On August 17 2026, Google changes how Target CPA and Target ROAS campaigns behave when they are limited by budget. Campaigns that have been beating their stated targets will start optimizing toward the target you actually set. The advice everywhere is to align your target with recent performance before the rollout so nothing changes.
That advice is not wrong. It is answering a smaller question than the one in front of you. Everyone is asking what target to set. Almost nobody is asking what happens after you reset it and uncap, which is the question that decides whether the reset was worth doing.
What The Budget Cap Was Actually Doing
When a campaign is limited by budget, the system lowers bids to make the money last through the day. Lower bids win a narrower set of auctions, and the ones it keeps winning are the cheapest, most predictable conversions available. That is a skim, not a sample.
Your reported CPA was the cost of the easiest conversions your budget could reach. It was never the cost of your demand.
The two numbers diverge more the harder your budget squeezes. A campaign capped at 30% of the spend it could absorb is skimming a much thinner layer than one capped at 90%. Both report a CPA. Only one is close to describing the market you are actually buying in.
The Number Everyone Is Missing
Here is the part the whole industry conversation is skipping, and it is the part that decides whether the promised scaling shows up in your account.
Your target is a blended average across every auction the campaign wins. Scaling is a marginal question. What matters is not what the average conversion cost, but what the next one costs.
The conversions you add when you raise budget are, by definition, the ones the system did not previously think were worth buying. They cost more than your average.
Google has now said this part out loud. In its August 5 community update, Ginny Marvin confirmed that a campaign holding its target will “likely mean entering different auctions than you previously competed in” to hit it. Different auctions is the whole point. The ones you were not competing in before are the more expensive ones, which is exactly why the next conversion costs more than your blend.
So a campaign at a €20 blended CPA can be picking up its marginal conversions at €35 or €40. Raise budget and the blended number drifts up, not because anything broke, but because the mix changed. The “more conversions at lower efficiency” that everyone is casually accepting is not a flat trade. It is a curve, and it is usually steeper than people expect.
This is the honest core of it. The predictable scaling being promised is really target adherence. Target adherence says the system will keep buying at your number. It does not say how much volume exists at that number, and for a lot of accounts the honest answer is: less than you would like.
But Not Every Account Feels This Equally
How steep your curve is depends on how hard you were capped. A campaign losing 60% of its impression share to budget was skimming a thin, cheap layer, and the demand sitting just past the cap is a lot more expensive. That curve bends hard. A campaign losing 10% of impression share to budget was already close to buying its full demand, so its marginal and blended costs are close together, and uncapping barely moves efficiency.
So before anything else, check impression share lost to budget over the last 90 days. This is the standard filter for finding genuinely budget-limited campaigns, and it matters more than it sounds: spending your full daily budget does not mean you are budget-limited, it just means you are spending it. Plenty of campaigns that look capped are actually rank-limited, and on those, Google’s change barely touches you and resetting the target just chokes volume for no reason.
The campaigns to worry about are the narrow set that are genuinely budget-limited and meaningfully over-performing. Those are where the curve is steep and where the marginal question actually decides your outcome.
How To Measure It Before You Commit
The marginal point is only useful if you can act on it, and measuring the cost of your next conversion is harder than reading a blended number. Demand fluctuates, seasonality moves, conversion lag blurs the window. Here is the cleanest approach I have found.
Do not uncap in one move. Raise budget on a single campaign in increments of 20% to 30%, and hold each step for two full conversion cycles so the lag resolves before you read it. Then look at the efficiency of the increment on its own, not the blended account number. The clean way to isolate it is a campaign experiment: change budget only in the treatment arm, so you can separate your change from ordinary week-to-week movement.
You are looking for how fast cost climbs as volume grows. If the increment holds near your target, the demand is there and scaling is real. If the increment costs far more than the blend, you have found the edge of your profitable volume, and “more conversions” past that point means very few conversions at much worse efficiency.
Why This Is Happening Now, And What It Signals
The stated reason is predictability. Advertisers raising budgets and watching efficiency move unpredictably is a real problem that generates a lot of support tickets. But this mechanic has behaved the same way for years. The question is not why Google would fix it. It is why now.
My read is that the target has become load-bearing in a way it was not before.
Google has spent two years building a planning and automation layer on top of bidding. Budget recommendations, performance forecasts, the results page, AI Brief, conversational campaign creation, a growing amount of agentic tooling: all of it depends on being able to project what happens when you change something. You cannot build reliable projections on a system with two behaviors. A bidding system that follows your target when uncapped and optimizes toward something else when throttled is not forecastable. Making the target binding in both states is a requirement for everything built on top of it.
