What should your Target ROAS be if your margin is 30 percent? The answer most people give is 333 percent, and as arithmetic it is correct. As a target it is a floor: the lowest return at which a credited dollar stops losing money, if the margin is real and every credited dollar is incremental. Both of those assumptions usually fail in the same direction, and since August a budget-limited campaign is delivered to whatever number you type.
Google is testing a way to compute that number for you. A paid search specialist, Arpan Banerjee, spotted it on October 6, and Search Engine Land and Techwyse both wrote it up. As reported, you enter a campaign's average profit margin before ad costs, or answer a few questions to estimate one, and the tool suggests a Target ROAS along with weekly estimates of clicks, revenue, spend and profit. Google has not disclosed how it gets from the margin to the target. Search Engine Land's example, a 15 percent margin giving a 667 percent target, matches one divided by the margin, but the author says that is his inference. It is a test, availability is unconfirmed, and I found no Google help page for it on October 24. Nothing in this essay depends on it shipping. The arithmetic is the same with or without the tool.
Here is the concession, and it is a real one. Break-even ROAS is the standard floor, and any retail practitioner knows to set the target above it. I also want to be careful about the bidder. Google's Smart Bidding page says it works in every auction, using signals available at auction time, and that its models predict how different bid amounts would affect conversions or conversion value. So the bidder does not bid on an average margin. The average lives in the number you type, not in the bid. Anyone who reads this as a claim about Google averaging across auctions has read it wrongly, and the mistake would be mine.
The sentence I am correcting is my own. In You Wrote That Number: Targets Are Contracts Now I said a target is an economic claim, priced against a real denominator, and I gave unit margin as the example. I did not say which margin, and I did not say which return. That left the reader with a rule that sounds finished and is not. This essay completes it.
What changed in August
I read Google's help page on target-based bid strategies on October 24. Starting August 17, Target CPA and Target ROAS campaigns that show "Limited by budget" optimise more consistently toward their set target. Before that, a budget-constrained campaign could beat its target, which Google says made results hard to predict when budgets changed. The global rollout was completed on August 27. The page says the auction itself is not changing, that the change will not directly increase spend, and that daily and monthly budgets are still respected. It also says campaigns that were overperforming their targets may see lower performance if the advertiser does nothing, and that Google will not adjust budgets or targets on its own.
The early evidence on what that did is thin but pointed. Mike Ryan and Christian Scharmüller of smec said on October 20, from their own data, that budget-limited campaigns which had beaten their ROAS target by a median of about 20 percent were beating it by roughly nothing by the end of August, and that CPCs rose on those campaigns. They did not disclose the sample, one UK retailer is the only named case, and they describe the figures as preliminary. I treat it as one analyst's early read, not proof. What it suggests is not surprising. The headroom between the number you typed and the number the campaign was achieving used to sit unspent. Now it gets spent.
That is the whole reason this is an essay and not a footnote. When the target was a hint, an error in how you calculated it cost you accuracy. When the platform delivers to the number, the error costs money, in proportion to the budget behind it.
Error one: one margin for many products
The test asks for a campaign's average margin. A Shopping or Performance Max campaign usually sells products with different margins, and the average the tool takes is a number you chose, with a definition you chose. Gross margin, margin after shipping, margin after returns and payment fees, margin after overhead: all of them get called "profit margin" in a planning meeting, and each gives a different floor. Techwyse makes the same point about shipping, returns and fees. The mix matters more than most people expect. If seventy percent of a campaign's revenue comes from a product with a 20 percent margin and thirty percent from one with 40 percent, the revenue-weighted margin is 26 percent, not the 30 percent someone remembered from a finance deck. The floor moves from 333 percent to about 385.
Error two: credited is not incremental
The return the bidder chases is credited conversion value, meaning what the account says the ads were responsible for. Some of that would have happened anyway: people who typed your brand name, people already in the basket, customers who were coming back. How much, nobody in the account can tell you without a test. Google's Conversion Lift help page, which I read on October 24, lists Search and Performance Max among the supported campaign types, and then rations it. It says Conversion Lift is not available for all accounts and tells you to contact your Google account representative. It asks for at least 1,000 observed conversions, and conversions with supplementary data do not count toward that. It asks for a minimum study budget of $5,000. A campaign can be in only one lift study at a time. Most accounts cannot run it, and the ones that can will not run it on every campaign. So the incremental share is unknown for most of the money, and the working assumption that fills the gap is 100 percent.
