Target ROAS vs Target CPA: When to Use Which (and When to Use Neither)

"which bid strategy should i use" is the wrong question, and most of the guides ranking for it answer a version of the question that stopped being true on august 17, 2026. a target in google ads is no longer a number the system tries to beat. it is the number the system delivers. this guide covers when target cpa fits, when target roas fits, when neither does, and how to set a target you can defend.

---

what target roas and target cpa actually are

both are conversion-based smart bidding strategies that set a per-auction bid to hit an average outcome across the campaign. target cpa aims at an average cost per conversion. target roas aims at an average conversion value divided by cost. neither is a cap on any single auction; both are averages the system steers toward over the conversion cycle.

what most guides get wrong is treating them as two flavors of the same thing. they are not, because they consume different data:

• target cpa needs conversions. every conversion counts as one. the system learns how likely a click is to convert and bids accordingly.
• target roas needs conversions with values, and the values need to vary. if every conversion is worth the same amount, the system has nothing to optimize toward beyond count, and target roas degrades into a less stable target cpa. if values are missing on a share of conversions, the system learns that some clicks are worth zero.

the four related strategies, as google labels them since june 2026:

the labels matter because they were confusing for years. from 2021 until june 2026, target cpa and target roas were shown as "maximize conversions with a target cpa" and "maximize conversion value with a target roas." google restored the standalone names in june 2026 and stated that the underlying bidding behavior is exactly the same (ppc land, june 2026). same math, cleaner menu.

---

the fundamental shift: a target is now the price, not the goal

before august 17, 2026: a campaign limited by budget on a target strategy would often beat its target. set a $10 target cpa on a campaign that could only spend $200 a day, and the system would frequently deliver $5 or $6, because the budget ran out before it had to reach for expensive conversions. advertisers learned to set loose targets and let the budget do the constraining.

after: google changed its bidding systems so that budget-limited campaigns on target cpa, target roas, and target cpc (demand gen) "perform more consistently toward your bid target, including when you make budget adjustments." google's own example: "if your campaign's target cpa is $10, but your recent actual cpa performance is $5, your campaign now delivers more closely to a $10 actual cpa" (google ads help, changes to target based bid strategies). the change applies to search, shopping, performance max, demand gen, display, hotel, and travel campaigns. app, video reach, and video view campaigns are exempt. google shipped a bid target adjustment tool on july 6, 2026 that shows recent actual performance, the current target, and a recommended target, and it stated it would not adjust any target automatically (optmyzr, august 2026).

the practical meaning: a target you never updated is now an instruction. if your target cpa is $40 because that was the number three years ago and you have been getting $22, you will get $40. ppc hero's summary is the right one: "targets are now reservation prices, not aspirations. the system delivers what you state, so state what you mean" (ppc hero, august 2026).

---

what this broke in existing accounts

1. loose targets that were doing nothing became targets that cost money

any budget-limited campaign carrying a stale, generous target saw cpa rise or roas fall toward the stated number in the weeks after august 17. nothing else changed. the target had been decorative; now it is enforced.

2. "set a high target roas so it stays efficient" flipped

on a budget-limited campaign, a high target roas used to be a safe way to keep spend efficient while the budget did the pacing. now a target roas above what the campaign can actually reach means the system bids down to hit it, spend drops, and the campaign shows as under-delivering with a clean roas nobody asked for. google's target roas page already warned that a high target may prevent the campaign from spending its budget (google ads help, about target roas bidding). after august 17 that warning applies with force.

3. target roas on campaigns without value variance

lead gen accounts that assigned one static value to every form fill and switched to target roas got a strategy that behaves like target cpa with more noise. worse, accounts sending values on some conversions and none on others taught the system that a large share of leads are worth nothing.

4. targets set from the dashboard rather than from reconciled economics

a target cpa derived from last month's platform-reported cpa embeds last month's attribution model, conversion window, and modeling. a target roas derived from platform-reported value embeds whatever value your tag sent, which for many accounts is order value before refunds, discounts, and cancellations. the target the system now enforces is a number that was never tied to margin.

5. constant target edits that reset learning

google lists four triggers for the learning status: a new or reactivated strategy, a setting change on the strategy, a composition change (campaigns, ad groups, or keywords added or removed), and ad group target changes on shopping campaigns (google ads help, learning period). a target edit is a setting change. accounts that "tuned" the target weekly never left learning.

6. portfolio strategies with bid caps nobody remembered

portfolio target strategies allow maximum bid limits. accounts inherited from a prior agency often carry a max cpc cap on a portfolio target roas set during a 2022 cpc environment. the system is being asked to hit a roas it cannot reach at the bids it is allowed to make, and the result looks like a demand problem.

