions were still maturing when we pulled the data, so if anything this overstates the fall. We'd rather publish the worst-case read than sit on it for another fortnight.

Plot each campaign as delivered performance divided by the target set and you can see exactly what the system did. Before the change, campaigns sat anywhere from 1.1x to 3.7x above the targets they'd been given. After it, most of them collapsed onto the 1.0 line. Demand Gen campaigns are shown as target CPA divided by actual CPA, so the direction reads the same way as the ROAS campaigns.

The size of the fall tracked the size of the gap, every time. One sub-brand brand search campaign had been delivering 3.7x its target — Google took the budget it had been sitting on and spent it buying down toward that target. Spend up 134%, ROAS down 67%. Campaigns already running close to their targets barely noticed.

Worth knowing before you panic: several campaigns overshot. Brand search went from 1.13x its target to 0.73x, and two non-brand campaigns ended up delivering at about two-thirds of what they'd been asked for. The correction carries past the mark while the model relearns. It's the reason Google's advice to wait one to two conversion cycles is worth taking rather than ignoring.

The most useful row in that chart is the campaign that wasn't flagged budget-constrained. It moved the other way entirely, 0.53x target to 1.36x. Same account, same tracking, same trading conditions, same fortnight. If measurement had broken, that campaign would have gone down with the rest of them. It didn't, and that's what tells you this is a bidding change rather than a tracking one.

The wider Signal portfolio says the same thing. Roughly half of those accounts had targets configured at all, and within that group just under 60% saw efficiency move without a deliberate budget change. The other 40% carried on as normal. Same caveat as above — provisional, and a worst-case read. A tracking fault takes down close to 100% of accounts. A bidding change only touches the ones carrying a gap between target and actual.

That distinction matters more than it sounds. If you conclude this is a measurement problem, you'll spend a fortnight in your tag setup while the actual cause carries on throttling your campaigns.

How to check your own account

Don't take our word for any of it. If you've got a Google Ads MCP connected to Claude, or any agent with Google Ads API access, paste this in and you'll know inside a minute.

Using my Google Ads account [CUSTOMER ID]

Using my Google Ads account [CUSTOMER ID]

Using my Google Ads account [CUSTOMER ID]

One line in there will save you more time than the rest put together. In the account above, 12 of the 14 campaigns carried their target on maximize_conversion_value.target_roas or maximize_conversions.target_cpa_micros. Only two used the pure Target ROAS strategy.

So if you check the obvious fields, or scan down the bid strategy column looking for rows that say "Target ROAS", you'll find two campaigns, decide you got off lightly, and be wrong by an order of magnitude. Maximise conversion value with a target ROAS is a target-based strategy. It's in scope, and it's easy to miss.

The other field to watch is primary_status_reasons. BUDGET_CONSTRAINED is what puts a campaign in the affected group, and it's also what separates this from a tracking problem, which is the whole job of step three in that prompt. Run it across a few accounts and the pattern either shows up or it doesn't.

If you're a brand, here's the order we'd work in

The first three steps exist so the fourth one doesn't send you chasing the wrong thing. It's tempting to skip straight to changing targets. Resist it for an afternoon.

Start by isolating the population. Campaigns that were "Limited by budget" in the 30 days before 17 August and running a target-based strategy, including the maximise-with-a-target variants. Everything else is a separate conversation.

Then compare like for like. Baseline 20 July – 16 August against 17 August onward. What you're looking for is efficiency that moved without anyone deliberately changing the budget. If spend scaled 40% or more and conversions climbed with it, that's diminishing returns doing its normal job, not this.

Rule out the confounders before you act. Tracking breakage, a consent platform migration, a promo calendar difference, plain September seasonality. Your unaffected campaigns are the control group and they're free — use them. If everything moved, this isn't your problem and you should keep looking.

Then audit the gap, and fix worst-first. Pull actual performance for the 28 days before 17 August, set it against the target that's configured, and rank by the size of the gap. Raise targets to recent actuals, or take them off brand campaigns altogether. Brand is where the measured return is most inflated and where a hard target hurts most. Google's Bid Target Adjustment Tool will do the arithmetic for you, behind the "Review your campaign targets" banner or under campaign settings via Bidding, then Review campaigns. It won't suggest anything for campaigns with fewer than seven conversions, so those are on you.

Three ways to get this wrong.

Don't rebuild campaigns. Google draws on 90 days of history and sometimes a full year. Rebuilding throws it away at the exact moment you need the model to recover quickly.

