The short version. Target CPA bidding lets you name a cost per conversion, and Google Ads sets bids automatically to hit it. It is sold as a promise: name your number, the machine holds it. Across $42M of real spend, that promise holds far less often than advertisers assume.
- Only 1 in 3 Target CPA campaigns (34%) actually lands within 10% of its stated target. 36% overshoot the target by more than 20%. The median campaign runs 10% above target. n = 1,950 campaigns.
- Target ROAS is leakier still: only 27% of campaigns hit or beat their ROAS target, with a median achievement of 86% of goal. Three in four fall short.
- The number you type is not a guarantee. It is the centre of a wide corridor. Plan your economics on the top of that corridor, not on the target.
Target-based bidding is still the right tool at scale. The failure is treating the target as a promise instead of a request the auction is free to miss.
What Target CPA bidding actually is
Target CPA is an automated bid strategy where you set a cost per acquisition you are willing to pay, and Google Ads adjusts bids in every auction to win the conversions it predicts will come in at or near that number. It replaces manual bidding with a model that reads signals you cannot see in real time: query, device, time, audience, and hundreds more.
The mechanism is genuinely powerful. It is also probabilistic. The system optimizes toward your target on average, across many auctions, using predicted conversion rates. When the prediction is good and the data is thick, it holds the line. When conversions are thin or their value varies, it drifts. The target does not break loudly. It erodes.
The promise gap: only one in three campaigns hits the target
The clearest way to test the promise is to compare each campaign’s actual cost per conversion against its own stated target. Across 1,950 campaigns and $42M in spend, only 34% landed within 10% of their target CPA. Update, July 2026: recomputed on a wider slice of the portfolio, 2,723 campaigns across the managed accounts, the on-target share came out at 26.8%, and the miss turned out to depend heavily on campaign type. See Performance Max overshoots its target CPA in 71% of campaigns. More than a third, 36%, overshot the target by more than 20%. The median campaign ran 10% above its target.
Read that as a forensic finding, not a complaint about automation. The target is not a fixed cost. It is the centre of a corridor that, in this data, runs roughly from 0.9 times the target to 1.4 times it. If your business only works at the target, you have built your economics on the middle of a distribution and ignored the top half. When those campaigns overshoot, the losses are invisible on a dashboard that reports the average.
Target ROAS is even leakier
If Target CPA is a wide corridor, Target ROAS is wider. Only 27% of Target ROAS campaigns actually reached or beat their target return. The median campaign delivered 86% of its goal, a 14% shortfall. Three campaigns in four came in under the return the advertiser was counting on.
The practical correction is direct. If your economics require a return of X, do not set the target at X. Set it near 1.15 times X, because the autobidder systematically undershoots. Setting the target at the number you actually need guarantees you will typically miss it.
Profit Forensics
I examine one week of your Google Ads account and find the money it is losing, or prove it is clean. Signed, either way. For accounts spending $50,000 or more per month.
Update, September 2026: the January to May 2026 check
We repeated the test on a newer slice, now published in our open dataset on Hugging Face: campaigns with an active target and at least five conversions between January and May 2026, measured against the targets in a July 2026 snapshot. Of 239 Target CPA campaigns, 33.5% came in more than 20% above target and 47.3% at or under it; the median campaign landed at 1.03 times its target. Of 64 Target ROAS campaigns, 31.3% missed by more than 20% and 56.3% hit or beat the target.
The advertiser view is calmer. Collapse each advertiser to the median of its own campaigns and 60.9% of Target CPA advertisers landed at or under target, with only 8.7% missing by more than 20%. Summarising each advertiser by its median campaign absorbs large campaign-level swings. Both numbers are true at once, and they answer different questions: whether a given campaign will hold its target, and whether an account as a whole will.
One limit carries over. Targets can move inside the window and the export holds no target history, so a campaign whose target changed in March is measured against its July value. The newer slice also keeps only campaigns with five or more conversions, which leans toward campaigns that work at all.
Why the autobidder misses, and when it is not the algorithm’s fault
Three causes account for most of the drift, and none of them is fixed by switching the strategy off.
- Thin conversion data. The model needs volume to predict well. A campaign with a handful of conversions a month gives it noise, and it chases the noise. This is the legitimate case for pooling into a portfolio strategy, covered in the companion piece on portfolio bid strategies.
