Target CPA and Target ROAS: How to Choose, Change and Judge a Google Ads Target When the Numbers Disagree

Last verified November 2026

A Target CPA or Target ROAS looks like one number. It is really three, and they rarely agree: what your business can afford per conversion, what the campaign has actually achieved, and what you type into the box. Type a number that ignores the first, and Google will diligently buy you conversions that lose money. Ignore the second, and the campaign can go quiet. And whatever you type, you won't know whether it worked until today's clicks have finished converting, which can take days or weeks.

This guide is for anyone who sets targets that decide real spend, whether you are about to set your first or have arrived because one "stopped working." It shows you how to line the three numbers up, choose a target the business can live with and the campaign can reach, change it without fooling yourself, and find what is really limiting a target that has gone wrong. It covers Search, Shopping and Performance Max campaigns, with standard and portfolio strategies, as Google Ads works in October 2026. Since June, the interface labels the strategies simply "Target CPA" and "Target ROAS," no longer "Maximize conversions with a Target CPA" and its value twin; bidding didn't change. In August, Google changed how budget-limited target campaigns bid.

The short version. If you read nothing else, read these seven lines; the rest of the guide shows why each holds.

1. Put three numbers in the same unit before you type anything: what the business can afford per counted conversion (or the lowest ROAS it can accept on the value you report), what the campaign has achieved on matured data, and the target itself.
2. Type a number between achieved and affordable, with a margin on the business side (step 5 of "How to choose"). Near achieved holds efficiency; nearer affordable buys volume, if the campaign has room to spend more.
3. If achieved is worse than affordable, no target can keep today's volume and meet the floor (the affordable number). Set the floor and accept lower, possibly near-zero, volume; change what the campaign sells, buys or measures; lower the floor explicitly; or spend less.
4. Read results on the conversion clock. A conversion cycle is the usual click-to-conversion time, including upload time for imported conversions. Judge a change after two cycles to settle, a window of two cycles, 14 days or 30 conversions (50 for Target ROAS), whichever takes longest, and one more cycle, against Avg. target CPA or Avg. target ROAS for the same matured dates, not against today's setting.
5. Let the clock pace your changes. Space changes one to two cycles apart, unless a business event (a price or offer change, a sale starting or ending, a measurement incident) requires otherwise. Changes on that spacing are judged together. To judge one alone, hold the next until that one is judged.
6. If the campaign is "Limited by budget," decide about the budget before the number. The decision is in "Target campaigns that are Limited by budget" below.
7. When the goal, the values or the catalog change underneath, re-price the target in the new unit, as part of the same change.

Already in trouble? A target that stopped spending or keeps missing: "Target CPA or ROAS not working?" A "Limited by budget" status: "Target campaigns that are Limited by budget." New conversion goals, values or products: "When conversions, values or products change under a target." More volume or more efficiency from a target that works: "How far and how often to change a target." A setting or limit: "Reference."

How a target works: a price, three numbers and a clock

Start with what the target actually instructs. With Target CPA, Google sets each bid from how likely that ad is to produce a conversion, aiming to get as many conversions as it can within your budget at the average cost you set. Some conversions cost more than the target and some cost less (Google: About Target CPA bidding). Target ROAS works the same way with value: Google predicts what each conversion is likely to be worth and tries to keep your average conversion value per dollar equal to the target (Google: About Target ROAS bidding).

Notice what the bidder controls: which auctions to enter and how much to bid in them. That gives a target one lever. A stricter target, meaning a lower CPA or a higher ROAS, makes Google pass on auctions it would otherwise have entered. Google says a CPA target set too low can forgo clicks that would have converted, leaving you with fewer conversions, and that a ROAS target set too high can limit your traffic. A looser target lets Google into more auctions and buys more volume at a worse average.

So the number in the box is an input to Google's bidding, not a promise about your results. Google's own page says your actual CPA depends on factors outside its control, such as your site, your ads and the competition in your auctions. A target moves the campaign along whatever trade between volume and efficiency the market offers. It cannot make that market better. Ask for a price the market rarely offers, and the campaign buys less, and usually not as much cheaper as the number asks. (Our story Who promised the $60 lead? follows what happens when a target number gets mistaken for a promise.)

That is why one number is never enough. Three numbers decide whether a target can work, and they mean something only when all three are in the same unit: the same conversion action, the same definition of value, the same dates.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Our answer on restarting a campaign after a 15-day pause uses the same three-way comparison to tell a too-strict target from a campaign that simply isn't serving.

To keep the arithmetic concrete, this guide follows one made-up account through about seven months. Kestrel sells refurbished office chairs online and takes bulk-order quote requests from businesses. An average chair order is worth $360. Kestrel can spend at most $72 in advertising per order and still keep the profit it needs; you'll see that calculation shortly. $360 of value for $72 of cost is a 500% return on ad spend, so 500% is Kestrel's floor for its chair campaigns. Over its last eight weeks of matured data, its Shopping campaign achieved 610%. Its Performance Max campaign achieved 380%.

Figure 1. Same floor, two campaigns. Shopping has room for a target between what the business needs and what the campaign has achieved. Performance Max has no number that is both affordable and reachable at the volume it has today.

The figure is the whole model in one picture. Shopping has a band where a target can go: anywhere in it is affordable, and the campaign has shown it can get there. The band starts at the floor itself; step 5 of "How to choose" takes a margin out of it, so the target you type sits a little inside the floor. Where in the band is a choice between volume and efficiency. Performance Max has no band. Any target Kestrel can afford asks for a return the campaign has never delivered, and any target the campaign has reached is one Kestrel can't afford. A number typed into the box can trade Performance Max's volume for efficiency, but it can't give Kestrel both.

