⏱ 5-minute read
The Limits of ROAS
Better ROAS does not mean a healthier business. It can mean the opposite, and the gap between those two things is where companies quietly get into trouble.
ROAS measures what happened on the order. It says nothing about the customer who placed it, whether they'll return, what they'll spend next, or whether the margin they generated will ever pay back what it cost to acquire them. That's not a flaw in ROAS. It's the limit of what it was designed to do. The mistake is using it as a business scorecard.
Cheap to Convert, Expensive to Keep.
When you lean into a discounted SKU, deepen promos, or shift spend toward bottom-funnel demand capture, ROAS goes up. The platform calls it a win. Customers become cheaper to convert.
What the platform never tells you: those customers may be more expensive to keep.
Lower-intent buyers convert on price, not brand. They return more. They repeat less. They don't migrate up the range. In-platform, every one of those decisions reads as efficiency. In the business, each one is quietly degrading the customer file.
That's the structural problem. ROAS is optimized at the order level. Cohort behavior plays out over weeks and months. The platform isn't lying to you, it's just only tracking one thing. You have to track the other.
What To Trust Instead
A cohort is a group of customers acquired in a defined window: a week, a month, a campaign, tracked forward in time. Where ROAS closes the book on acquisition day, cohort behavior keeps score on what actually happened:
Did they buy again within 60 days?
What did repeat rate look like by day 90?
What was net contribution margin after returns, discounts, and shipping?
How many days until the cohort paid back CAC?
Two cohorts with identical day-one ROAS can look completely different by month three. One compounds. One plateaus. The platform never told you which was which because it stopped paying attention after the conversion.
Four Signals That Matter
Net CM payback. How many days until CM1, after discounts, returns, COGS, and shipping, exceeds CAC. Healthy is under 60 days. Above 90, you're looking at cash risk. If ROAS improved but payback extended, you didn't get more efficient. You got more exposed.
Repeat rate by cohort. If repeat rates on a recent cohort are tracking below your baseline, acquisition quality dropped, regardless of what ROAS showed.
Cohort by acquisition source. Some channels recruit returners or discount shoppers at a structurally higher rate. If conversion only happened because the product was cheap or the discount was deep, you may have improved ROAS while lowering future LTV. That's borrowed revenue, not growth.
Entry product migration. Your ad account is not qualified to choose your best growth product. Your cohort economics are. An entry product with weak repeat, high returns, or poor payback isn't a hero, it's a trap, regardless of what it does to day-one ROAS.
The Takeaway
Stop reporting ROAS in isolation. Pair every ROAS movement with at least one cohort signal, repeat rate, payback window, or return rate by source. When they move together in the same direction, the efficiency gain is real. When ROAS improves but cohort signals weaken, something was traded away to get there.
The brands that scale without breaking cash flow aren't ignoring platform metrics. They're just not letting platform metrics have the final word. ROAS tells you what the platform did. Cohort behavior tells you whether it was worth it.
If you want to know which cohort signals I'd look at first in your data, reply COHORT and I'll tell you exactly where I'd start.
