New vs. Returning Customer ROAS, and Why Blended Hides an Acquisition Problem

Blended ROAS has sat at 3.8x for five months. Spend is up 60% over the same period. Leadership reads this as clean, efficient scaling.

Underneath it, new-customer ROAS has fallen from 2.9x to 1.6x, and the only reason the blended number held is that a growing base of returning customers — acquired last year, at last year's costs — is buying again. The business is spending more to acquire fewer, worse customers, and the headline metric is smoothing it into a flat line.

This is the most common way a blended number conceals a problem it is arithmetically incapable of showing.

Why blended cannot show it

Blended ROAS puts all revenue over all spend. Returning-customer revenue is in the numerator; almost none of the cost of generating it is in the denominator, because you acquired those customers in a previous period and paid for them then.

So as your customer base grows, the numerator gains a component that the denominator never sees. Blended ROAS structurally improves with age, independent of how well acquisition is working. On a growing business it is a metric with a built-in upward bias, and that bias masks exactly the deterioration you need to catch.

The definitions, precisely

The split is easy to get subtly wrong, and the errors all push in the flattering direction.

New customer

A customer whose first ever paid order falls in this period. Not "first order this period" — a customer who bought in March and again in August is a returning customer in August, not a new one.

The identity has to be stable: deduplicate on email or phone, not on customer_id, because guest checkout mints a fresh id every time. A business with 30% guest checkout and id-based deduplication will overstate new customers by roughly the guest repeat rate, which flatters new-customer CAC and ROAS simultaneously.

Acquisition spend

Spend aimed at acquiring, not retaining. In practice this means prospecting campaigns, and it means excluding retargeting and retention — a retargeting ad shown to a past purchaser is retention spend regardless of which platform it ran on.

The clean version requires your campaign naming to distinguish the two, which is one of the underrated returns on a disciplined UTM and campaign taxonomy.

New-customer ROAS

New-customer ROAS = First-order revenue from new customers ÷ Acquisition spend

First-order revenue only. Including their later orders turns it into a lifetime metric, which is a legitimate thing to measure but is not this — and mixing them makes the trend uninterpretable, because recent cohorts have had less time to place later orders.

The query

Reading the four patterns

Once both series exist, the shape of the divergence tells you what is actually happening.

The other two are simpler. Both falling is usually a market-wide cost increase or a margin/pricing change — check CPMs and AOV before touching campaign structure. Both rising is genuine improvement, and the only thing to check is whether you are underspending.

The second-order metric worth adding

Once the split exists, one more number turns it from a diagnostic into a decision tool: new-customer contribution against the cost of acquiring them, over a bounded window.

A cohort_payback_ratio above 1.0 means the cohort repaid its acquisition cost in gross profit within 90 days. Tracked month over month, this is the single most informative acquisition metric most teams do not have — and it is immune to both attribution disputes and the returning-customer masking effect, because the cohort is defined by when you bought it.

What to report

Four lines. Reported together, they make the masked decline impossible to miss — and any one of them alone can be read as good news while the business is deteriorating.

Frequently asked questions

How do I split acquisition from retention spend if my campaigns are not named for it?

Approximate it first: treat prospecting audiences as acquisition and retargeting or customer-list audiences as retention. It will be imperfect at the edges — a broad prospecting campaign reaches some past customers — but a rough split beats no split by a large margin. Then fix the campaign naming so the next quarter is clean.

Should first-order revenue include the customer's second order if it lands in the same month?

No. First-order revenue means the first order, full stop. Their second order is returning revenue even if it happens nine days later. Mixing them makes months with faster repeat cycles look like better acquisition months.

What about customers acquired through channels I do not pay for?

They are still new customers and belong in the numerator, which will make new-customer ROAS look better than paid acquisition alone. If organic acquisition is a meaningful share, split it out too — an organic new-customer count rising while paid falls is a very different situation from both falling.

Is a falling new-customer ROAS always bad?

Not necessarily. Deliberately scaling into a broader, more expensive audience lowers new-customer ROAS by design, and can be correct if the customers acquired are worth it. The check is the 90-day cohort payback ratio: if it stays above your threshold while ROAS falls, you are buying more expensive customers who are still worth buying. If both fall, you are simply buying worse ones.

How does this relate to blended CAC?

Blended CAC divides total spend by total new customers — so it already isolates new customers in its denominator, but keeps retention spend in its numerator. New-customer ROAS goes one step further by isolating acquisition spend as well. Use blended CAC for the board; use new-customer ROAS and CAC for the acquisition decision.

The summary

• Blended ROAS contains returning-customer revenue that this period's spend did not buy, so it improves with age regardless of acquisition health.
• Split it: first-order revenue from first-time customers, over acquisition spend only. Deduplicate customer identity on email or phone, never on an id that guest checkout regenerates.
• Blended flat + new-customer falling is a masked decline, and it surfaces in the headline roughly two quarters late.
• Blended falling + new-customer flat is a retention problem; cutting ad spend makes it worse.
• Add the 90-day cohort payback ratio — it is immune to both attribution disputes and the masking effect, because the cohort is defined by when you bought it.