Then there is the clock. Google says the bidder reacts to a new target within minutes, but it can take one to two conversion cycles before results reflect it, because conversions arrive late (Google: How to make target adjustments). A conversion cycle is the time it takes for a click to become a conversion; if you import conversions, it includes the time they take to reach your account (Google: How our bidding algorithms learn). Google's own example of the bid strategy report, a 2020 screenshot, shows what that looks like: a recent date with an estimated 60% of its conversions received, and 99% expected five days later. Judged on the first day, that campaign would have looked like it was failing. It was waiting.

The clock has a second consequence. If the target changed during a period, the number Google was working toward is no longer the one in the box. The columns that record what it was actually working toward are Avg. target CPA and Avg. target ROAS: traffic-weighted averages of every target that applied in the date range. Google recommends comparing your results with that average, not with today's setting. You add these columns to the Campaigns table from the Performance category.

One limit sits outside the three numbers: the budget. When a campaign is "Limited by budget," its daily budget runs out before it has bought everything its target would allow. Since August 2026, Google has made budget-limited target campaigns perform more consistently toward their target, so campaigns that used to beat their target now move toward it (Google: Changes to target-based bid strategies). That changes which decision comes first, and "Target campaigns that are Limited by budget" below walks you through it. Our essay What a Target Does When the Money Runs Out explains the reasoning behind it.

With the model in hand, you can predict four things the rest of this guide relies on:

• A target stricter than anything the campaign has achieved will buy less, and usually not as much cheaper as the number asks.
• A target change shows up in bids almost at once, and in results one to two conversion cycles later.
• The same typed number means different things in different units: 500% on revenue and 500% on profit are different targets.
• When achieved is worse than affordable, no number delivers both the current volume and the floor.

Should this campaign have a target? Target CPA, Target ROAS or neither

Before choosing a number, decide whether the campaign should have one, and which kind.

A target or no target. Use Target CPA or Target ROAS when an average cost per conversion, or an average return, is the limit you will defend, and you accept that the campaign may spend less than its budget when it can't find enough conversions at that price. Use Maximize conversions or Maximize conversion value when spending the budget for the most conversions or value matters more, and the business accepts that CPA or ROAS will move with the budget. That is the division in Google's own strategy table (Google: About Target ROAS bidding). Either way, a target is an average: it does not cap the cost of any one conversion or click.

CPA or ROAS. Choose Target ROAS when all of these hold:

• the conversions you count carry meaningfully different values (a $150 chair and a $900 chair, not identical form fills);
• those values reach Google Ads accurately, in the basis you will price against (revenue or profit);
• the campaign meets Google's eligibility (below);
• the values have been reported for the period Google advises before you adopt Target ROAS ("When conversions, values or products change under a target" covers it).

Otherwise use Target CPA on one stage of the funnel, after checking that the campaign type offers it.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Sources: About Target CPA bidding, About Target ROAS bidding, Automated bidding for Shopping, Target ROAS for Shopping.

Low volume doesn't forbid a target; it delays your ability to judge one. Google says you can start Target CPA with no conversion history, and calls it effective for campaigns of all sizes. Judging is a different matter: Google recommends evaluating results over periods with at least 30 conversions, typically a month or longer, or 50 for Target ROAS (Google: About Smart Bidding). This guide calls that the evaluation minimum. The monthly thresholds you'll find in practitioner guides are those authors' recommendations; for Target CPA, Google sets no minimum. With thin volume you can still use a target; you will read it later and less precisely.

Kestrel's choices follow from the table. Its chairs sell for very different prices and order values reach Google Ads, so Shopping and Performance Max bid on value, toward Target ROAS; Target CPA isn't among the strategies Google lists for Standard Shopping in any case. Its quote requests carry no value until a sale closes weeks later, so the quote campaign bids on one counted action, toward Target CPA. In every case Kestrel is defending an average, so a target fits better than Maximize.

How to choose your starting Target CPA or ROAS

Use this section when you are setting a first target, moving a campaign off Maximize conversions or Maximize conversion value, or re-setting a target after something changed. It is the hardest decision in the job, because the three numbers usually disagree: what the business can afford, what the campaign has achieved, and the target you are about to type. Google will also offer a recommended target, which turns out to be a reading of achieved rather than a fourth opinion (step 4). The seven steps below put everything in one unit and end with a number, or with the honest conclusion that no number will work yet.

If the campaign shows "Limited by budget," do steps 1 to 4, then read "Target campaigns that are Limited by budget" before step 5.

Step 1: Confirm what the campaign is counting

Every number below is "per counted conversion," so first establish what is counted. In the campaign, open Settings → Conversion goals and look inside the selected goals at the actual conversion actions. Bidding optimizes toward the actions that land in the campaign's Conversions column, and that can include more than you expect: a selected custom goal can include actions marked Secondary (our answer on making SQL the Primary conversion shows how). Then check three things:

• Each conversion counted once. If the same order or lead can be counted twice, history flatters you: CPA looks lower and ROAS higher than they are, and a target set from that history will be too strict for the real campaign.
• Imports arriving. If conversions come from a CRM or another system, confirm they are still reaching Google Ads.
• What "value" means. For Target ROAS, find out whether the conversion value is revenue, revenue after discounts, or profit. The floor you calculate in step 2 must use the same basis.

You're done when you can say, in one sentence, what one conversion is and what its value represents. Kestrel's Shopping and Performance Max campaigns count one online purchase, valued at order revenue excluding tax and shipping. Its quote campaign counts one submitted quote request, with no value.

If you can't write that sentence, repair the count first, or choose an earlier action you can count dependably for now (our answer on form fills and CRM sales as Primary conversions covers that choice). Pricing a target before then is setting a price for something you haven't defined.

Step 2: Price the conversion the campaign actually counts

This is the affordable number, and it comes from your business, not from Google Ads.

