One Account's Blended 4.0x ROAS Hid a 2.8x New-Customer Number Underneath.
A healthy blended return on ad spend can sit directly on top of a new-customer number that would read as a problem on its own.
Returners convert faster, so the algorithm finds them first
An ad platform optimizing for conversions will find the easiest conversion available, and a customer who has already bought once tends to convert three to five times faster than someone new. Even with exclusion campaigns running, roughly half of attributed sales in one documented operator's example turned out to be people who had already purchased before, quietly propping up the blended number without anyone deciding that was acceptable.
One account, two very different ROAS numbers
That operator's own worked example: a $250,000-a-month account blended to a 4.0x return on ad spend. Split new customers from returning customers against the same spend, and the new-customer ROAS came out to 2.8x while the returning-customer ROAS was 5.5x. The blended figure was real. It just was not the number that mattered for actually growing the business.
A heavier survey is still a rented number
The usual fix on offer is a heavier measurement stack: a multi-touch attribution overlay, a post-purchase survey asking where someone first heard about the brand, tuned inside the ad platform itself. That is a real improvement over reading the blended number blind, and it is still an estimate assembled on someone else's platform, gone the day the subscription lapses. A new-versus-returning split that actually holds up is one resolved against the client's own real orders, in a warehouse the client owns, not reconstructed from a survey layered on top of a platform's existing blind spot.