tCPA vs tROAS: which should you use?
The short answer
tROAS only works when the value data feeding it can be trusted, since it optimizes toward a revenue number the algorithm has to believe. tCPA fits accounts with a fairly uniform average order value, where cost per action is a reasonable stand-in for value. Below roughly $100 a day in spend, neither target gets enough conversion signal to learn from, so an underperforming test at that budget is usually a false negative, not proof the strategy failed.
tCPA and tROAS are both automated bidding targets, and the choice between them comes down to what your data can support and what you’re actually optimizing for.
What each target optimizes for
tCPA (target cost per acquisition) tells Google’s bidding system to get as many conversions as possible at roughly the cost per conversion you set. It treats every conversion as equally valuable, because it has no other value signal to work with.
tROAS (target return on ad spend) tells the system to get as much conversion value as possible relative to a return target you set. It needs a real value passed back on every conversion… order revenue, not just a fired pixel… or it’s optimizing against noise.
| tCPA | tROAS | |
|---|---|---|
| Optimizes for | Volume of conversions at a cost target | Conversion value relative to a return target |
| Needs | Any conversion signal | Accurate, consistent value data on every conversion |
| Fits best | Fairly uniform average order value | Wide variance in order value, where not all conversions are worth the same |
| Breaks when | Order values vary a lot and volume optimizes for the wrong orders | Value data is missing, duplicated, or wrong on a meaningful share of conversions |
The decision rule
The real question is whether your value data is good enough to trust. If order values swing widely and you’re tracking them accurately, tROAS should outperform, because it’s optimizing for the thing you actually care about. If order values are fairly uniform, or your value data has gaps, tCPA is the more honest choice, since there’s nothing for tROAS to gain by chasing value differences that aren’t reliably there. This is worth deciding deliberately, since both targets can technically run on the same account, and defaulting to whichever one launched first is how good accounts end up optimizing for the wrong thing for months.
The budget floor that neither fixes
Below roughly $100 a day in spend, this decision matters less than it seems, because neither target gets enough conversion volume to learn from. Automated bidding needs a steady flow of signal to find a stable pattern, and a thin account produces too little of it. A campaign that “doesn’t work” at that budget is often just under-signaled, not a bad fit for the target you picked… the fix is more budget or a longer test window, not switching targets again.
Targets are constraints, not goals
The most common mistake with either target is treating the number you set as a wish instead of a constraint. A tCPA or tROAS target should describe the floor or ceiling you can tolerate, with the algorithm free to maximize volume or value inside it, not a number you set once and defend regardless of what the account is telling you. When performance allows it, tightening the target should follow, not lead.
What changed in August 2026
Google’s August 17, 2026 bidding update changed how budget-constrained campaigns behave against these targets: instead of drifting past a stated tCPA or tROAS when a campaign is capped by budget, the system now converges toward the target itself, per coverage from Optmyzr and Keyweo. That makes the target a more reliable constraint on budget-limited campaigns than it used to be, and it’s one more reason to treat the number as a boundary rather than a goal.
Decision table by account condition
| Account condition | Better fit | Why |
|---|---|---|
| Uniform AOV, reliable conversions | tCPA | Value data adds little; volume at a cost target is the simpler, more stable goal |
| Wide AOV variance, clean value tracking | tROAS | The algorithm can meaningfully tell a good conversion from a mediocre one |
| Under roughly $100/day spend | Neither, yet | Not enough signal for either target to learn reliably |
| Budget-constrained, target set as a floor or ceiling | Either, tightened gradually | Post-August-2026, budget-constrained campaigns converge toward the target instead of drifting past it |
Rise
Our own doctrine, across every account we manage, is to maximize conversion value subject to a ROAS floor rather than chase a target for its own sake. If you’re not sure which target fits your account’s data, that’s the kind of question a teardown is built to answer.