For an online store, start with a realistic order: a typical basket and product mix, not your best-margin item. Subtract the costs that follow it, including goods, fulfillment, payment fees, shipping you absorb and an allowance for returns. What remains is the order's contribution before advertising. Decide how much of that the business needs to keep; the rest is what you can spend to win the order. Then turn it into the ROAS the campaign needs:

Allowable advertising cost per order = contribution before advertising − contribution you need to keep

ROAS needed = value reported per order ÷ allowable advertising cost × 100

That is the method in our guide to planning Google Ads before a sale, which also works through what a discount does to it. If your conversion value is contribution rather than revenue, put contribution in the second line instead; the same economics then give a much lower percentage.

Kestrel keeps a third of each order's contribution for overhead and profit. On its average chair order, that works out as follows.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Keep both numbers. 500% is the floor. 333% is break-even: below it, every order costs more in advertising than it contributes, before any overhead or profit.

For leads, price the counted action from the sale it leads to. Take the advertising cost the business can carry per won sale and multiply it by the share of counted leads that become sales, measured on leads old enough to have finished their sales process. That is the approach in our answer about a $335 click after fixing conversion tracking. Work out the advertising cost per won sale the same way as for an order: contribution before advertising, minus what the business keeps. Break-even works the same way too: contribution per won sale, keeping nothing, times the same share.

A won bulk order contributes $900 to Kestrel before advertising. Keeping a third leaves $600 for advertising, and one in five quote requests becomes an order:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

If you don't yet know what share of your leads becomes sales, you don't yet have a defensible number for the target; you have a guess, and you should treat it as one. "If you can't fill in a number yet," at the end of this section, says what to do.

You're done when you have one affordable number per campaign, in the unit that campaign bids on, and its break-even beside it.

Step 3: Measure what the campaign has achieved, on matured data

Now the history. Three choices decide whether you can trust it.

Where the window ends. Leave out the most recent stretch whose conversions are still arriving. Google's example: with a two-day conversion delay, look at 28 days and leave out the last 2 (Google: How our bidding algorithms learn). The bid strategy report shows your conversion delay where data is available (Google: What the bid strategy report shows); open it from Tools → Budgets and bidding → Bid strategies, through the link in the Bid strategy type column (Google: Finding your bid strategy report). Where it doesn't, segment conversions by Days to conversion (the segment icon, then Conversions) over a date range that ended at least 30 days ago (Google: Find out how long it takes customers to convert). If you import conversions and your delay figure ends on the sale date in your system, add your upload days. That is the conversion cycle this guide's timing rules use.

What the window covers. Use a period comparable to the one you are pricing for: the same counted action and value basis, a similar price level and the same structure. Leave out periods covered by a data exclusion, a seasonality adjustment or Smart Bidding Exploration, since each changes what bidding was told. For a campaign that is limited by budget, use only data from after August 27, 2026, when Google's change finished rolling out. Data from before a change like these describes a different campaign. "All-time or the last 90 days?" usually has the same answer: neither, the period since things last changed.

How much it must contain. Aim for the evaluation minimum (30 conversions, or 50 for Target ROAS). A window with fewer is still evidence, but treat the achieved number as provisional and give your margin in step 5 more room.

Then read the number. For ROAS, add the Conv. value / cost column (it's among the Conversions columns) and multiply by 100 (Google: About Target ROAS bidding). For CPA, use Cost / conv.

You're done when you have the achieved number and the dates it covers. Kestrel's chair purchases mostly arrive within five days of the click, so its window is the eight weeks ending five days ago:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Step 4: Read Google's recommendation for what it is

Google will usually offer a target, and it is worth knowing where that number comes from. The recommended Target CPA is your average CPA over the last 30 days, adjusted for conversion delay (Google: About Target CPA bidding). The recommended Target ROAS is based on your actual ROAS over the last few weeks, leaving out the last few days (Google: About Target ROAS bidding).

In other words, the recommendation is your own recent history, handed back to you as advice. It is a good description of last month. It has never seen your invoices.

So use it as a second reading of step 3, never as the affordable number. If it is close to your achieved figure, it confirms your window. If it is far from it, find out why before you use either; a different date range, the delay adjustment or a recent change are the usual reasons.

Step 5: Compare the numbers and choose

Every target goes a margin inside the floor, so start with the margin, then sort the campaign into one of three situations.

Put the margin on the business side. A target is an average, and even an unchanged target produces a wandering result. In smec's data on 600 Performance Max campaigns, the middle half of campaigns whose target ROAS did not change still saw actual ROAS move by roughly 10% in either direction, and the outer range ran to about 39% (smec: Target ROAS changes FAQ). So don't set the target exactly at the floor, where much of that wandering lands below it. Set it a margin inside:

• for ROAS, the floor × (1 + margin);
• for CPA, the affordable cost ÷ (1 + margin).

The best margin is your own campaign's normal swing: compare matured actual results with Avg. target across three or four consecutive windows of equal length, and use the largest gap you saw. Without that record, use 10% if each window would hold the evaluation minimum, and 20% if it wouldn't. Both are our judgment. They are informed by smec's data, where the typical wander was about 10%, the outer range far wider, and campaigns with more than 60 conversions a month moved more steadily. Replace them with your own swing as soon as you have three windows.

Note where the margin goes: it protects the floor, not your history. A common rule of thumb says to open 10–20% looser than history. It never asks what the business can afford, so when history sits close to break-even it can set a target that loses money on every sale.

Now find your situation. "Better" means a higher ROAS or a lower CPA than the floor.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Then choose within the band. A number near the achieved end holds the efficiency you have; typing achieved itself is what Google calls adjusting your target in line with recent performance. A number near the floor-with-margin end lets Google into more auctions and buys more volume at a lower average return. Choose that end only if the business wants the volume, the campaign is not limited by budget, and there is room for it to spend more. Where a target simulator is available, it estimates conversions at different targets from the last seven days of data (Google: About simulators). You'll find it in the bid strategy report's Simulators section, or through the simulator icon in the Campaigns table's Budget column.