That also explains why the language has been so heavy through this rollout. As execution moves further behind automated surfaces, the target is close to the last real control advertisers hold. A control that is not actually binding is not much of a control, and Google needs it to be one before the layer above it can be trusted.
There is a commercial read too, worth stating. Campaigns skimming cheap conversions under a cap buy less inventory than they could. Google has been explicit that the change does not alter spend, and for a campaign that stays capped, that is true. The mechanic is neutral. The guidance shipped alongside it is not: remove budget limits, capture demand you are missing, treat the cap as the thing between you and growth. Both things can be true at once. The engineering problem is real, and the fix also happens to favor the company selling the inventory.
The Tool Contradicts The Principle
Google’s own position, stated repeatedly through this rollout, is that your target should reflect your business goals. Profit, margin, customer growth, the numbers your leadership team reviews and believes. I agree with that completely. It is the same argument I have made for years about connecting automated bidding to real business outcomes rather than platform metrics.
Then look at what shipped. The Bid Target Adjustment Tool applies your trailing 28-day performance as your new target, in one click, in bulk. Those are two different instructions. One says your target should express your economics. The other says your target should describe your recent behavior. And by Google’s own account, that recent behavior is an artifact of bid suppression under a cap. So the default path takes a number produced by throttling, installs it as your efficiency goal, and calls it alignment with your business objectives.
If your margin math and your suppressed CPA happen to match, fine. That is a coincidence worth confirming, not a setting worth inheriting.
Doing Nothing Is Not A Neutral Choice
Most coverage frames this as advertisers losing a little overperformance. That undersells it, and the arithmetic is simple enough to do in your head.
Take a campaign with a $10 Target CPA, a $1,000 daily budget, and an actual CPA of $5. Today that buys 200 conversions. Google’s stated claim is that after the update, your actual moves toward the target you set. If it goes all the way to $10, the same $1,000 buys 100.
In practice it lands near the target, not always exactly on it, so treat that as the worst case. But the direction is fixed: as much as half your conversions, for the same spend, triggered by changing nothing. Leaving it alone is exactly what produces that outcome.
This is not a scaling question and it does not require you to raise a budget. It lands on any budget-limited account with a loose target that has been quietly beating it. The most exposed advertisers set a target long ago, watched it outperform, and never went back because there was no reason to. The reason arrives on August 17.
One warning that follows directly from the marginal argument: the tempting move is to dodge this by raising budgets. On a budget-capped campaign that is over-performing, adding budget releases the very cushion you are trying to protect. You uncap into the expensive part of the curve. The cleaner defense is anchoring the target to a number you can actually defend.
What To Actually Do
In order, and the last step is for after the rollout.
Filter to the campaigns this actually affects. Budget-limited, by lost impression share to budget rather than by spend, and meaningfully over-performing. Everything else is do-nothing.
Work out the target from margin, not from history. What CPA or ROAS does the business need to make money at the volume it wants? That number lives in the finance function, and it is the only defensible answer.
If you cannot get that number in time, treat recent performance as a provisional floor, and label it as such. Applying the trailing figure is a reasonable holding position, not a decision, and it should not be recorded as one.
Test marginal efficiency before you uncap. Increments of 20% to 30%, two conversion cycles per step, a campaign experiment to isolate your change from Google’s. Read the increment, not the blend.
Write down what you expect to happen, before the 17th. When performance shifts in September you want your prediction on record, not a reconstruction. Cheapest step on the list and the one most people skip.
After the rollout, check whether cost moved or quality moved. If CPCs rise while conversion rate holds, the auction got more expensive and you are paying more for the same traffic. If CPCs hold while conversion rate falls, the system expanded into weaker demand and your landing pages are absorbing visitors they were not built for. The blended CPA on your dashboard will not tell you which. Check them separately in the first two conversion cycles.
One Note For Anyone Automating This
If you run any automated or agentic budget allocation, this change raises the stakes on the input rather than lowering them. An agent moving budget on blended efficiency makes exactly the mistake above, faster and with more conviction than a person would. Marginal efficiency belongs in the objective function before automation is allowed to move money. That is also the whole point of the load-bearing target: Google is rebuilding the bidding system so its own automation layer can trust the number, and your automation needs the same clean input for the same reason.
The Question Worth Asking Instead
The industry conversation is about what target to set before the deadline. That is a settings question with a settings answer, and the tool will hand you one. The better question is what your next conversion costs. Answer that and the target follows from it. Skip it and you are picking a number because it was already on the screen.
Google is right that budget caps were never meant to be an efficiency lever. It is also asking you to replace that lever with a number your budget cap produced. The advertisers who come out of August in good shape will be the ones who knew their margin, and knew the shape of their own demand curve, before the tool offered them an alternative. The rest will keep reporting the skim as the yield.
