LinkedIn
Most agencies won’t tell you this: that 400% Target ROAS you typed into Google Ads can make an account smaller without making it more profitable. The number doesn’t order Google to deliver a 400% return. It tells the bidder how much value it must expect before a click is worth buying.
Here is the fact everything else follows from: raise the target and you reduce what the bidder can afford to bid for the same predicted value. A useful model is predicted conversion value divided by target ROAS. If a click looks worth $50 and your target is 500%, the model gives you $10 to bid. Google describes the objective as keeping conversion value divided by cost near your target over time, not promising that every auction will return your number. I used to tell clients to set the target at break-even and leave it there. I was wrong about the leave it there part.
First, price the click before you choose the target
Google starts with an estimate, not your margin spreadsheet. It considers the chance of a conversion and the value that conversion might bring. The estimate can change from one auction to the next: device, query, location, time and audience all matter. Google lists real-time signals including device, browser, location, time of day and remarketing lists.
Then the target constrains what Google can bid against that estimate. In the simple division model, a click with $60 in predicted value supports a $20 bid at a 300% target. At 600%, it supports $10. Same predicted value. Half as much room to compete. Actual auction bidding is more complicated than a fixed CPC ceiling, but this model captures the decision your setting pushes: a higher target makes marginal auctions harder to buy.
It also explains why spend may fall sharply rather than neatly in proportion to the target change. Some auctions have plenty of predicted value relative to their cost. Others barely clear the line. Tighten the target and the borderline auctions disappear first; keep tightening and there is less left to buy. The practical question isn’t, “What ROAS would I like?” It’s, “Which clicks am I willing to stop buying?”
Why the report doesn’t land exactly on your number
The prediction happens before the sale. The ROAS report happens after it. Google says individual auctions can return more or less than the target while the system aims for the average. Add delayed purchases and the newest days become particularly easy to misread. If a customer buys three days after clicking, spend appears before that conversion value does. I treat ROAS from the last 3 to 7 days as provisional and wait for a full conversion cycle before judging a target change.
Selection changes the report, too. Raise the target and you may stop buying weaker-looking clicks. Reported ROAS can rise even as revenue and customer volume fall. Google warns that a target set too high can limit traffic and ads served. Lower it and you may buy more marginal clicks, bringing the average down while expanding total sales. A target miss is information about predictions, prices and the auctions you bought. It isn’t Google being disobedient.
Before turning the dial, check the values behind it
The divisor gets attention. The numerator deserves the first inspection. If every purchase reports as $50, the bidder cannot distinguish a small basket from a large one using the value you supplied. I’ve seen a static purchase value on Shopify stores whose actual baskets range from $29 to $240, and lead-gen accounts that assign the same value to a demo and an ebook. Neither problem yields to a cleverer target.
Check that real order values flow back and that feed prices and IDs are correct before adjusting ROAS. Where value differs by geography, device or customer type, conversion value rules can reflect those differences. If margins differ across products, your revenue total may still be a poor description of what those sales are worth to you. Garbage values in, obedient but wrong bids out.
Now calculate your economics. Break-even ROAS is 1 divided by your profit margin. For a $100 shoe at a 30% margin, that’s roughly 333%. Say you’re spending $20k a month selling it. A 333% target sounds like discipline: why ask the bidder to buy a click that looks unprofitable?
Because a target is a constraint on predicted returns, not a guarantee of actual returns. Predictions are uncertain, reported values can lag, and some useful auctions will look weaker than they turn out to be. Setting the target at break-even can restrict volume before you’ve established whether those excluded auctions actually lose money. But the opposite mistake is just as real. At a 30% margin, $20k in ad spend returning 280% produces $56k in revenue and $16,800 in contribution before ad cost. That does not cover the $20k spend. Calculate break-even, write it down, then test volume against actual contribution. Don’t turn “below break-even target” into permission to accept below-break-even results.
Separate conversion lag from a real bidding problem
Say a customer typically buys three days after the click. Today’s spend is visible; some of the value it produces will arrive later. That does not mean Google simply sees three days of worthless clicks and panics. It means your freshest report is incomplete, which is a different problem.
Google advises excluding the recent conversion-delay period when you use historical conversion value divided by cost to choose a target. I compare the last seven days with an earlier seven-day window that has had time to mature. If the older window clears the target and the recent one doesn’t, I wait before diagnosing a stall. Lowering the target solely because delayed revenue hasn’t reached the report can buy more clicks at higher prices to solve an accounting illusion. Let the values arrive, then decide.
Sparse conversions make every prediction shakier
The bidder needs examples. Google’s stated minimum for Search and Shopping tROAS is 15 conversions in the past 30 days, while a practitioner guide suggests 50 conversion events in 30 days before relying on it. Eligibility and a comfortable amount of data are not the same thing.
With 12 conversions a month, one unusually large order can exert a lot of influence. The next change in spend may look like a verdict on your target when the model has little to generalize from. For a thin campaign, start with Maximize Conversion Value without a target and add one when volume supports it. Practical takeaway: if the account barely produces conversion examples, stop trying to tune the divisor to compensate.
