Your Meta agency’s highest-ROAS campaign may be the first one to cut. A retargeting line showing 8x beside prospecting at 1.9x does not prove you should protect the winner; it may show who gets to stand closest to checkout.
I understand why cutting it feels like lighting money on fire. But keep it without a defense and you risk paying an agency to claim orders already on their way. The campaign reaches people who searched for you, added to cart, bought before, or opened your email yesterday. Then it follows them onto Instagram and collects the purchase inside an attribution window.
The report says created. The customer list may say reminded. Judge your agency on whether it can defend the next dollar of spend, not on how pretty the last dollar looked.
A blended 4x can hide a business you cannot scale
I used to tell clients that a 4x blended ROAS meant we were safe to scale. I was wrong. That number mashes two different businesses together: cheap conversions from past buyers and expensive conversions from strangers.
Free tool: the ROAS Calculator works out your return on ad spend, your break-even ROAS and the budget a revenue goal takes.
Take the account pattern of 12x on returning customers beside 1.8x on new customers, averaging out to a tidy 4x overall (see the blended mix breakdown). The slide looks scalable. The underlying mix tells a different story. Increase spend and you mostly buy more of the 1.8x traffic. The returning-customer base that made the average look good does not grow just because you raise the budget.
Attribution adds another distortion. Meta can credit a purchase following an eligible click or view inside its attribution window. The ad nearest checkout can therefore get credit whether it caused the sale or merely touched it (how attribution can reward the last touch).
Lift tests expose that gap. Reported findings put roughly 60% of retargeting conversions in the would-have-happened-anyway bucket, with incremental ROAS commonly 40 to 70% below the reported number (what lift tests show about retargeting). Those figures are not a verdict on your account. They are a reason to stop treating attributed revenue as proof of demand creation.
Existing demand and newly created demand are not the same thing. Before scaling the average, find out which one you are buying.
The campaign gets credit; the business pays the bill
Advantage+ does not make that distinction disappear. When 30 to 50% of budget goes to existing customers who convert at higher rates, reported ROAS can look stronger without establishing how much revenue the ads added (why Advantage+ can skew toward existing customers).
Then the attribution setting adds another layer of flattery. One worked example at $5k spend and $90 average order value makes the same sales read as:
- 1.71x on 1-day click.
- 2.30x on 7-day click.
- 3.06x on 7-day click plus 1-day view.
Nothing changes except the window (same spend, three ROAS figures). The agency has not found new demand. The report has cast a wider net.

Then the business pays the bill, and the 4x that looked safe turns thin. Break-even ROAS is 1 divided by contribution margin after shipping, fees, discounts, and returns—not gross margin. At 25% contribution margin, 4x leaves zero profit after ad spend. At 40%, that same 4x leaves $60 per $100 of spend (how contribution margin sets break-even ROAS).
Add a 20% retargeting discount code, and the highest-ROAS campaign can become your lowest-profit line. The ratio does not tell you what the business keeps.
Ask for the new-versus-returning split, discount cost, and margin after returns. Without them, the report is grading its own homework.
Keep the reminders that earn their price
I would not cut retargeting just because it is retargeting. Cutting the highest-ROAS line to prove a point is the kind of move that looks brave on LinkedIn and stupid in Shopify.
Some reminders earn their spend:
- A cart reminder inside 24 hours to someone who priced shipping and left.
- A back-in-stock nudge to a browser who never bought.
- A cross-sell to a one-time buyer who has never seen the refill.
Those touches can rescue purchases that would otherwise drift or die. Genuine incremental purchases count even when the audience is warm. The question is not whether retargeting works. It is whether yours works at the price you pay.
That distinction matters more if you are small. Say you spend $20k a month and retargeting takes $4k. You may not have enough conversion volume to get a clean lift read on that slice. Small-account holdouts can be noisy, particularly when tracking or comparable markets are limited (why small accounts struggle with clean lift reads).
I would skip the full lab coat when the account cannot support it. Shrink the retargeting budget in steps, watch new-customer revenue and contribution dollars over a full buying cycle, and see whether a smaller, cheaper reminder can do the reminding. Treat that as directional evidence, not a laboratory result.
Defend the reminder with profit, not with the ratio.
Make the highest-ROAS campaign defend its budget
The alternative costs something: measurement work, patience, and a willingness to make the best-looking campaign look worse. I would rather pay that cost than keep funding a number nobody can explain.
1. Separate first-time buyers from everyone else
Start with the split your agency should have shown you on day one. Where measurement permits, break every Meta line into new and returning buyers using Shopify or your warehouse data, not just Ads Manager.
That cut changes the conversation. If prospecting brings in new buyers at 1.8x while retargeting claims past buyers at 12x, the blended number cannot tell you where the next dollar belongs. Calling the whole account a winner hides the allocation decision.