Here is Kestrel's choice, in month 0:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

What Kestrel did not do matters as much. It did not type 610% for Shopping because that's what the campaign "does," and it did not type 500% because that's the floor. Each would have been a decision made by one number while ignoring the other.

Step 6: When achieved is worse than affordable

This is the case most advice skips, and it's where a target does the most damage if you treat it as a dial. Kestrel's Performance Max campaign needs 500% and has achieved 380%.

First, check that the comparison is valid. Is the window matured? Are the counted action and value the same as in step 1? Could the achieved figure be distorted by device adjustments, ad-group targets or Smart Bidding Exploration (see "Reference")? Then place the result against break-even. At 380%, Kestrel spends about $95 in advertising per $360 order. Performance Max is above break-even: it contributes about $13 per order toward overhead and profit, less than the $36 Kestrel needs.

You have four honest options. Only the first uses the target field, and it trades volume away to do it.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Below break-even the choice narrows. If matured, correctly measured results are below break-even, don't keep buying at the current target: set the floor and accept that delivery may collapse, restructure, or pause, unless the business has written down a funded reason to accept the loss and how much of it.

Two tempting moves are missing from that table. The first is typing 380% "to keep the volume coming." That tells Google to buy more of exactly the sales the business has said it can't afford. The second is re-running the report over every possible date range until one of them shows 500%. That is the reporting version of asking the other parent. The answer you finally get has the right number in it, but it describes a campaign you don't have.

Kestrel sets the floor. It types 500% on Performance Max and changes nothing else. "Learning status and when to judge a new target" shows how review dates are set, and what the review found.

You're done when each campaign has either a target with its reasons, or a written decision about why it has none yet.

Step 7: Make the change and book the review

Where you make the change depends on the strategy:

• A standard Search or Shopping campaign: Campaigns, select the campaign, then Settings; enter the new target and click Save. To test the new target against the old one rather than swap it in, choose Save as experiment where it's offered (Google describes experiments for Search and Display; check whether your Shopping campaign shows it).
• A portfolio strategy: Tools → Budgets and bidding → Bid strategies. A new target there applies to every campaign and ad group using the portfolio. You can't change an existing portfolio's type, from Target CPA to Target ROAS for example; that takes a new or different portfolio.
• Performance Max: the campaign's Bidding section. While the June labels settle in, Google's own page shows both forms: "Target CPA" or "Target ROAS," and a "Set a target cost per action" or "Set a target return on ad spend" checkbox under Conversions or Conversion value.

Sources: Edit your target CPA, Create a portfolio bid strategy, Performance Max bidding.

Then record five things for each campaign: the target you typed, the affordable number and how you calculated it, the achieved number and its dates, today's date, and a review date. The next section shows how to set the review date from your own conversion delay.

The record does two jobs. It stops a red dashboard on day three from triggering a change the clock hasn't earned. And it shows the next person why the number is what it is, so a target set from a floor isn't mistaken for a number someone copied from last month.

If you can't fill in a number yet

If you don't know what you can afford, usually because you don't yet know what share of leads becomes sales, don't type a target. Set a spending limit you can live with, keep the campaign untargeted, and gather outcome data until enough leads have finished their sales process to measure the share; our $335-click answer shows how to bound that spend. Then return to step 2.

If you know what you can afford but have no usable achieved number (a new campaign, a recent tracking fix, a major catalog change), choose by what matters more:

• A hard average-cost limit matters more than volume: start at the floor-with-margin, using step 5's margin rule, and watch delivery from the first days.
• The business can afford its full budget at an uncertain CPA: start without a target for at least two conversion cycles to measure what the campaign achieves. (Google also suggests Maximize clicks or Manual CPC as ways to build conversion data.)

Then return to step 3 with the matured window, treating the result as provisional if it is small. The decision in step 5 is the same one; you are making it later, with the missing number filled in.

Learning status and when to judge a new target

You've changed a target, or switched a campaign onto one. Two questions follow: what does the Learning label mean, and when can you believe the numbers?

Learning is a reason, not a verdict. A bid strategy can show the status Learning, and hovering over it shows why: a new or reactivated strategy, a change to the conversion goals or actions the campaign optimizes to, or a change in the campaign's composition (Google: About bid strategy statuses). Read it as context for interpreting early results. It is not a sign that the change failed, not a sign that it worked, and not permission to ignore a spending limit or a campaign that isn't serving.

A target change never triggers a Learning status and never removes your historical data (Google: About Smart Bidding learning). The same page advises against multiple target changes within a single conversion cycle unless there is a clear business need. Nothing on that page supports the common rule that every target change restarts learning, so targets must creep by 10–15% every two weeks. What actually limits you is the clock.

Set the review date from the clock. First let performance settle on the new target for two conversion cycles; Google gives one to two. The window you judge starts the day after settling ends. It runs for the longest of two cycles, 14 days, or the days needed to reach the evaluation minimum. The review date is one conversion cycle after the window closes. If the window holds fewer conversions than the evaluation minimum on the review date, extend it by the days its own daily rate needs to cover the shortfall, and review one cycle after the new end.

Kestrel switched its Shopping campaign to a 550% Target ROAS in month 0. Its purchases mostly arrive within five days, so the change settles until day 10. At about 19 orders a week, the 50 conversions Google suggests take about 19 days, longer than two cycles or 14 days, so they set the window: days 11 to 29. Kestrel books its review for day 34, one cycle later. (Google's Shopping page separately suggests allowing about 15 days before evaluating a Shopping Target ROAS campaign; Kestrel's date satisfies both.)

On day three, the Shopping dashboard showed ROAS well below 550%. Nothing was wrong. The last few days' purchases hadn't arrived yet, so their cost was in the report and much of their value wasn't.

Read the result against the right number. On the review date, open the bid strategy report or the campaign table, set the date range to the matured window, and compare actual ROAS (or CPA) with Avg. target ROAS (or Avg. target CPA) for the same dates. If you changed the target partway through, the average reflects that; today's setting doesn't. Then make a second, separate comparison: actual against the floor. The first tells you whether Google delivered what you asked for. The second tells you whether what you asked for is worth buying.