---

the data

google's own eligibility floors. for target roas, google requires at least 15 conversions in the past 30 days at the conversion tracking level for search and shopping campaigns, 15 conversions with valid values for display, 30 in 30 days for video action, 50 in the past 35 days with 10 in the past 7 for demand gen, and 50 per week for hotel and travel. it also advises waiting one to two conversion cycles after you begin reporting new conversion values before adopting the strategy. (google ads help, about target roas bidding.) for target cpa, google recommends measuring performance over the last 30 days "including at least 30 conversions," and its recommended target is the average cpa of the last 30 days adjusted for conversion delay (google ads help, about target cpa bidding). note that 15 is an eligibility floor, not a performance floor.

the optmyzr dataset. optmyzr analyzed 14,584 google ads accounts, each at least 90 days old with conversion tracking configured and monthly spend between $1,500 and $5 million, published september 24, 2024 (optmyzr, the impact of bidding strategies on google ads performance). the findings that matter here:

• "the threshold for any bidding strategy to be predictably successful is 50+ conversions" in 30 days, with performance improving meaningfully above 25.
• maximize conversion value outperformed maximize conversions on roas, cpa, and cpc.
• advertisers using maximize conversions without a target saw better roas and cpa than those who set a target. advertisers using maximize conversion value without a target showed improved results on all metrics versus those with a target.
• manual cpc delivered strong performance on some metrics but struggled on cpa, ctr, and conversion rate.
• 48.42% of accounts ran multiple bidding strategies at once.

read the third bullet carefully. it does not say targets are bad. it says the population of advertisers who set targets set them badly often enough that the no-target group won on average. after august 17, 2026, the cost of a bad target went up, not down.

data-driven attribution volume. google's attribution guidance recommends at least 200 conversions and 2,000 ad interactions within 30 days for data-driven attribution to perform well, while stating all conversion actions are eligible regardless of volume (google ads help, about data-driven attribution). since target strategies bid on attributed conversions, a thin-volume campaign is running a target on top of an attribution model that is also short on data.

---

what changed in 2026 specifically

june 2026: standalone "target cpa" and "target roas" labels returned. no behavior change. existing campaigns were relabeled.

july 6, 2026: bid target adjustment tool launched in accounts. it shows recent actuals, the current target, and a recommended target with one "apply" action.

august 17, 2026: budget-limited campaigns on target strategies deliver to the target. display and hotel already behaved this way; search, shopping, performance max, demand gen, and travel joined them. google warned of "temporary performance and traffic fluctuations" during the transition.

ongoing: google's stated remedy for advertisers who liked the old overperformance is to raise budgets, not to loosen targets. that is a revenue-positive recommendation for google. it is only correct for you if the marginal conversions at the target are profitable, which is a reconciled-economics question, not a dashboard question.

---

what actually works now

1. pick the strategy from the data you have, not from the outcome you want

• fewer than 25 conversions in 30 days: no target strategy. maximize conversions with no target, or consolidate campaigns until volume clears 25 to 50, or move the optimization event one step up-funnel where volume exists.
• 25 to 50 conversions, values flat or absent: target cpa, set from reconciled cost per qualified outcome, or maximize conversions with no target on a fixed budget.
• 50 or more conversions with real, varied values (ecommerce with a range of order values, lead gen with value rules tied to qualification or close rate): target roas or maximize conversion value.
• values exist but are unreliable (order value without refunds, static lead values): fix the values first. see enhanced conversions and google ads attribution setup.

2. decide whether the campaign is budget-limited before you decide on a target

if the campaign will always spend its full budget, a target is now a throttle on volume, not a guardrail on efficiency. for a fixed budget where you want every conversion the budget can buy, maximize conversions or maximize conversion value with no target is the honest configuration. google's documentation says these strategies are "designed to spend the full daily budget, and are limited by budget by design" (google ads help, maximize conversions). if instead you have more budget than efficient demand and want the system to stop when the marginal conversion gets too expensive, a target is the right tool and the budget should be set above what it will spend.

3. set the target from reconciled numbers

target cpa: the cost per outcome your finance model says you can pay, translated into the platform's conversion definition. if platform conversions overcount qualified leads by 30% (compare crm-qualified leads to platform conversions over 90 days), the platform target is 70% of the true allowable cpa. target roas: contribution margin divided by allowable acquisition cost, expressed as revenue divided by spend, adjusted for the gap between platform-reported value and collected revenue after refunds and cancellations. for subscription businesses use the subscription attribution guide to get first-payment value versus ltv straight.

4. move targets in steps of 10 to 20% and wait a cycle

every target edit is a setting change and restarts learning. change once, wait one to two conversion cycles, evaluate on 30 or more conversions, then change again. the august change made the target binding, so a 40% jump is not a test, it is a new price.

