Build in public · Bid Pilot

I scanned 59 accounts before Google's 17 August bidding change. Here's what was exposed.

Maria Waters · 27 August 2026 · 6 min read

On 17 August 2026 Google changed how Target CPA and Target ROAS behave on budget-limited campaigns. If you run Smart Bidding on more than a handful of accounts, it is the most consequential quiet change of the year, and most accounts I look at were sitting directly in its path.

What actually changed

Before 17 August, a budget-limited campaign on tCPA or tROAS routinely beat its target. You set a £40 target CPA, the campaign ran out of budget by lunchtime, and Google — with more demand than money — happily delivered leads at £22. The target was a ceiling you never touched.

From 17 August, Google optimises those campaigns to deliver at the stated target. Google's own example in the notice: target CPA $10, recent actual $5 — the campaign will drift towards $10 unless you update the target. It applies to Search, Shopping, Performance Max, Demand Gen and Travel campaigns. Google does not adjust the targets for you.

The target stopped being a hint and became a promise. A loose target that used to be harmless now costs you the difference.

The fix is not complicated, but it is counterintuitive: tCPA goes down to recent actual, tROAS goes up to recent actual. Most people's instinct is the opposite, because for years a low tCPA was the thing that throttled volume.

What the scan found

I built a scan into Campaign Owl in July to find every campaign this would bite: budget-limited in the last twelve months, on a target-based strategy, beating its target by more than 10%. It runs nightly on every connected account. Here's the fleet as of this week, anonymised:

Measure	Value
Accounts scanned with at least one exposed campaign	59
Campaigns exposed	140
On Target CPA	96 (69%)
On Target ROAS	44 (31%)
Reviewed and adjusted before 17 August	27 (19%)

Two things in that table matter more than the totals.

First, the exposure is concentrated. In the July scan, 57% of the total monthly money at stake sat in a single campaign. That is normal — most accounts have one campaign doing the heavy lifting — but it means "check a few campaigns" is not a strategy. You need the list.

Second, brand campaigns were the worst offenders, and nobody worries about brand. The clearest example: a brand campaign with a tROAS target of 800% that was actually returning around 4,630%. Perfectly healthy for years. From 17 August, Google has permission to spend six times more per conversion on that campaign and still call it "on target".

Only one in five were adjusted in time — and that's fine

19% adjusted before the switch sounds like a failure. It isn't, for a reason worth being honest about: the fix has a cost. Moving a target is a bid change, and every bid change puts the campaign back into learning. On an account that was working, a pre-emptive move risks trading a theoretical problem for a real one.

So the rule I settled on, and built into the tool, is a review cadence, not a panic: every campaign is reviewed every 21 days, the target moves at most 20% per cycle, and nothing moves within 14 days of the last change. The 17 August exposure list simply told us which campaigns to look at first.

What to do this week if you haven't yet

1. Pull every campaign that is budget-limited and beating its target — "Limited by budget" status at any point in the last year, on tCPA or tROAS, with actual better than target by 10%+. This is a twenty-minute job in a spreadsheet if you have fewer than ten accounts.
2. Check the direction before you touch anything. tCPA: lower it to recent actual. tROAS: raise it to recent actual. If you find yourself doing the opposite, stop.
3. Set the target at the performance you want to keep, not the one you'd tolerate. The old habit of leaving headroom is exactly what the change punishes.
4. Watch the next three weeks, not the next three days. Learning periods make the first week noisy. Judge the move on days 8–21.
5. Brand campaigns first. They're the ones with absurd headroom and the ones nobody reviews.

What happens next

Because the scan banks a snapshot every night, I can see what each campaign did after the switch — before-and-after on every one of the 140. That's the follow-up post, once there are three clean weeks of post-change data rather than one noisy one. If you want it when it's ready, the free audit checklist signup below gets you the email.

Numbers are from Campaign Owl's nightly scan across connected accounts, as of 26 August 2026. Client accounts are never named. Google's notice: "Changes to target based bid strategies" in Google Ads Help; the Bid Target Adjustment Tool went live 6 July 2026.