What to do with the reading:

• Actual within your margin of Avg. target: leave it. That is a target working.
• Actual worse than Avg. target by more than the margin, two matured windows running: diagnose before you change anything ("Target CPA or ROAS not working?"). Two windows is our judgment, meant to avoid acting on one noisy reading.
• Actual better than Avg. target by more than the margin, and the campaign is not limited by budget: check the count first. Is each conversion still counted once with the same value, and is a bid limit or inventory limit shown? (These are checks 4 and 6 of "Target CPA or ROAS not working?".) If the count and value hold and no other limit is shown, the target isn't what's holding the campaign back; it found fewer auctions than the target would allow. Treat the result as a new achieved number and run step 5 again. Then choose:
  • If you want efficiency, move the target toward the result; Google says aligning a target with actual performance shouldn't cause noticeable volatility.
  • If you want volume, loosen within the band, and expect little extra volume from it (our judgment: a campaign already underspending at its target has shown how few extra auctions there are).
• The same, but the campaign is limited by budget: that's not a fault to diagnose. Go to "Target campaigns that are Limited by budget."
• Spending or stock limits reached, a broken checkout, or broken measurement: act at once. None of these waits for a window.

Until you have a matured window of reasonable size, read results as provisional. Watch spend and serving, fix defects you find, and leave the target alone.

Kestrel's reviews came back as follows.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Performance Max did what a stricter target does: it bought less, and got more efficient, but by less than the target asked. "When to drop the target" picks up what Kestrel did next.

How far and how often to change a target

The target is working, and the business wants something different: more volume, or better efficiency. Two decisions follow, how big a step and how soon the next one.

How far. Google sets no maximum. Its August FAQ says targets can be adjusted as business needs dictate, and that Smart Bidding reacts in real time to large or small changes (Google: FAQ about changes to target-based bid strategies). What Google warns about is the consequence: a major target change can have a similarly sized impact on spend or volume (Google: How to make target adjustments), and in rare cases a large change that moves the campaign into a new set of auctions brings volatility for a conversion cycle or two (Google: How our bidding algorithms learn). So size the step by the change in spend or volume the business can absorb, not by a percentage someone made standard.

Expect the result to be comparable, not proportional or predictable. In smec's Performance Max data, campaigns tracked target ROAS decreases closely, but actual ROAS lagged larger increases: a 30% target increase showed a median actual change near 12%. Kestrel's Performance Max floor test went the same way. Google's own pages describe raising volume by "gradually reducing" a target ROAS (for Target CPA, gradually raising it). The 10–15% step caps in many practitioner guides are those authors' practice, not a Google rule.

Before a large move, use the target simulator if your campaign has one; it is often unavailable for campaigns that reached their budget in the last week, use a shared budget, or ran an experiment recently (Google: About simulators).

How often. Give each change time to show in matured data before you make the next one: one to two conversion cycles. Google's general guidance on measuring Smart Bidding gives this for all four strategies after major bidding or budget changes (Google: Tips on measuring Smart Bidding performance). For Target ROAS on Search and Shopping, Google also advises against several target changes within one conversion cycle unless there is a true business need. No Google page sets a calendar cadence; "every two weeks" rules are practitioner habit. The real exceptions are business events: a price or offer change, the start or end of a sale, a measurement incident, a budget emergency.

That spacing allows the next change, but each change's review date comes later, so steps made on that spacing are judged together. To judge one by itself, make the next change after its review date, as our answer on when you can judge a change explains. A target changed every Monday belongs to a campaign permanently reporting on last week's argument.

Change one thing at a time. If you need to know what a change did, don't put a target change, a budget change, a restructure and a goal change in the same evaluation window. The exception runs the other way: when a change underneath alters what the target is measured in, re-price the target as part of that same change ("When conversions, values or products change under a target").

There is a limit to loosening. Don't loosen the target past the floor-with-margin to buy volume, whether directly or through something that lowers the effective target, such as Smart Bidding Exploration or positive device adjustments on Target CPA. Past that point, more volume has to come from somewhere else: more budget, but only where the campaign is actually limited by budget; otherwise the offer, the coverage or the structure, or an explicit business decision to lower the floor.

Short events. For a short, predictable and evidenced jump in conversion rate, ideally one to seven days, use a seasonality adjustment rather than a target edit (Google: About seasonality adjustments). If the event changes the economics of an order, such as a discount, re-price the floor and the target for exactly its duration and restore them when normal prices return; our before-the-sale guide works through both. Without evidence of a conversion-rate change, change neither.

In month 2, Kestrel's owner asks for more chair orders from Shopping. At 550% it is already at the floor-with-margin and isn't limited by budget, so no target or budget move can deliver them. The quote target, though, could loosen from $105 to about $109. Kestrel holds it at $105, because it plans to change what the quote campaign counts in month 4, and a step taken now would still be maturing when the count changes underneath it.

When to drop the target

Removing a target, by switching to Maximize conversions or Maximize conversion value, changes what the budget is asked to buy. It is the right move when using the whole budget matters more than the average cost, and the business accepts CPA or ROAS moving with the budget. Expect spend to rise to the budget if the campaign was underspending. Judge the switch on matured business outcomes per useful result, not on the platform's conversion count alone.

It is the wrong move when the target is missing the floor. Dropping the target changes the objective, not the economics: an untargeted campaign spends its budget at whatever return it finds. Our answer on dropping Target CPA after a CPA jump works through a case where removing the target was worth testing, and what had to be true for it.

