Google publishes a number for Target CPA and almost nobody checks their account against it.
The number is 30. Google’s own Target CPA documentation recommends evaluating performance over the last 30 days including at least 30 conversions. Below that, the machine learning has too few examples to find a pattern, and you get bidding that swings around, budgets that behave strangely, and performance data you cannot trust enough to act on.
We ran our own accounts against that number expecting to find one or two offenders. We found seven out of ten.
This is the second piece built from a campaign-level export across the lead generation accounts we manage. The first looked at our PMax versus Search cost per lead analysis and found that one account with a broken conversion action was distorting the headline figure by 5x. This one looks at bidding, and it turns out the two problems are connected in a way we did not expect.
The dataset
We pulled a campaign-level report from our manager account on 24 July 2026, covering 1 June 2025 onwards, which works out at roughly 13.8 months. That gave us 36 campaigns across 13 US accounts, all lead generation, carrying $163,732.04 in spend and 6,616 platform-reported conversions. The businesses are mostly home services, meaning plumbing, fencing, iron doors, windows and doors, remodeling and tree removal, plus counselling, psychiatry, payroll services and a dental practice.
When we say conversions per month, we mean total campaign conversions divided by 13.8. It is the fairest approximation available over a period this long and it smooths out seasonality, but it cuts both ways: any campaign that launched partway through the window looks worse than it actually was. Where that matters to the argument, we say so.
One gap we are not hiding. Our manager account reporting shows more total spend than these 36 campaigns account for, so this is a subset rather than the complete book. Every number here refers to the $163,732 in the export. We work white label, so no client is named.
What the account was actually running
First, the spread of bid strategies across all 36 campaigns.
| Bid strategy | Campaigns | Spend | Conversions |
|---|---|---|---|
| Maximize Conversions | 18 | $59,346 | 1,242 |
| Target CPA | 10 | $91,029 | 5,288 |
| Maximize Clicks | 4 | $11,429 | 64 |
| Target Impression Share | 3 | $1,918 | 22 |
| Manual CPC | 1 | $10 | 0 |
Target CPA is on ten campaigns but carries $91,029, which is 56% of all spend. The most demanding strategy is running most of the money.
That is not automatically wrong. It is only wrong if those campaigns cannot feed it.
Seven of ten were below the threshold
Here is every Target CPA campaign in the set, sorted by conversions per month.
| Business | Campaign type | Spend | Conversions | Per month |
|---|---|---|---|---|
| Home remodeling | Search | $10,650 | 10 | 0.7 |
| Payroll services | Search | $3,614 | 28 | 2.0 |
| Tree removal | Search | $2,585 | 30 | 2.2 |
| Plumbing (water heater) | Search | $13,817 | 44 | 3.2 |
| Iron doors | Search | $10,114 | 51 | 3.7 |
| Fencing | Search | $3,039 | 58 | 4.2 |
| Counselling | Search | $11,479 | 314 | 22.8 |
| Plumbing (local PMax) | PMax | $4,781 | 533 | 38.6 |
| Plumbing (local PMax) | PMax | $12,046 | 635 | 46.0 |
| Plumbing (competitor PMax) | PMax | $18,904 | 3,586 | 259.9 |
Seven of ten sit below 30 conversions per month. Six sit below ten. One sits below one.
The six worst account for $43,819 of spend, or 27% of everything in this export, running on a bid strategy that Google’s own documentation says needs roughly ten times more data than they were producing.
The home remodeling campaign is the one that stings. Ten thousand six hundred and fifty dollars, thirteen months, ten conversions. Target CPA was being asked to learn a pattern from less than one example per month.
The part we did not expect
Look at which three campaigns cleared the threshold.
All three are Performance Max campaigns in the plumbing account. The same plumbing account that, as the first article in this series documented, had a phone number click set as its primary conversion.
Those campaigns were not producing 38, 46 and 260 leads a month. They were producing phone taps, counted as leads, at a volume high enough to clear Google’s data threshold.
So the honest summary of this dataset is not “seven of ten campaigns were under-fed.” It is worse than that. The only campaigns with enough conversion volume to justify Target CPA had that volume because they were counting the wrong thing.
Every campaign measuring something real was below the threshold. Every campaign above the threshold was measuring something that was not a lead.
That is an uncomfortable finding to publish about your own accounts. It is also the single most useful thing we learned from this exercise, because it means the two problems are the same problem. Weak conversion tracking does not just produce misleading reports. It produces campaigns that appear to qualify for advanced bidding strategies they have no business running.
