Target Based Bid Strategies: What Changed in August 2026
On 17 August 2026 Google changed how target-based bid strategies behave. Budget-limited campaigns that used to overdeliver against Target CPA and Target ROAS — quietly producing conversions cheaper than the number you set — now optimise much closer to that number. Google did not change anyone’s targets or budgets automatically. But since that date plenty of accounts have seen cost per conversion climb, volume shift, and month-over-month comparisons stop making sense. Here is what actually changed, who it hits, how to separate it from seasonality, and what to do with your target based bid strategies this week.
Target based bid strategies: the old behaviour and the new
Target CPA and Target ROAS run on an agreement: you name a number, the system tries to hold it while buying as much volume as possible. There was a well-known quirk for anyone running on a constrained budget.
When a campaign’s budget ran out before the system could buy all the traffic available at your target, the algorithm behaved conservatively. It cherry-picked the cheapest, highest-probability conversions, and reported a CPA well below target. On paper that looked like a gift: “I set a $40 target and I’m getting $28.” In practice it meant your campaign’s real target was $28, and you did not know it.
After 17 August 2026 the system optimises more consistently toward the target you actually set, including when you adjust budgets. The campaign that used to deliver $28 against a $40 target is now more likely to move toward $40 — buying more volume at a higher price.
| Before 17 Aug 2026 | After 17 Aug 2026 | |
|---|---|---|
| Budget-limited campaign | Frequently overdelivered: actual CPA below target | Tracks closer to the stated target |
| Response to a budget increase | Unpredictable — CPA could jump | More consistent movement toward target |
| What your target meant | A ceiling the system often stayed well under | A goal the system actively works toward |
The one-line version: your target used to be a ceiling you rarely reached. Now it is a destination. If you set targets “with headroom” knowing you would actually pay less, that headroom has stopped working in your favour.
Who this affects
The change applies to Search, Shopping, Performance Max, Demand Gen, Display, Hotel and Travel campaigns. App campaigns, Video Reach and Video View campaigns retain their previous bidding behaviour.
Within the affected types, exposure is uneven. The sharpest impact lands on:
- Campaigns whose impression share lost to budget sits consistently above 20–30%. That is the working definition of budget-limited. How to read the metric is covered in the piece on impression share and auction insights.
- Accounts where targets were set by feel and never revisited. Those targets are usually well above what the economics support; the account survived on overdelivery.
- Performance Max campaigns running several asset groups on one shared budget, where internal reallocation becomes more visible.
Accounts where budget does not constrain delivery will barely notice. If lost IS to budget is near zero, the system was already working toward your target and nothing meaningful has changed.
Diagnosing your own exposure
Step 1: find the budget-limited campaigns
Add the “Search lost IS (budget)” column, sort by cost, and flag everything above 20%. Those are your candidates.
Step 2: compare actual CPA to target, before and after
Take two windows: 18 July to 16 August and 18 August to today. For each campaign compute actual CPA divided by target CPA. If that ratio used to sit at 0.7–0.85 and now sits at 0.95–1.05, you are looking at this change and not at seasonality.
Step 3: check whether volume grew
This is not purely a cost story. Conversions frequently rise alongside CPA, because the system now buys traffic it previously declined as too expensive. If you are getting more conversions on the same budget at a CPA that has risen to your target, the system did precisely what you asked it to do.
Step 4: rule out everything else
August also brought DSA-to-AI Max migrations, competitive shifts and pre-season ramp-up. To avoid attributing all of it to one cause:
- Use a control group — campaigns that are not budget-limited. If they show the same trend, the driver is external, not bidding.
- Check the change history for edits you made yourself around the same date.
- If a search campaign migration overlapped, separate the two effects deliberately. Context is in the guide to the DSA to AI Max migration.
Three scenarios and what to do in each
Scenario A: the target had headroom your economics never justified
The most common case. You set a $40 tCPA, got $28, and every downstream number — margin, payback, forecast — was quietly built on $28. Now the system is heading for $40 and the margin disappears.
Fix: recalculate the target from unit economics rather than habit. The maximum allowable cost per lead is straightforward:
Max CPA = Average order value × Gross margin × Lead-to-sale rate × Share of margin you will spend on acquisition
Worked example: $200 order value, 45% margin, 30% lead-to-sale, willing to spend half your gross margin on acquisition → 200 × 0.45 × 0.30 × 0.5 = $13.50. That is your ceiling per lead, not whatever number looked reasonable in the interface. The full method is in the piece on unit economics and LTV in media buying.