That was Kestrel's position with Performance Max in month 2. At its 500% floor it had reached about 420%, on far less volume. Dropping the target would have let it spend its full budget at whatever return it found: the opposite of what Kestrel needs. So Kestrel takes step 6's second option, as a separate change. Its higher-margin chairs, averaging $420 an order with $150 of contribution, can carry more advertising than the average chair. Kestrel builds a new Performance Max campaign for them. In Performance Max the target and the budget belong to the campaign; asset groups are collections of creative, not separately bid units. So a separate campaign is what lets these chairs carry their own target (Google: About Performance Max campaigns; Google Ads API: bid strategies are set at campaign level). Keeping a third of their $150 contribution leaves $100 for advertising (step 2's method), so their floor is $420 ÷ $100 × 100 = 420% on revenue, or 504% with step 5's 20% margin for low-volume windows. The new campaign has no history of its own, so Kestrel starts it at 504%, as "If you can't fill in a number yet" says for a missing achieved number, and pauses the original campaign's remaining lower-margin chairs.

Target campaigns that are Limited by budget

Use this section when a target campaign shows "Limited by budget," or when your diagnosis of a target that isn't working has led you here. The decision comes first; what changed in August, and the details, follow it. The reasoning behind the decision is in our essay What a Target Does When the Money Runs Out; you can act without it.

In short: from August 17, 2026 (rollout complete August 27), Google made campaigns that are Limited by budget and use Target CPA or Target ROAS perform more consistently toward their target, including when the budget changes. Campaigns that used to beat their target now trend toward it (Google: Changes to target-based bid strategies). Google doesn't change your targets or budgets for you.

The decision

1. Confirm the budget is what's limiting. The campaign shows "Limited by budget," and on Search and Shopping its Search lost IS (budget) column (impression share columns are under Competitive metrics when you add columns) shows it losing a material share of impressions to budget (Google: Impression share). That column isn't established as a diagnostic for Performance Max, where the status is your evidence. Status can change from day to day, so check the whole period you're judging, not just today. Don't use the column with untargeted Maximize conversions or value, which spend the whole budget by design; for those, compare budget with spend.

2. Place matured actual (after August 27) against the floor and against Avg. target. These terms come from "How to choose your starting Target CPA or ROAS": matured windows from step 3, the floor from step 2, and the floor-with-margin from step 5.

• Worse than the floor: go to step 6 of "How to choose your starting Target CPA or ROAS." A bigger budget only buys more of what the business can't afford.
• Anything else, whether beating the target or missing it: continue. If it's missing, don't loosen the target to close the gap: in a budget-limited campaign, loosening raises the cost per conversion without adding spend.

3. Is the current target the business's real number? Run the floor method from step 2 of "How to choose." A target left loose for years may never have been checked against margins; a target at the floor-with-margin is the business's number, whatever the campaign was achieving.

4. Can the budget grow? Not "would more budget be nice," but whether the business can fund more spend when the campaign finds conversions at the target.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

The options in the table are Google's (Google: Changes to target-based bid strategies), with the floor method deciding between rows. When the budget can't grow, Google also lists consolidating campaigns under a portfolio strategy or a shared budget, and switching to Maximize conversions or Maximize conversion value if the business accepts CPA or ROAS moving with the budget (Google: FAQ).

At a fixed budget, choose by what the same money buys. While the campaign keeps spending its budget, conversions are the budget divided by the cost per conversion; for ROAS, conversion value is the budget times the ROAS. Google says a budget-limited campaign performs consistently toward its target. So, at a fixed budget:

• Keep a target looser than recent performance, and performance moves toward that target: the same spend buys fewer conversions, or less value. It's the right choice when you are about to be able to raise the budget, so that the target becomes the price of the extra volume, or when the business is content with the target's cost per conversion and wants nothing more from the money. That is our judgment.
• Match recent performance, and the money keeps buying what it buys now. Google says aligning a target with actual performance shouldn't cause noticeable volatility.
• Tighten past recent performance to get more out of the money, for example to win back the efficiency the campaign had before August. Google says you can set a more efficient target at any time, but that daily spend will likely fall. Tighten in steps, reading each on matured data. Each step pays off only while the campaign still spends close to its budget: once spend drops well below it, the budget is no longer the limit, the campaign isn't budget-limited any more, and the rest of this guide applies. Pre-August performance is where you're aiming, not evidence of what the campaign will achieve now.

Kestrel met this cell in month 3, its busiest month. Its Shopping campaign showed "Limited by budget" while achieving 640% on matured data from after August 27, well above its 550% target. The 550% is its real number, and the month's $300 daily budget couldn't grow. At 640%, $300 a day buys about $1,920 of sales; at 550%, the same $300 buys $1,650, so letting performance drift to the target would cost about $270 of sales a day. Kestrel moves the target to 640%, which still clears the floor-with-margin. When the month ends, the budget returns to normal, the campaign is no longer limited, and Kestrel restores 550%, as its record said it would.

Two cautions. Don't add data exclusions or new bid limits just because of this change; Google advises against both, though a genuine tracking outage still warrants an exclusion. And if you compare impression share across August 2026, don't read high pre-change loss to rank on a budget-limited campaign as evidence that its target was too strict; vendor data from smec suggests these campaigns were losing auctions to rank because their bids were held down before the change (Search Engine Roundtable).

What to watch afterwards

• Spend against budget, daily. A higher budget may not be spent in full; Google says not to expect that under what it calls demand-led budgeting. A stricter target will likely reduce daily spend.
• Matured actual against Avg. target, on the change's review date.
• The status itself. If "Limited by budget" disappears, the rest of this guide applies as normal.
• Cost per click and, in Performance Max, channel mix. smec's data showed median CPC rising for budget-limited campaigns after the change, and Google warns that Performance Max traffic may shift between channels. Judge both by matured cost per conversion, not by CPC alone.

Scope and structure. The change covers Search, Shopping and Performance Max (and Demand Gen and Travel, which this guide doesn't cover). Campaigns not limited by budget didn't change. For portfolio strategies and shared budgets, make adjustments at the portfolio or shared-budget level: in a portfolio with separate budgets only the constrained campaigns are affected, and with a constrained shared budget the effect spreads across all its campaigns. Campaigns on a campaign total budget are unaffected.

