There is no right number for the target box

One week after the change I have had time to reflect and see what is happening. There are complaints about campaigns tanking, budgets not spending, and lower quality leads. Personally I have seen a lot of fluctuation in CPAs and I am hoping it will settle down, but I fear the step up in CPA is going to be the new normal.

What changed

On 17 August 2026, Google changed how bidding works for campaigns that are limited by budget and use a target-based bid strategy such as Target CPA or Target ROAS. The change applies to Search, Shopping, Performance Max, Demand Gen and Travel campaigns. Display and Hotel campaigns already worked this way. App and video campaigns are excluded.

Before the change, a budget-limited campaign with a target behaved like a bargain hunter. The system bought the cheapest conversions available first, and the budget usually ran out before the average cost ever approached the target. Google's own example: a campaign with a Target CPA of $10 that was actually converting at $5. The target sat in the settings doing nothing. The budget was the real constraint.

After the change, the target means what it says. Google's help documentation states that budget-limited campaigns "will more consistently perform toward your bid target." In the same example, the campaign "will deliver more closely to a $10 actual CPA." The target has changed from a ceiling into a setpoint. The system now delivers to it from both sides.

Why this matters: the same budget buys less

Cost per acquisition is simple arithmetic. CPA equals budget divided by conversions. If the budget stays fixed and the CPA rises from $5 toward $10, the number of conversions falls. Same money, higher prices, fewer results. Spend does not change. What changes is what the spend buys.

This is why the framing of "more consistent and predictable performance" deserves scrutiny. The performance becomes predictable at the level you typed into the box, which for an overachieving campaign is a worse level than the one you were getting.

The obvious fix, and why it fails

Google's recommended response is a tool called the Bid Target Adjustment Tool. It offers to lower your target to match your recent actual performance. Target was $10, actual was $5, so set the target to $5 and nothing changes. That is the advice most of the industry repeated.

It fails for two reasons.

First, the $5 was an average produced by a system that was free to hunt below it. Cheap auction days, quiet weeks, seasonal dips: the old system followed prices down and those savings accrued to you. Set $5 as a target under the new rules and the system delivers to $5 from both sides. When auctions dip cheaper, you no longer follow them down. The drift that produced your $5 is the exact behaviour you have switched off. The target now blocks the mechanism that created the number you set it to.

Second, the $5 average came from skimming the cheapest inventory. The marginal conversion, the last one your budget bought, cost more than $5. Meanwhile every other affected advertiser is repricing upward at the same time, because their campaigns are also being walked toward their targets. Auction prices inflate. A $5 target in an inflating auction starts to bind, the campaign underdelivers, and budget goes unspent.

So a target tuned to your actuals caps your upside and keeps your downside. Every branch of that decision lands at $5 or worse.

There is no correct number

Put the two failure modes together. Leave the target loose and the system spends your budget at worse efficiency, and volume falls. Tighten it to your actuals and you freeze yesterday's cream-skimmed average into a moving auction: unable to do better, still able to do worse.

A theoretical right answer exists: your marginal CPA at the point of budget exhaustion. That number is not shown anywhere in the interface, it moves every day with auction conditions, and being wrong in either direction costs you. A number you cannot observe and cannot hold is not a number you can type into a box.

The conclusion is that the box itself is the problem, for one specific group of advertisers.

No account has one CPA

Everything so far treats the account as if it has a single cost per acquisition. Real accounts do not. Performance varies across campaigns, across channels, and across days, and that variance is where a target does its worst work.

Take a live example. A travel account running three campaign types side by side: over the last seven days, Demand Gen converting at $23, Search at around $32, Performance Max at $40. The account's trailing average sits at roughly $40 for the month.

Now set a single target at that average, which is exactly what the adjustment tool and the trailing data will suggest. The Performance Max campaign delivering at $40 keeps delivering at $40. The Search campaign at $32 gets walked up to $40. The Demand Gen campaign at $23, the best thing in the account, gets walked up to $40. A target can never pull anything down below the number, and after 17 August it pulls everything beneath the number up to it. A target set from an average locks your worst campaign's economics onto your best campaigns.

The same logic runs through time instead of across campaigns. Auction prices fluctuate. Some days conversions are available at $20, some days at $50. That fluctuation is usually described as a problem the target will solve, and it is the opposite: the fluctuation is cheap days existing, and the ability to buy them is worth money. On expensive days the target stops you spending, and that volume never comes back. On cheap days the system now delivers at your number instead of the market's. An uncapped campaign harvests the cheap tail of the distribution automatically. A target is the act of selling that option back to Google for nothing.

Putting it to a Google rep

On a routine account review call, a Google rep suggested adding a Target CPA to the Demand Gen campaign above, to control its fluctuating cost per conversion. So the argument got tested live. Screen shared, one question asked: you name the number. What target goes in the box?