Do not slash the target in one move. Rough guide: steps of no more than 10–15%, with 7–10 days or 30 conversions between them so the system can settle.
Scenario B: the target was realistic, the budget was the constraint
Your conversions were cheap not because the target was inflated but because the budget only covered the top of demand. Now the system reaches deeper at a price closer to target.
Fix: decide whether you want that incremental volume. If your max CPA supports it, raise the budget and take the volume — this is the case where the change works for you. If it does not, lower the target to match reality and accept smaller volume. What a budget increase might buy can be modelled in the Performance Planner, remembering that its output is a scenario, not a promise.
Scenario C: you genuinely do not know what the right target is
Then the problem is not the target, it is the signal. If you send one generic “form submitted” goal with no value attached, any target is arbitrary — the system optimises for event count, not money. Before touching numbers, fix:
- Value differentiation across conversions, via conversion value rules and value-based bidding.
- Real closed deals flowing back in, via offline conversion import.
- Click identifier integrity, via GCLID, GBRAID and WBRAID tracking templates.
While the signal is broken, tuning targets is steering a car with no wheels attached.
Max CPA is calculated differently by business model
The formula above is the base case: one transaction, right now. In roughly half of accounts it produces a number that is too low, because it ignores repeat purchase. Use this as a reference:
| Model | Basis for max CPA | Common error |
|---|---|---|
| One-off purchase, retail | Gross margin on the first order × acceptable acquisition share | Calculating from revenue rather than margin — target ends up roughly double what it should be |
| Repeat purchase | Margin across the first 6–12 months of customer life | Counting only the first order — target too low, volume left on the table |
| Subscription | Margin × average tenure − cost to serve | Using undiscounted lifetime value — the money arrives years out, the invoice arrives this month |
| Lead generation | Deal margin × lead-to-sale rate | Ignoring that lead-to-sale varies sharply by campaign and geography |
| Marketplace or aggregator | Your take rate × transactions within the payback window | Modelling on platform basket size rather than on your own commission |
Payback horizon deserves its own line. If you are willing to recover acquisition cost over three months rather than on the first transaction, your max CPA rises — sometimes by two to three times. That is a legitimate way to raise targets and take volume, but only under two conditions: you have verified repeat-purchase data, and you have the working capital to survive the gap between spend and return. Without the second, “we bid on LTV” is just a cash flow problem with a strategy label on it.
Target ROAS: same logic, trickier arithmetic
The change works identically on tROAS, but the failure mode is easier to hit, because the number is more legible and people tend to set round ones. Three things worth checking on ROAS campaigns specifically.
ROAS is only as honest as the value you send
If your site passes order totals including tax and shipping, your reported ROAS is inflated by exactly that amount. A 400% target on revenue can be a 130% target on margin. The cleanest fix is to send margin as conversion value instead of revenue — breakeven then sits at 100% and no mental arithmetic is required. Setup detail is covered in conversion value rules.
An average ROAS hides the spread
A campaign averaging 500% may be half traffic at 900% and half at 100%. While the system was overdelivering, it stayed in the upper half of that distribution. Now it reaches into the lower half too. Segment by product category or ad group to check; if the spread is 2× or wider, the answer is splitting the campaign, not nudging one blended target.
Seasonality adjustments still exist
For short, predictable spikes — a weekend sale, a three-day promotion — there is a dedicated seasonality adjustment mechanism. It exists precisely so you do not have to move a target for an event that ends in 72 hours. Editing the target instead creates two disruptions: one when you raise it, another when you put it back.
Performance Max deserves a separate look
PMax is among the affected types, and the change plays out with its own quirks there.
- Shared budget across asset groups. The campaign reallocates between them on its own. With tighter target adherence, that reallocation becomes more visible: an asset group that was quietly harvesting cheap traffic may now get less.
- Brand traffic inside PMax. If brand is not excluded, it drags the campaign’s blended CPA down and masks what is happening. Verify brand exclusions before drawing any conclusions about the target.
- The feed as a ceiling. With a narrow catalogue there simply is no additional volume available at a higher target — the campaign hits demand limits, not budget limits, and raising the target achieves nothing.
Structure and signal handling in PMax are covered in the complete Performance Max guide.
When to drop target-based bidding altogether
There are situations where Target CPA and Target ROAS now perform worse than strategies without a target.
- Low conversion volume. Rough guide: under 15–30 conversions per campaign per month. The model never accumulates enough data and the target becomes an arbitrary number. Maximise conversions with a firm budget is more honest here.
- You want all available demand. Brand campaigns and narrow niches with capped volume. A target only gets in the way.