When conversions, values or products change under a target

A target is a price in a unit. Change the unit, and the price in the box becomes wrong even if nobody touches it. Re-price the target as part of the same change: leaving the old number is itself a mis-set target.

When the counted action changes. Moving from form fills to qualified leads, or from leads to sales, changes what one conversion is. Recompute the affordable number with the new stage's rate (step 2), and price the target from the new stage's own matured history, recorded before the swap wherever possible. Google's method for staying on Target CPA through such a move (Google: lower-funnel transition):

1. Make sure the lower-funnel action is categorized as Purchases, Qualified lead or Converted lead, and upload offline outcomes consistently, ideally daily.
2. If it's categorized as something else, recategorize it first. The tracking period below starts again after you do.
3. Mark the lower-funnel action Primary, but don't use it for bid optimization yet.
4. Track it for at least two to three conversion cycles; Smart Bidding trains on this in the background.
5. Swap the goals, and set the CPA target directly to the lower-funnel action's historically achieved CPA.
6. Monitor spend closely and adjust as needed.

After any goal change, Google expects Smart Bidding to need one to two conversion cycles to learn in most cases, and Performance Max may take longer (Google: Changing conversion goals). Judge on matured data from after the swap settles. Our answers on making SQL the Primary conversion and form fills and CRM sales cover the goal and import mechanics.

In month 4, Kestrel moves its quote campaign onto qualified quotes, which its sales team confirms within a day or two. One in two qualified quotes becomes an order, so the affordable number becomes $600 × 0.5 = $300 per qualified quote (break-even $450). In the preparation window the campaign spent $3,200 and recorded 15 qualified quotes: about $213 each. At the swap, Kestrel follows Google's step 5 and sets $213, treating it as provisional, since 15 conversions is below the evaluation minimum. The $105 still in the box would have asked for qualified quotes at about half what they had ever cost. Once two matured windows have come in, Kestrel re-runs step 5 with the wider low-volume margin: $300 ÷ 1.2 = $250, so the band runs from about $213 to $250.

When the value definition changes. Before adopting Target ROAS on new or changed values, Google advises reporting them for a period first. Its pages put that period differently, from 3 to 4 weeks or from 1 to 3 conversion cycles, each with "whichever is longer" (Google: About Target ROAS bidding; Google: Tips on measuring Smart Bidding performance). Google also advises against backfilling historical values. For a campaign already on Target ROAS, re-price the target into the new value basis at the moment the values change, then judge it on its review date.

In month 5, Kestrel starts reporting each chair order's contribution instead of its revenue, because margins vary by model. The floor in the new basis is $108 ÷ $72 × 100 = 150%, and 165% with the margin. That is the same economics as 550% on revenue, now expressed in profit. Left at 550%, the Shopping target would have asked for more than three times the return Kestrel needs. The new Performance Max campaign's 504% on revenue converts to 180%: the same 150% floor, with its 20% margin.

One floor serves every chair only because Kestrel keeps a fixed share of contribution, a third, rather than a fixed amount. If a business must keep the same dollars on every order, the contribution floor changes with each product's margin. Keeping $36 on every order, a chair contributing $150 would need only about 132%, while a chair contributing $60 would need 250%; a single 150% floor would let that cheaper chair keep $20 instead of $36. If your business keeps fixed amounts, give products whose floors differ their own campaigns and floors.

When conversion data was wrong or missing. If a tag failed, an upload broke or duplicates crept in, protect bidding with a data exclusion over the affected clicks, then adjust the target and budget as needed. Exclusions apply to clicks in a date range and the conversions attached to them, so cover the conversion delay; Google advises covering at least 90% of affected clicks, never removing an exclusion once applied, and avoiding frequent or prolonged use (Google: About data exclusions). Don't exclude results that were recorded accurately but were bad; a disappointing month is not a tracking outage. Late but accurate imports aren't an exclusion case either. One narrow exception is an accurately measured, now-fixed, one-off site or checkout failure, where excluding the affected clicks is a deliberate forecasting choice you should record. Our essay The First Fix Changes What Later Truth Can Repair explains what each kind of fix leaves behind.

When the catalog or structure changes. After a large catalog change, a new campaign split or a merger of campaigns, judge the new state on matured data from after the change settles. Compare it with a matured window from before, mapped to the same coverage. Our answer on revenue falling after merging campaigns shows how to map one to the other.

Target CPA or ROAS not working? Find what is actually limiting it

A target "stopped working" can mean it stopped spending, or that it keeps missing. The cause can sit in any of seven places, and the fix for one makes another worse: raising a CPA target won't help a campaign whose ads aren't serving, and raising its budget won't help a campaign whose target is too strict. So check in order, and fix each failure you find before moving to the next.

Three terms carry the checks. Matured results come from a date range old enough for its conversions to have arrived (step 3 of "How to choose" shows how to pick it). Avg. target is the column recording what the strategy actually aimed for over those dates. The floor is what the business can afford per conversion, from step 2.

First, stop any exposure the business can't tolerate. If continued spending would breach a limit, pause or reduce it now and diagnose afterward.