5. use portfolio strategies for volume pooling, not for control

a portfolio target cpa or target roas across several low-volume campaigns pools their conversions for learning. that is the legitimate use. audit any bid limits on the portfolio annually and remove caps set in a different cpc environment. demand gen campaigns do not support portfolio bidding.

6. keep manual cpc for the cases it still fits

brand terms where you want a specific position, campaigns whose conversions cannot be tracked in the platform at all, and short flighted campaigns that will never reach learning. optmyzr's data shows manual holds up on some metrics and loses on cpa and conversion rate; treat it as a deliberate exception with a written reason.

7. after august 17, audit every "limited by budget" campaign with a target

pull 30 to 90 days before and after august 17. compare actual cpa or roas to the stated target. where actuals were better than target and you liked them, either lower the target to match (accepting the learning restart) or raise the budget and keep the target. where actuals were worse, the target was already aspirational and nothing changed except the label.

---

what to stop doing immediately

• stop leaving a target on a campaign whose target you cannot explain.
• stop using target roas on campaigns where every conversion carries the same value.
• stop setting targets from last month's platform-reported cpa or roas without reconciling to crm or revenue.
• stop editing targets weekly. each edit restarts learning.
• stop treating a high target roas as a safe efficiency setting on a budget-limited campaign.
• stop ignoring max bid limits on inherited portfolio strategies.
• stop running any target strategy on a campaign with fewer than 25 conversions in 30 days.
• stop "testing" targets by changing them 40% at a time.

---

the orbit agent instructions

{
  "guide": "target-roas-vs-target-cpa",
  "agent_tasks": {
    "audit_conversion_volume_floor": {
      "check": "for every active campaign on target cpa or target roas, count conversions in the last 30 days at the campaign level and, for target roas, count conversions that carried a nonzero value",
      "flag": "fewer than 25 conversions in 30 days on any target strategy, or a target roas campaign where more than 10% of conversions have no value",
      "recommend": "move the campaign to maximize conversions or maximize conversion value with no target, consolidate it into a portfolio strategy to pool volume, or shift the optimization event up-funnel until volume clears 25 to 50"
    },
    "audit_value_variance_for_target_roas": {
      "check": "for every target roas campaign, compute the distribution of conversion values over 30 days: min, median, max, and the share of conversions at a single identical value",
      "flag": "more than 80% of conversions at one identical value, or value present on fewer than 90% of conversions",
      "recommend": "either implement value rules or dynamic values tied to order value or lead quality, or switch the campaign to target cpa until values are real"
    },
    "audit_budget_limited_target_campaigns": {
      "check": "list campaigns marked limited by budget that run a target strategy, and compare actual cpa or roas for the 30 days before and after august 17, 2026 against the stated target",
      "flag": "any budget-limited target campaign whose actuals were more than 15% better than target before august 17 and have moved toward the target since",
      "recommend": "present two options with the numbers: lower the target to recent actuals and accept a learning restart, or raise the budget and keep the target if the marginal conversions are profitable"
    },
    "audit_target_vs_reconciled_economics": {
      "check": "compare each campaign's target to the allowable cpa or required roas derived from the brand's playbook goals or from revenue and qualified-lead data where connected",
      "flag": "a target cpa more than 20% above allowable cpa, or a target roas more than 20% below the break-even roas implied by margin",
      "recommend": "propose the reconciled target, state the formula used, and stage the change in steps of no more than 20% per conversion cycle"
    },
    "audit_learning_restart_frequency": {
      "check": "from change history, count bid strategy setting changes, target edits, and campaign or ad group composition changes per campaign over the last 60 days",
      "flag": "more than two learning-triggering changes in any 14-day window on a campaign with fewer than 100 conversions per month",
      "recommend": "freeze bid strategy settings for one to two conversion cycles and evaluate on at least 30 conversions before the next edit"
    },
    "audit_portfolio_bid_limits": {
      "check": "list portfolio bid strategies with maximum or minimum bid limits, the date each limit was set, and the current average cpc of the campaigns inside",
      "flag": "a maximum bid limit below 1.5x the current average cpc, or any limit older than 12 months without a documented reason",
      "recommend": "remove or raise the limit, document the reason if it stays, and re-evaluate the strategy after one conversion cycle"
    }
  }
}

---

the bottom line

target cpa when you have 25 to 50 conversions a month and no reliable value. target roas when you have 50 or more conversions with values that actually vary. neither when the campaign will spend its whole budget anyway; maximize conversions or value without a target is the honest setting there.

since august 17, 2026 a target is the number you get, not the number you hope to beat. set it from reconciled economics, change it in small steps, and give the system a full conversion cycle before you judge it.

---

built by orbit - the performance marketing intelligence layer. questions? jose@orbitllm.com