The best legitimate performer in the set is the counselling account at 22.8 conversions per month. Real conversions, decent volume, and still short of 30.
What happens when you run Target CPA below the threshold
The failure mode is not obvious, which is why this persists in so many accounts.
The campaign does not stop. It does not throw an error. It spends its budget and reports numbers, and if you only look at the summary row it can appear to be working.
What actually happens:
Bidding becomes erratic. With a handful of conversions to learn from, the system cannot separate signal from noise. One good week produces aggressive bidding, one quiet week produces retreat. You see CPCs swing without any change on your side.
Delivery chokes if the target is wrong. This is the most common version we inherit. Somebody set a $60 target on a campaign whose real cost per lead was $140. With no data to argue otherwise, the system simply stops bidding on most auctions rather than exceed the target. Impression share collapses, spend drops, and the account looks “efficient” while doing almost nothing.
Every change resets the clock. Adjust the target, edit the budget, swap the ads, and you are back in a learning period. On a campaign producing two conversions a month, the learning period effectively never ends. The campaign lives permanently in a state of recalibration.
The data you use to diagnose it is unreliable. This is the trap. You look at a low-volume tCPA campaign, see a bad cost per lead, and start making changes. But at four conversions a month, the difference between a good month and a bad month is one or two leads. You are optimising against noise, and every optimisation restarts the learning.
It looks like a campaign problem when it is a structure problem. Which leads directly to the next article in this series, because the reason most of these campaigns had so few conversions was not that the accounts were small. It was that the accounts were split into too many campaigns.
We are not the only ones seeing this
The pattern we found in our own accounts is one that consultants who audit other people’s accounts describe constantly.
Boris Beceric, a Google Ads consultant and coach, was asked by Search Engine Land what advertisers get wrong with Smart Bidding. His answer was that most try it too early, without enough conversion volume, and that the usual fix is to consolidate campaigns so more data flows through a single campaign. He also points to portfolio bidding as the same idea applied at the bid strategy level, and to adding the micro conversion with the most volume and the closest relationship to a real conversion where volume genuinely cannot be found.
That last suggestion deserves a caveat given what we found in the first article of this series. Adding micro conversions to reach threshold is a legitimate technique, but only if you understand what you are teaching the system to buy. A phone tap is a micro conversion with plenty of volume and almost no relationship to a booked job. Choose the micro conversion carefully or you end up exactly where our plumbing account ended up.
Search Engine Land‘s own coverage of campaign structure describes a scenario close to ours: an account with twelve separate Search campaigns, one per product category, each averaging eight to twelve conversions a month, with Smart Bidding enabled across all of them and none consistently exiting the learning phase. Their conclusion is the same as ours. The bid strategy is not the problem. The number of containers is.
The consistent thread across practitioner commentary is that this is a structure problem wearing a bidding problem’s clothes, which is why the next article in this series is about campaign structure.
The sequence we use instead
We do not put a target on a campaign until the campaign has earned it. The progression is boring and it works.
Stage 1: Maximize Clicks, briefly
New campaign, no conversion history, nothing for any algorithm to learn from. Maximize Clicks buys traffic and generates the data that everything else depends on.
Two rules. Set a maximum CPC bid limit, because without one this strategy will happily buy the most expensive clicks available. And treat it as temporary. Maximize Clicks optimises for volume, not quality, and if you leave it running for six months you have bought a lot of traffic and learned very little about who converts.
In our export, four campaigns were still on Maximize Clicks, carrying $11,429 and producing 64 conversions between them. Three of those are plumbing service campaigns that have been running long enough to have graduated.
Stage 2: Maximize Conversions, until volume builds
Once conversions are arriving, move to Maximize Conversions without a target. The system optimises for conversion volume within the budget, and crucially it does not have a target to choke against. You get a real cost per lead rather than an aspirational one.
Stay here longer than feels comfortable. This is where the data that makes stage three possible gets built.
Stage 3: Add a target, once and only once you are above 30
When the campaign is consistently producing 30 or more conversions per month, and consistently is doing a lot of work in that sentence, you can add a target.
Set it at or slightly above your actual achieved cost per lead from the previous 30 days. Not your desired one. If your campaign is producing leads at $95 and the client wants $60, setting the target to $60 does not produce $60 leads. It produces almost no leads.
Then move it down in increments of 10 to 15% and wait a full learning period between changes. Practitioner consensus on this is fairly settled and it matches what we see: gradual reduction works, dramatic reduction chokes delivery.