1. Serving and eligibility. Are the ads, ad groups, keywords or products actually eligible to show? A campaign marked Enabled can still contain a paused ad group, a disapproved ad or ineligible products. Our answer on restarting after a 15-day pause walks through these checks. A target can't spend money the campaign isn't allowed to show for.
2. What changed, and when. Select the campaign's checkbox on the Campaigns page, then Change history on the menu bar above the table; the changes appear on a timeline against performance (Google: Change history). Target, budget, goal, feed, site and tracking changes near the date things went wrong are leads, not verdicts.
3. Maturity. Is the period you're worried about old enough for its conversions to have arrived? If not, wait; see "Learning status and when to judge a new target."
4. The counted action and its value. Did what counts as a conversion, or what its value means, change? Duplicate counting, a broken tag or a stalled import changes the number the target is compared against. See "When conversions, values or products change under a target."
5. The site's conversion rate. Compare conversions per click in two matured windows, before and after things went wrong. If fewer clicks now become customers while cost per conversion also worsened, and the mix of traffic (queries, channels, devices) is comparable, look at the site rather than the bid: the landing page, checkout, prices, stock and delivery terms. A lower conversion rate alongside a steady or better cost per conversion is the bidding buying cheaper clicks, not a site problem (Google: Fix low traffic for Target CPA). Google lists changes to your website among the things outside its control that move your actual CPA, and a target can't bid its way past a page that stopped converting.
6. Which limit is active. The bid strategy status, and the reason shown when you hover over it, is the first place to look (Google: About bid strategy statuses).
   • The budget is the active limit when the campaign shows "Limited by budget" and, on Search and Shopping, Search lost IS (budget) is material. Make the decision in "Target campaigns that are Limited by budget."
   • The target is the likely limit when spend is below budget and matured actual sits at or worse than Avg. target, especially if Search lost IS (rank) rose after a tightening. Google's troubleshooting page says a target well below historical CPA may not be attainable while keeping reasonable traffic (Google: Fix low traffic for Target CPA).
   • Neither: check demand and inventory (a small market, out-of-stock products) and any bid limits, which show as "Limited (bid limits)" in the status.
7. Achieved against affordable. If the campaign is reaching its target but the target itself sits below the floor, or the campaign can't reach the floor at any target, the problem is the one in step 6 of "How to choose your starting Target CPA or ROAS."

When the target is the limit, loosen it, within bounds. Loosening is the right move when the campaign isn't budget-limited and the loosened number stays within the floor-with-margin. It is most clearly called for when the target is stricter than anything the campaign has achieved on matured data. After a data exclusion (see "When conversions, values or products change under a target"), Google recommends adjusting the target while performance settles: usually a few days, or one to two conversion cycles if a week or more of clicks were affected. Reverse that temporary adjustment once bids have recovered (Google: once the exclusion takes effect). Restoring a target to what the floor method supports, as Kestrel does in month 6, is not such an adjustment and is not reversed.

A diagnosis Kestrel needed. In month 6, after a site theme update, Kestrel's Shopping campaign reported a contribution ROAS of about 340% for two weeks, double its usual 170%. Someone took the result at face value and raised the target to 280%, and spend fell to a trickle. Working down the list: serving was fine; Change history showed the theme update a few days before the target edit; the window was mature. Check 4 found it: the theme update had installed a second purchase tag, counting every order twice and doubling its reported value. Kestrel fixed the tag and applied a data exclusion covering the affected clicks and their conversion delay. It then set the target back to the 165% its floor method supports. The target hadn't stopped working. It had been asked for a return that existed only in the duplicated data.

Two other symptoms have their own answers. A campaign suddenly spending twice its daily budget is covered in what to check when Google Ads starts spending twice its daily budget. A drop in reach after a goal change is covered in our SQL-Primary answer.

Reference: settings, limits and where to go instead

Settings that change what the target means

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Standard or portfolio

A portfolio strategy applies one target across several campaigns. Editing its target changes it for every campaign and ad group using it, and August-change adjustments for portfolios are made at the portfolio level. Performance Max can't use a portfolio strategy. Google's pages conflict on whether a portfolio Target ROAS is available for Standard Shopping, so check the bid-strategy options in your own account. For when pooling thin campaigns helps, see our answer on low-volume B2B Search campaigns.

Capping the cost of a click

A CPA or ROAS target can't cap a click; it is an average. Portfolio Target CPA and Target ROAS strategies offer bid limits, but Google doesn't recommend them: they apply only in Search Network auctions, the minimum isn't absolute, and Google's API documentation says they are disregarded for portfolio Maximize conversions and portfolio Maximize conversion value strategies on Search campaigns. If you set one, confirm it binds: the status shows "Limited (bid limits)." For a firm per-click limit, use a campaign type that offers Manual CPC (or Maximize clicks with a maximum CPC limit), with no positive bid adjustments and Search Partners excluded. Performance Max offers neither. Our answer on unwanted or expensive clicks in Performance Max sets one up.

Numbers to know

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Where things are in Google Ads

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Names

In June 2026, Google renamed "Maximize conversions with a Target CPA" to "Target CPA," and "Maximize conversion value with a Target ROAS" to "Target ROAS," in the Google Ads interface; Google describes the change for Search and similar campaign types. Bidding didn't change. Older labels can linger in the API, Google Ads Editor and the mobile app. The rename and the August change to budget-limited campaigns are separate events.

Where to go instead

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Your monthly target check

Run this for each target campaign once a month, and again after any change to its goal, values, catalog or budget. It takes the guide down to the decisions you will actually repeat.

1. The count. Settings → Conversion goals: the same actions as last month, each counted once, imports arriving, value in the same basis?
2. The floor. Has anything changed the order economics or the lead-to-sale rate: prices, costs, the close rate? If so, recompute the affordable number.
3. The window. Pick the latest matured window: ending before the conversion delay, after the last material change settles, holding the evaluation minimum if you can get it.
4. Two comparisons. Actual against Avg. target for that window: inside your margin, or outside it for a second window running, and in which direction? Then actual against the floor: still worth buying?
5. The budget status. Limited by budget now, or during the window? If so, make the budget decision.
6. Learning and changes. Any Learning status, and its reason? When was the last change, and has its review date passed?
7. The record. Write down what you found, what you changed and why, and the next review date.

Here is Kestrel's check in month 7, after everything above. All three campaigns now report contribution, or count qualified quotes.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Most months, the check ends where Kestrel's did: three campaigns left alone, for reasons someone can read. When it doesn't, the step that failed tells you which section to open.