- Conversion values that vary. If a conversion sometimes means a $40 lead and sometimes a $4,000 sale, a single CPA target averages them and misprices both.
- A target set from history, not economics. Copying last quarter’s comfortable CPA into the target field sets a wish. The auction does not care what you paid last quarter.
| Strategy | Campaigns that hit the target | Typical miss | How to set the target |
|---|---|---|---|
| Target CPA | 34% within 10% | 36% overshoot by 20%+ | From conversion value; plan on 1.4x the target |
| Target ROAS | 27% hit or beat | Median 14% short | Set near 1.15x the return you actually need |
How to set a Target CPA that survives the auction
- Derive the target from what a conversion is worth, not from last month’s average. Start from margin and close rate, work back to the most you can pay and still profit, and set the target below that.
- Plan your economics on the upper edge of the corridor. If the target is $100, budget as if many campaigns will run at $140. If that still works, the target is safe. If it does not, the target is too high.
- Feed the model clean, consistent conversions. One conversion action, one definition, values that reflect real worth. Garbage signals produce garbage bids.
- Give it enough volume, or pool it. If a campaign cannot support a stable target alone, combine it with others that share the same economics.
- Stop resetting learning. Every target change and strategy swap restarts the learning phase. Set it deliberately, then leave it alone long enough to judge.
When this does not apply
Your campaign changed its target during the window. A target moved mid-period makes the comparison between goal and outcome meaningless; the figures here assume the target that was in place.
You have very low conversion volume. With a handful of conversions a month, the distance between target and actual is mostly noise, and reading it as a miss will send you chasing randomness.
You care about profit rather than about hitting the number. A campaign that overshoots its target CPA while buying profitable volume is not failing at anything that matters; the target is an instruction to the bidder, not a definition of success.
Key takeaways
- Only 34% of Target CPA campaigns land within 10% of their target; 36% overshoot by more than 20%.
- Target ROAS is leakier: 27% hit their target, with a typical 14% shortfall.
- The target is the centre of a corridor, not a guarantee. Plan on the upper bound.
- For Target ROAS, set the goal near 1.15x the return you actually need.
- Most misses come from thin data, variable conversion values, or targets copied from history. Fix those before blaming the algorithm.
Frequently asked questions
Why is my Target CPA higher than the target I set?
Because the target is a request, not a cap. In real accounts, only a third of campaigns land within 10% of their target and over a third overshoot by more than 20%. Common causes are thin conversion data, conversion values that vary, and a target set from history rather than economics.
What is a good Target CPA to start with?
Work back from economics. Take your margin and close rate, find the most you can pay per conversion and still profit, then set the target below that. Do not copy last quarter’s average, and budget as if campaigns will run above the target, because many will.
Is Target CPA or Target ROAS more reliable?
Target CPA holds its target more often. In this data, 34% of Target CPA campaigns landed within 10% of target, versus 27% of Target ROAS campaigns hitting their goal. If you use Target ROAS, set the target near 1.15 times the return you actually need to offset the systematic shortfall.
How much conversion data does Target CPA need?
Enough that the model predicts from signal, not noise. A campaign with only a few conversions a month will drift. If a single campaign cannot support a stable target, pool it with campaigns that share the same economics using a portfolio strategy.
Method and sources
Doctor Ads Profit Forensics is the research desk of a Google Ads practice with 19 years and $770M managed spend behind it.
The September 2026 update uses the published open dataset (Hugging Face ivitskiy/google-ads-benchmark-2026, files smart_bidding_promise_gap and smart_bidding_promise_gap_advertiser_level): January to May 2026, targets from a July 2026 snapshot, campaigns with an active target and at least five conversions; advertiser-level figures take the median of each advertiser’s campaigns (239 Target CPA campaigns from 23 advertisers, 64 Target ROAS campaigns from 15).
The proprietary figures here come from a forensic analysis of Google Ads accounts representing $133M in managed spend, September 2024 to February 2025. The Target CPA figures cover 1,950 campaigns and $42M in spend; the Target ROAS figures cover 622 campaigns and $16.9M. Each campaign’s actual cost or return is measured against its own stated target, so the ratio is self-normalized and comparable across accounts. Related reading: portfolio bid strategies and bid adjustments.
