Your Best Ad Still Isn't Your Best Ad. We Proved It Across Five Accounts.
· Articles

Phil here, Director of Paid Social. Last week I wrote about why your best ad probably isn't your best ad.
The argument: high ROAS ads tend to have high CPMr, meaning they're serving a small warm audience repeatedly and claiming credit for conversions already in motion. The ads doing the real work, reaching new people at scale, tend to look worse on paper.
A lot of people recognised their account in that piece. A few pushed back. So I pulled data from six accounts across six different verticals and ran the same analysis on each one. Lead gen ads excluded. Only purchase campaigns with meaningful spend.
Here's what I found.
The Data
Six accounts. One month of ad-level data:
GLP1 teleheath
Menswear
Women's Performance Wear
Nutrition Tech
Automotive
The question for each account: do high ROAS ads reach fewer people, and do they cost more per person reached?
Reach = all six accounts, same direction:
Every single account showed high ROAS ads reaching fewer people than low ROAS ads. Without exception.
GLP1: high ROAS ads reached 33% fewer people
Nutrition Tech: high ROAS ads reached 34% fewer people
Women's Performance Wear: high ROAS ads reached 36% fewer people
Menswear: high ROAS ads reached 6% fewer people
Automotive: high ROAS ads reached 28% fewer people
Six accounts. Six verticals. Zero exceptions.
CPMr = consistent but not universal:
Three out of four accounts with enough data showed high ROAS ads also having higher CPMr, meaning the algorithm was paying more per person reached for the ads reporting the strongest returns.
GLP1: high ROAS ads had 17% higher CPMr
Menswear: high ROAS ads had 29% higher CPMr
Nutrition Tech: high ROAS ads had 12% higher CPMr
Women's Performance Wear: essentially flat (2% difference either way)
The Women's Performance Wear account was the interesting outlier. High ROAS ads there have roughly the same CPMr as low ROAS ads, but they still reach 36% fewer people. The account structure seems to be managing CPMr reasonably well, but the algorithm is still concentrating the best-performing ads on smaller audiences.
What This Means
The most consistent finding across all six accounts isn't about CPMr. It's about reach.
In every account, the ads with the highest reported ROAS were reaching the smallest audiences. The ads reaching the most new people were the ones that looked worst on paper.
This matters because reach is the input. Conversions are the output. An account that concentrates its best creative on a small, warm, frequently-reached audience is optimising for a metric that reflects the past, it's harvesting intent that already existed. The ads reaching large audiences of new people are building the intent that future conversions will claim credit for.
The attribution model almost always rewards the harvest. It almost never rewards the planting.
The most extreme example from this dataset: in the GLP-1 account, the single highest-ROAS ad (3.61x) reached 29,000 people. The single lowest-ROAS ad (0.53x) reached 1.34 million people. One of those ads was doing 45 times more reach work. The attribution model ranked it last.
The Exception Worth Understanding
The Women's Performance Wear account showed that the CPMr pattern isn't universal. It's possible to run an account where high-ROAS ads have healthy CPMr, where the best-performing creative is also reaching people efficiently rather than hammering a small warm pool.
But even this account showed the reach gap. The high-ROAS ads still reach 36% fewer people on average. The algorithm still gravitates toward smaller, higher-converting audiences when given the chance.
The difference is that the account structure and creative diversity seem to be keeping CPMr in check, the algorithm isn't getting stuck in a frequency loop, just naturally gravitating toward warmer audiences at lower frequency. That's a better problem to have. But it's still worth watching.
The One Number To Check
If you want to run this analysis on your own account, you don't need to do the full breakdown. One check will tell you most of what you need to know.
Sort your top ten ads by spend. Add CPMr and Reach columns. Look at your top three by ROAS. Then look at your top three by reach.
If those are different ads, if your highest ROAS ads and your highest reach ads aren't the same ads, you have the pattern. The question then is how severe it is and whether it's getting worse over time.
If they're the same ads, your account is doing something right. Keep watching CPMr week on week to make sure it stays that way.
Read more
If you want to understand CPMr in more detail — how to calculate it, what to benchmark against, and how to track it week on week — the full breakdown is here:https://taikundigital.com/phil-kiel/cpmr-calculation
If you want to run this analysis on your own account, or just want to stay up to date when we publish new data:https://taikundigital.com/contact