The exception: portfolio bid strategies
If you genuinely need target-based bidding but no single campaign clears 30 conversions, a portfolio strategy pools conversion data across several campaigns and bids against the combined pool.
This is the right answer for accounts with several small campaigns serving one business goal. It is not a workaround for bad structure. If the campaigns should have been one campaign in the first place, consolidate them instead of pooling them.
How to run this check on your own account
Twenty minutes, and you can do it in the interface without an export.
Step 1. Campaigns view, last 30 days. Add the Bid strategy type column if it is not showing.
Step 2. Sort by Conversions, ascending.
Step 3. Read down the list. Any campaign showing Target CPA or Target ROAS with fewer than 30 conversions in that window is a candidate.
Step 4. Before you change anything, check the conversion column is telling the truth. Segment by Conversions > Conversion action and confirm those conversions are real leads rather than phone taps, page views or form starts. A campaign that appears to clear the threshold on soft conversions does not clear it.
Step 5. For each campaign below threshold, check impression share lost to rank. A tCPA campaign that is choking will show high lost impression share to rank while sitting well under budget. That combination is close to diagnostic.
Step 6. Move the under-fed campaigns to Maximize Conversions. Expect a wobble for a week or two while learning resets, and do not judge anything inside 30 days.
Step 7. Ask the structural question. If five campaigns are each producing four conversions a month, the bid strategy is not really the problem.
What this means for anyone reading benchmark data
Published cost per lead benchmarks, including the ones we are producing in this series, are built on platform-reported conversions from accounts whose bid strategy configuration nobody audited.
If a meaningful share of campaigns in any large sample are running target-based bidding below the volume it requires, some of those campaigns are choking and some are bidding erratically, and both distort the averages in ways nobody can see from the outside.
We are not arguing benchmarks are worthless. We are arguing that when your numbers look worse than published averages, the first question is whether your campaigns are structured to let any bid strategy work, not whether your bids are wrong.
In our own accounts, the answer was no more often than yes.
Methodology and limitations
This report covers 36 campaigns across 13 US lead generation accounts running between 1 June 2025 and 24 July 2026, representing $163,732.04 in spend and 6,616 platform-reported conversions.
Conversions per month means total conversions divided by 13.8 months. Campaigns that launched or paused partway through are understated by that method. We do not have reliable launch dates for every campaign and decided a stated approximation was more honest than adjusting some campaigns and not others.
The 30-conversion threshold comes from Google’s own Target CPA documentation, which recommends evaluating performance over the last 30 days including at least 30 conversions. Google separately allows the strategy to run below that, so nothing in this article describes a rule being broken. It describes a gap between what the platform permits and what the platform recommends, which is a gap most accounts fall into without anyone noticing.
The sample is small. Thirteen accounts, one of which carries over half the spend, all US, mostly home services in the southeast. Nothing here is an industry benchmark. It is a description of one agency’s accounts, published because the pattern turned out to be far more widespread than we expected and we doubt we are unusual in that.
Several of the campaigns above are mid-remediation as this is published. We will report what happened.
FAQ
What is the minimum conversion volume for Target CPA? Google’s documentation recommends at least 30 conversions in the last 30 days at campaign level for evaluation. The strategy can technically run below that, which is why so many accounts do.
What about Target ROAS? Higher. Practitioner consensus lands around 50 or more monthly conversions with real value data attached, because tROAS has to learn which conversions are worth more as well as which are likely, and that needs a bigger sample.
My campaign has 20 conversions a month. Should I use Target CPA? No. Use Maximize Conversions, or pool it with related campaigns in a portfolio strategy. Twenty is close enough to be tempting and far enough to be unstable.
Can I just count more things as conversions to get above 30? This is the trap our own data walked into. Adding form starts, page views or phone taps will get the number up and will teach the system to buy more of those, which is not what you want. If you need more volume, get it from structure or budget, not from a looser definition of success.
What is the fastest fix if I find campaigns below threshold? Switch them to Maximize Conversions today. It is reversible, it removes the choking risk immediately, and it costs you nothing but a learning period. Then work on the structural question.
How long before I can judge the change? Thirty days minimum. The learning period alone runs one to two weeks, and comparing the fortnight after to the fortnight before will tell you nothing useful.
Does this apply to Performance Max? Yes, and PMax makes it harder to see because it aggregates across placements. The three PMax campaigns in our set that cleared the threshold only cleared it on inflated conversions, which is exactly the failure this article is about.
About The Author