The rep checked the account data and suggested around $40, the trailing average. Which would walk the $23 campaign to $40, the $32 campaign to $40, and leave the $40 campaign at $40. When that was pointed out, the rep conceded the mechanics in full: whatever had been described was correct, "it works like this only." Then came the more revealing admission. Targets get suggested because some advertisers are afraid of cost per acquisition rising, so reps are expected to raise the option. The suggestion is a script written for advertisers who fear an uncapped number, and it answers the fear rather than the maths.

Nothing about that account is special. The same spread, with a trailing average sitting above the best performers, exists in almost every multi-campaign account, so the same demonstration works almost everywhere. Ask anyone recommending a target to commit to a specific number, then hold that number against your best campaign's current CPA. The recommendation refutes itself in the time it takes to read one column.

The optimisation argument

Here is the core of it, and it takes four sentences.

Call the budget B. Maximise Conversions is asked to get the most conversions possible while spending B. A Target CPA adds a second constraint to that same problem. A constraint can never improve the optimum of the problem it is added to: at best it does nothing, at worst it removes auctions you would have won and the volume goes with them.

So the decision comes down to one question. Is your real objective to get the maximum output from a fixed budget? If yes, the target should come off. Switch to Maximise Conversions, or Maximise Conversion Value for revenue-based accounts, and let the budget do the constraining. That is what a budget is for. If your real objective is an efficiency number, then CPA is your true constraint, the budget was never the binding one, and you have a different problem with a different answer.

This is Google's own documented option. The help page lists switching to Maximise Conversions or Maximise Conversion Value as a way to "capture the highest volume of conversions or conversion value for your set budget." It appears fourth in a list of five, after three variations of adjusting the number, and it is the only option printed with a warning attached.

Honest limits of the argument

Three concessions keep this claim honest.

The argument is about the objective, and it says nothing about next Tuesday. Google's bidder is a learning system. Removing a target triggers a settling period of days to weeks, and performance during that window proves nothing in either direction. Judge the change after the system has resettled.

The proof is a single-period argument and the system is a learner, so what it observes shapes what it can do next. The effect should be small when the budget binds, because both strategies buy from the top of the same ranked list of auctions, but it cannot be shown to be exactly zero.

And a target can be a legitimate stop loss. If your business genuinely cannot survive the top end of the CPA range that an uncapped campaign might produce, or if you are contractually held to a CPA number, then CPA is a hard constraint for you and keeping a target is defensible. The August change makes that insurance more expensive than it used to be, but for some businesses insurance is worth buying. The same applies to campaigns that are only intermittently limited by budget: on the days the budget has headroom, a target is doing real work.

None of these change the core answer for the advertiser whose budget genuinely binds and whose goal is maximum output. They define who that advertiser is.

Why the right answer stayed buried

No conspiracy is required. The incentives explain it.

For Google, a budget-limited campaign pays the same regardless of bidding strategy, so the money is somewhere else: the next budget conversation. An account with no target is capped at its budget, and the only way to grow it is to send more money and hope. An account with a target and headroom scales on rails: spend rises with auction prices, with query volume, and with every recommendation that says more conversions are available at your target. The entire help page funnels toward "confidently increase your budget." The change converts budget-constrained accounts into candidates for target-constrained accounts, and target-constrained is the posture where Google holds the growth lever.

For agencies and consultants, "remove the target" is self-erasing advice. Nothing to tend, nothing to optimise, no number to report against. Client accountability is denominated in CPA targets, and deleting the target moves accountability onto the budget, which the client owns.

For the industry conversation, "here is the number to set before the deadline" makes an actionable post. "The box should be empty" makes a worse one, so it lost the distribution war.

And underneath all of it, loss aversion. A cap that has never once bound still feels like protection, and removing it feels like risk. Nobody gets fired for keeping a decorative safety rail.

What to do

Pull every campaign with a Limited by Budget status running Target CPA or Target ROAS. For each one, answer the question: is the objective maximum output from this budget?

If yes, remove the target. Switch to Maximise Conversions or Maximise Conversion Value, expect a settling period of one to two conversion cycles, and judge the result after it. If the resulting CPA is genuinely unacceptable, that reaction is information: it means CPA was your real constraint all along, and you should manage it as one.

If no, because a contract, cashflow, or genuine tail risk makes CPA a hard limit, keep a target and set it deliberately as the constraint it now is, knowing it will bind. Or reduce your budget and use that to limit the CPAs.

Portfolio bid strategies and shared budgets are in scope, and changes there happen at the portfolio or shared-budget level. Multi-channel campaigns such as Performance Max may also shift how traffic distributes across channels, which is worth watching separately.

The one thing not to do is nothing. Any target is bad if you are working with a fixed budget and over-performing. Most advertisers are in that category and should remove the targets.

17 August: Google converted a setting into a job · Budget is the limiter · Targets from unit economics