Walk the dial without fooling yourself
Return to the $100 shoe and its 30% margin. Suppose an account spends $20k and reports 300% ROAS. That’s $60k in revenue and $18k in contribution before ad cost: $2k short of covering the spend. Now suppose a tighter target cuts spend to $13k and reported ROAS rises to 390%. Revenue falls to $50,700, but contribution before ad cost is about $15,210. After the $13k in ads, the account has about $2,210 left.
That tighter setting is a win on these numbers, even though volume fell. The loss of volume is a warning to investigate, not proof that loosening the target will earn more. If I called the higher ROAS a failure just because fewer auctions cleared, I’d be making the same mistake as someone who calls every higher ROAS a success. The margin calculation gets the vote. A comparison of $10k spend at 500% with $20k at 400% makes the same point from another angle: efficiency alone cannot tell you which level produces more profit.
Now consider a move from a 333% target to roughly 265%, about 20% lower. In our simple model, a click predicted to bring $50 in value supports about a $15 bid at 333% and close to $19 at 265%. That extra room may bring back auctions the tighter setting priced out. It may also buy clicks whose eventual return doesn’t cover their cost. Shopify describes how distribution can contract when actual ROAS falls below target; easing the constraint can expand it again. The question is whether the added sales improve contribution after ad cost. Watch that alongside spend, reported ROAS and new-customer volume. Use the target to test how much volume is available; use your economics to decide whether you want it.
What a stalled account is trying to tell you
The stall has a recognizable shape. Spend stops reaching budget, reported ROAS looks cleaner, and prospecting has thinned out while brand and returning visitors remain. I’ve seen that pattern on a €15k-a-month store. A high tROAS can constrain spending because fewer auctions look able to support the required return. The account may look more efficient partly because it stopped taking the harder tests.
I don’t respond by yanking the target down on sight. First I check conversion values and the lag-adjusted window. Then I ask whether the remaining volume is actually more profitable, as it was in the shoe example. A smaller account can be healthier; a prettier ROAS report cannot tell you that on its own.
The standard advice is to move targets in 10 to 15% steps and wait two weeks. I like small steps when they help me read cause and effect, not because the number itself has a magic learning speed. Google recommends using historical conversion value divided by cost and testing large changes as an experiment. For a larger shift, such as 333% to 265%, I would test against the old target and allow a full conversion cycle before calling it. Change one knob, then read the result that matters.
Give temporary changes and mixed campaigns their own treatment
A short promotion does not automatically call for a permanent new target. If you expect a substantial, temporary change in conversion rate, seasonality adjustments are designed for short windows, ideally 1 to 7 days and less useful beyond 14. I used to rewrite targets for promo week and spend the next two weeks undoing it. Now I separate a known, brief conversion-rate change from a lasting change in the account’s economics. Routine holiday drift isn’t, by itself, a reason to grab the dial.
Pooling campaigns calls for similar care. A portfolio tROAS strategy can put one target across campaigns, giving thin campaigns more data to work with. Portfolio tROAS has availability limits, including Performance Max, Hotel and Travel campaigns. But one target also applies one constraint to auctions with different economics. Brand search with an 800% natural return and prospecting with a 250% natural return do not become comparable because both appear in one portfolio. I keep brand and non-brand, and Search and Shopping with different margins, on separate targets. Pool volume where intent and value are similar, not where the report looks tidy.
When the target is the wrong tool
If the job is to turn a fixed $20k budget into as much conversion value as possible, ask whether you need an efficiency constraint at all. A very low tROAS target isn’t the same setting as uncapped Maximize Conversion Value. Start without a target when maximizing value within budget is the job; add a realistic one when you need to enforce efficiency. Then judge the result against margin, not the satisfaction of having typed a number into a field.
Lead gen exposes the other failure. If a closed-won deal is worth $4,000 and an ebook download is worth $20, but both arrive with the same reported value, tROAS has no honest basis for preferring the deal. Import real offline values first or stay with tCPA until the values can distinguish the outcomes. No divisor fixes a false numerator.
Run one new case through the model. Your Shopping campaign has a 400% target, impression share lost to rank sits near 45%, and the last seven days report 340% ROAS. Don’t touch the dial yet. Check conversion value divided by cost over the last 30 days, excluding the most recent seven to allow for lag. If that mature window reads 380%, you have a possible constraint to test, not proof that every lost impression was worth buying. In the simple bid model, moving from 400% to 320% gives the same predicted value 25% more bidding room. Run the lower target as an experiment, wait a full sales cycle, and compare total contribution and new-customer volume with the old setting.
This is the loop I stopped doing by hand at 1am. groas runs it continuously, with values, lag and bid changes checked, changes logged, and a strategist responsible for the outcome. That beats paying a retainer to move the dial twice a quarter. Set the target to test the volume you can profitably buy. Then make sure the values you feed it are worth trusting.