- Sharp demand swings. During promotional peaks, a target calibrated on normal weeks throttles volume exactly when you want volume.
The trade-off is real: strategies without a target spend the full budget and swing more on CPA when budgets change. You are trading predictability for volume, deliberately. A full comparison sits in the guide to Google Ads bidding strategies.
How to change targets without breaking learning
- One variable at a time. Change the target, and leave budget, structure and assets alone that day. Otherwise attribution is impossible.
- Steps of 10–15%. Anything above 20% in one move reliably triggers several days of instability.
- Wait 7–10 days or 30 conversions, whichever comes later. Reading earlier is reading noise.
- Write it down. Date, campaign, old value, new value, hypothesis. Without a change log you cannot reconstruct the month.
- Test, do not observe. If meaningful spend is involved, run the change through Experiment Center rather than flipping it across all traffic.
What this means for budget planning
A side effect worth naming: planning got more honest. Previously, setting a budget gave you two levers — the budget itself, and a hidden discount from target overdelivery. Now there is one lever each: budget controls volume, target controls price.
| What you do | What to expect now |
|---|---|
| Raise budget, leave target | More conversions at a price close to target |
| Lower target, leave budget | Fewer, cheaper conversions; budget may go unspent |
| Raise target | More volume at higher cost, up to available demand |
| Lower budget | Less volume; CPA stays near target rather than dropping as it used to |
That last row is what breaks old habits. Cutting budget used to be a way to “make it cheaper.” It no longer works that way — price is the target’s job now.
This week’s checklist
- Export every tCPA/tROAS campaign with lost IS to budget and actual CPA across two windows, before and after 17 August.
- Flag campaigns where the actual-to-target ratio moved by more than 10 percentage points.
- For each, recalculate max CPA from economics rather than from history.
- Where the target exceeds max CPA, begin stepping it down 10–15% at a time.
- Where the target is below max CPA and you want volume, raise the budget and take the demand.
- For campaigns under 15–30 conversions per month, consider dropping the target entirely.
- Record the new baseline so next month’s comparison means something.
To stop rebuilding this analysis by hand, put the ratios into calculated columns — the method is in the guide to custom columns and the Report Editor. The broader account review sequence is in the Google Ads account audit checklist.
If you manage a portfolio through Google Ads agency accounts, run this centrally. The change hit every campaign at once, and working through one account a week means paying above your economics for a month. The underlying theory on bids and budgets is covered in the Google Ads guide.
FAQ
What exactly changed on 17 August 2026?
Budget-limited campaigns on target-based bid strategies now optimise more consistently toward the target you set, including through budget adjustments. Previously such campaigns often delivered a CPA well below target.
Did Google change my targets or budgets?
No. No automatic adjustments were made to targets or budgets. What changed is how the system pursues the numbers you already set.
Which campaign types are affected?
Search, Shopping, Performance Max, Demand Gen, Display, Hotel and Travel. App, Video Reach and Video View campaigns keep their previous bidding behaviour.
How do I know if a campaign is budget-limited?
Look at impression share lost to budget. Values consistently above 20–30% mean budget is genuinely constraining delivery — those campaigns feel the change most.
My CPA went up. Is this definitely the cause?
Check with a control group: campaigns that are not budget-limited, same period. If they show the same trend, the driver is external — season, competition, or your own edits. The change history covers the last of those.
Should I cut targets immediately?
Not immediately. First calculate your maximum allowable CPA from economics. If the live target exceeds it, step it down 10–15% at a time with 7–10 days between moves.
Why not halve the target in one move?
A large jump throws the system into instability: volume drops, price swings, and several days of data become unusable for evaluation. Gradual steps give the model time to adapt.
I got more conversions but they cost more. Good or bad?
It depends on your economics. If the new cost per conversion is still below your maximum allowable CPA, you are simply buying more available demand — that is a win. If it is above, cut the target.
What does lowering the budget do now?
It reduces volume without reducing cost per conversion the way it used to. Trimming budget is no longer a lever for getting cheaper conversions; that is what the target is for.
When should I move off target-based bidding?
With low conversion volume (rough guide: under 15–30 per campaign per month), when the goal is to capture all available demand, and during sharp demand peaks. In those cases untargeted strategies usually behave more predictably.
Does this affect Performance Max?
Yes, PMax is among the affected types. The effect is most visible where several asset groups share one constrained campaign budget.
How do I validate a target change properly?
With a traffic-split experiment rather than a before-and-after read. Rough guide on duration: three to four weeks, or until each arm has 50–100 conversions.