Give your budgets room. This only bites budget-constrained campaigns, so taking off the constraint takes you out of the affected group entirely. Keep daily budget comfortably above average daily spend and a lot of this stops being your problem.

If you run portfolio bid strategies or shared budgets, make the changes at that level. Campaign-level edits won't hold, and it's an easy hour to lose.

Then leave it alone for one to two conversion cycles before you judge it. Genuinely leave it alone.

If you're an agency, the job is a bit different

Everything above still applies per account. At portfolio scale, three things change.

Triage by exposed spend, not by whoever's shouting. The loudest client usually isn't the most exposed one. Run the diagnostic across the book and rank by the spend sitting inside budget-constrained campaigns that were delivering above target. That's the number that tells you where the hours go. The concentration can be brutal inside a single account — in two accounts in the Signal portfolio, the share of campaigns running targets tighter than actual performance was 72% and 86%.

Split affected from unaffected before you touch anything. When several accounts degrade in the same fortnight, every instinct says shared cause: a tracking release, a consent platform update, a container change. It's the most expensive wrong turn available here. If every account moved, it isn't this. If the ones carrying target gaps moved and the rest held steady, it is.

Get to the client before the monthly report, not inside it. On a month-end deck this looks precisely like an agency that's dropped the ball: efficiency down, volume down, spend flat. There's no version of that meeting that goes well if the client sees the number before they hear the explanation. Lead with the diagnostic, the control group, and what you're doing about it. The performance-to-target chart is the one to put up, because it shows a mechanism rather than a mistake.

Two practical notes. Accounts running through Search Ads 360 won't see the Bid Target Adjustment Tool in Google Ads at all — it's behind the "Review bid strategies" banner in SA360 instead. And where you don't control the budget decision, put the budget-headroom recommendation in writing now. It's the structural fix, and it needs a client to say yes to it.

The bit that doesn't go away

Here's what's really changed, underneath the immediate mess.

Under the old behaviour, incomplete conversion data cost you efficiency at the margin. The bid strategy worked with a partial picture, you absorbed some wasted spend, and the budget constraint hid a lot of it.

Under a hard target, incomplete conversion data throttles delivery. Google optimises against the conversions it can see. When a decent share never arrive — consent loss, blocked client-side tags, broken event mapping, poor match rates — your measured ROAS sits below commercial reality. The system reads that as missing the target and cuts bids to close a gap that only exists in the data.

So you pay twice. Once for the conversions you never recorded, and again for the delivery Google pulls back to compensate.

For most brands this isn't hypothetical. We audited conversion tracking across 500 of the UK's biggest Shopify brands for the UK State of Tracking Report 2026: 98% had at least one critical tracking issue across Meta, Google Ads or GA4, and 52% were failing to track core events on their own site — the exact events a bid strategy optimises against. That 52% is the number that matters here. Those are brands whose targets are now being enforced against a systematically incomplete picture of what's actually converting. You can read the summary here or download the full report.

Before 17 August, under-measurement was a reporting problem you could work around. Now it's an execution problem you can't. Setting a hard target on top of partial event coverage is asking the system to optimise toward a number you already know is wrong.

Before Black Friday

This is the part worth acting on this month rather than next.

Your ROAS improves during peak — higher intent, better conversion rates, stronger AOV. Which means a target set at business-as-usual performance will throttle your campaigns at precisely the moment they're at their best. You get held at your November average while the auction is offering you the strongest week of your year.

The curve below is illustrative and your own peak shape will differ, but the shaded area is the part a business-as-usual target won't let you keep.

Set targets above what you expect to hit in peak, not at it. If you're expecting 6–8x through Black Friday week, put 10x in and let the campaign find its own ceiling. Do it in late October rather than the week before, because relearning costs days you can't spare in November — and as that account showed, the correction can overshoot before it settles.

And check your conversion tracking before you touch any of it. Under the new behaviour, a hard target sitting on top of incomplete event data is an instruction to throttle your best week of the year.

Google has taken away a bit of accidental upside that a lot of accounts had quietly been living on. Hard to argue it's wrong to make campaigns deliver what you asked them for. But it does mean your target is a live commercial input now rather than a safety rail, and the quality of the conversion data sitting underneath it decides whether the system delivers your intent or a distorted version of it.

If you inherited your targets from whoever built the account, they were almost certainly set as a floor and then forgotten about. Go and look at them this week. They're ceilings now.