The same campaign at four targets
The table is illustrative. I invented the campaign so that the arithmetic can be checked on a napkin. It spends $10,000. The remembered margin is 30 percent. The true revenue-weighted margin is 26 percent, from the mix above. Sixty percent of credited revenue is incremental, an assumption I picked, not a measurement of anything.
| Target typed | Assumptions behind it | Credited revenue | Contribution | Result |
|---|---|---|---|---|
| 333% | 30% margin, every credited dollar incremental | $33,333 | $10,000 | $0 |
| 333% | True mix margin of 26% | $33,333 | $8,667 | -$1,333 |
| 333% | 26% margin and 60% incremental | $33,333 | $5,200 | -$4,800 |
| 641% | 26% margin and 60% incremental, target set to break even | $64,103 | $10,000 | $0 |
The first row is the number the margin tool would give you, on the assumptions the tool has to make. The last row is where break-even sits when the two assumptions are loosened. Even 641 percent earns nothing; it is still a floor. A real target sits above it by whatever you want the campaign to earn. I am not saying 641 is the right target, because the campaign may not be able to deliver it at any useful volume, and a higher target buys less traffic. I am saying that the distance between the first row and the last is where the money goes when the platform spends to the number you typed.
There is a second thing the table cannot show, which is that the average return on a whole campaign and the return on its next dollar are different numbers. A target governs the average. A budget decision is made at the margin. A campaign that earns 500 percent across all of its spend might earn 250 percent on the last tenth of it, and the bidder, delivering to a target, will push spend toward wherever the last dollars are still clearing the number you typed. That is the reason I ask for a budget step in the Monday move below. It is a crude instrument, but it is the only one most accounts can run this week that says anything about the return at the edge.
If you cannot run a lift study, there are cheaper ways to put a rough number on the incremental share, and none of them gives you the number itself. You can read brand and non-brand returns separately, since brand is where credited and caused diverge most. You can look at the share of conversions from new customers. You can take a campaign the business can bear to see go quiet, pause it for a defined window, and compare orders with the weeks around it. Each of these is a guide, and each has an obvious way of being wrong. What matters is that the share you write in the table is labelled as a guess, carries a date, and has a name beside it, so that it can be revised when better evidence arrives. A labelled guess can be re-priced. An unlabelled 100 percent cannot, because nobody knows it was a guess.
A target is a price
The best objection is that everyone already knows this. Practitioners set the target above break-even, usually by feel, and a campaign manager with ten years in the account will pick a number that works. I believe that, and I have done it. But ask what the headroom is for. If the answer is "a cushion", then nobody has priced it, and a cushion that nobody priced is what the binding target essay called aspiration debt: a number written once and never re-priced, now enforced. Headroom ought to be sized to two things you can write down, the error in the margin and the incremental share, and each of those has an owner who is not the media buyer. The margin belongs to finance. The incremental share belongs to whoever runs measurement.
That is why I would call the target a transfer price. It is the rate at which the company agrees to buy credited revenue from the platform. In most companies that price is set by the person who runs the account, in a field, without a finance co-signature, and the field is now executed as written. The connection to the rest of the P&L is the one I made in Marketing on the P&L: a marketing decision does not stop at marketing, so neither does its price.
The Monday move
The account-level playbook for Google's August change is on Rollout Brief, under Google Ads bid target adjustment, and I will not repeat it here. This is the company-level version, using only terms already on this site. For every target that binds, write four things next to it. The margin used, and whether it is an average or a contribution margin by product. The return it is measured on, attributed or incremental. The result of a fourteen-day budget step, which means moving the budget a little and writing down what happened to the return at the margin; longer than fourteen days where your conversion lag is longer, because Google's own page says to wait one to two conversion cycles before judging. And the name of the finance partner who co-signs the number. A row with a blank in any of those columns is a target whose re-pricing condition nobody has written.
On April 15, 2027, open Google's documentation for margin-based Target ROAS suggestions. My claim: it still takes a single average margin for a campaign, and it does not accept margin by product or segment or refer to incremental return.
On October 24 I found no Google help page for the feature, only trade-press reports of a test. If there is still no page on April 15, the claim does not resolve, and I will reset the check to thirty days after one appears.
A target is a price, and a price that finance did not sign is an unpriced decision with a platform attached to it. The platform will now spend toward it faithfully, which is a service worth paying for only if you know what you agreed to buy. The company that writes down the margin it used and the return it counted, and gets a finance partner to initial the number, loses nothing by it. The company that does not will find out at the next budget review what the cushion cost, and which account it came out of.