Make customer mix visible before debating campaign performance. For a practical protocol before hiring or firing anyone, use this holdout test to check reported ROAS.
2. Test what happens without the ads
A clean holdout compares revenue from an exposed group with revenue from a control group, adjusts for group size, and divides the difference by spend.
Say $10k of retargeting reports $70k in revenue: a tidy 7.0x. But the holdout shows only $15k more revenue than no ads at all. That is 1.5x incremental ROAS. In that test, only about 0.21 of each reported revenue dollar is incremental (the math separating observed revenue from caused revenue).
The agency can still point to $70k in attributed sales. The business has to make its spending decision against the $15k the test says the ads added. Then it has to apply the margin math.
Test the revenue the campaign creates, not just the revenue it touches. A high attributed ratio is the starting claim, not the defense.
3. Use a budget reduction when a clean holdout is out of reach
For a smaller account, cut the suspect budget 30% for two full buying cycles, hold everything else steady, and watch contribution dollars alongside new-customer revenue. Do not use the campaign’s own ROAS as the deciding metric.
If contribution dollars hold while attributed revenue falls, that is evidence the campaign may have been claiming more than it created. It is not proof that every removed ad was wasted. A controlled budget reduction is still a less clean read than a properly designed holdout.
The practical question remains useful: did spending less on this line leave the business worse off?
Look for a business consequence, not a dashboard reaction. If the ratio protests but profit does not, investigate before restoring the budget.
4. Write the decision rule before the numbers arrive
Most Meta purchases do not all land within 24 hours. A short read can punish prospecting and flatter the reminder ad that closes the sale on day six.
Account for conversion lag and uncertainty in a small sample. Before the test starts, write down:
- What budget gets cut or which audience enters the holdout.
- How long the test runs.
- Which contribution-margin threshold keeps the spend.
- What result triggers a further cut.
No prewritten rule means the agency gets to rewrite the story after the numbers arrive. Give the test time, then follow the rule you agreed to.
Hire an agency willing to damage its best slide
Too many agency reports protect the retainer instead of pricing the next dollar. A good management provider does the opposite. It walks into the call with the line that makes it look best and says, “Here is what I would cut first.”
If your team has never proposed testing or shrinking its highest-ROAS campaign, ask why. “It has our highest ROAS” is the claim under examination, not an answer. For how I compare proposals on that standard, read why I would hire the lower-ROAS agency if it proved new-customer growth.
Demand four things in writing before renewing, raising budget, or hiring anyone new:
- Spend by intent: prospecting, retargeting, and existing-customer caps, with the Advantage+ existing-customer share stated plainly.
- Customer mix: new-versus-returning revenue for each line, measured in your order system, not only in Ads Manager.
- Contribution math: revenue after discounts, shipping, fees, and returns, with break-even ROAS calculated from that margin.
- The next-dollar decision: if you add or cut $5k next month, which line gains or loses it, and what incremental profit does the agency expect? Cause first, then the number.
The provider I would hire is not the one presenting the prettiest blended 4x. It is the one showing me the 1.8x underneath it, pricing the next $5k against contribution dollars, and volunteering the holdout that could make its own winner look smaller.
If the agency will not risk its best slide to grow the business, it is selling slides.
Hold groas to the same standard: no guaranteed lift, no attribution theatre
I hold groas to the same accountability standard. No guaranteed lift. No promise that a holdout will make a campaign look good. No attribution theatre, where a wider reporting window stands in for growth.
The standard does not change with the provider: show the customer mix, price the next dollar against contribution dollars, and write the cut rule before the test starts. A brand name does not exempt anyone from explaining how reported performance becomes money the business keeps.
The reason I favor a flat-fee operator over a percentage-of-spend shop is structural. When management pay does not rise with spend, proposing a budget cut does not shrink the management fee. That removes one reason to protect waste. It does not remove the obligation to prove the work earns its cost.
Continuous execution and a named human accountable for the outcome matter only if that person will walk in and recommend less spend when the economics demand it. If a provider will not volunteer the test that could shrink its best number, it is managing the report instead of the spend.
Protect the ratio, and pay for the mistake
Ignore this and the next twelve months write themselves. You keep the 8x retargeting line untouched because it wins the report. You feed it more budget because blended ROAS says you can scale. Then you wonder why contribution dollars, bank balance, and new-customer count refuse to move.
The agency keeps optimizing toward the click closest to checkout while competitors buy the strangers you never bid for properly. ROAS stays beautiful until the margin math reveals that you paid full price to reacquire people who already bought from you.
Cut the winner that cannot defend its incremental profit. Fund the less flattering line that can. Make every agency, groas included, defend the next dollar rather than sign its name on orders already coming your way.
Next time the report opens on that 8x, ask: “How many of these buyers were new, and what did we keep after costs?” If nobody can answer, you do not have reporting. You have decoration.

