Mizzen+Main makes performance menswear, dress shirts and tailored clothing built from technical fabrics, sold direct to consumer on Shopify. Paid acquisition is run by adMixt, a performance marketing agency and an INTERLUXE group company.
Years of Disciplined Audience Testing, and a Ceiling That Wouldn't Move
Mizzen+Main's paid program was not broken. It was well run: a dedicated team, constant creative iteration, a full audience testing history, and every dollar measured in a best-in-class attribution platform. Over the years the team had tested the audiences on every serious brand test: behavioral lookalikes built from past purchasers, value-based seeds, broad prospecting at multiple age bands.
The pattern that emerged is one most operators will recognize. Broad delivered volume at a high cost per new customer. Behavioral lookalikes did better, but they carried a built-in limitation: a seed list of "past purchasers" blends very different kinds of buyers into one audience, full-price loyalists and discount-triggered shoppers, occasional gift buyers and repeat customers, so the lookalike Meta builds inherits an average of everyone rather than the shape of the best.
Measurement was never the gap. The team could see exactly what everything cost. What no attribution tool could tell them was who their best customers actually were, and how to manufacture more of them.
Personas as Seeds: Targeting Built From Who Customers Are, Not Just What They Did
Mizzen+Main connected their store to GoAudience, which generated customer personas from their real buyers: clusters built from purchase behavior and enriched demographic data, each persona a distinct kind of customer with its own economics, full-price share, repeat pattern, and lifetime value.
The personas surfaced something the team's audience lists had been hiding for years: demographically similar customers with opposite buying behavior. Two personas can share an age range, income band, and geography, while one buys almost entirely at full price and the other almost never does. Every "past purchasers" seed the account had ever used contained both, which is precisely why those lookalikes plateaued.
The play was simple: seed lookalikes from the personas with the strongest economics instead of from blended lists. adMixt pushed two GoAudience personas to Meta and built a lookalike from each, in its own ad set with its own budget, running the account's existing creative. The lookalikes were built at 20% depth through Meta's API, a breadth only available programmatically, wide reach, carried by seed quality rather than audience narrowness.
Then the market answered. Within days, both persona-seeded ad sets were outperforming everything else in the account. adMixt scaled the first seed's budget aggressively, and performance held. A week after the first launch, the team shipped the second seed unprompted. Within two weeks, the persona seeds had become the largest allocation in the account, budget following evidence.
"We test audiences constantly, and these behaved differently from day one. Broad 20% lookalikes seeded from GoAudience personas beat everything else in the account on new-customer CAC, and held while we scaled spend 3x. Our own Northbeam data confirmed it and its been a long time since I have seen a LAA perform this well week over week at scale."
The Brand's Own Attribution Stack Confirmed It
Two independent measurement systems, with different methodologies, told the same story.
Northbeam, the team's source of truth for attribution, measured on a strict clicks-and-deterministic-view model with a new-customer CAC basis: persona-seeded audiences acquired new customers 22% cheaper than all other targets combined, at 10% higher average order value, on roughly a fifth of account spend. When the team roughly tripled persona-seeded spend in a single week, CAC on those audiences improved rather than fatigued, the opposite of what scale usually does. Ranked by new-customer CAC across every active target, both persona seeds landed in the top half of the entire account.
Meta's in-platform reporting, read directionally and separately (platform attribution and third-party attribution measure differently, and the two should never be merged into one table), showed the same shape: the two persona-seeded ad sets ranked as the cheapest acquisition sources in the account, at roughly 2x the return of the account's best alternative audience, at identical CPMs, meaning the difference was seed quality, not cheaper inventory. The cleanest proof point: the identical ad creative, running simultaneously in a persona-seeded ad set and in other ad sets, converted at roughly one third the cost inside the seed. Same ads. Different audience. Different economics.
"We bet on GoAudience to understand not only WHO our customers are but also HOW they shop. Persona characteristics don't always determine shopping patterns, and GoAudience unlocks that difference and sends signals to platforms like Meta that are just as strong as anything else we feed them, if not stronger."
Cheaper Customers, Bigger Baskets, and a New Default for Prospecting
- 22% cheaper new-customer CAC, verified in Northbeam
- 10% higher AOV on persona-acquired customers
- Roughly 8x return on persona-seeded spend in Meta during the launch window
- Spend scaled 3x week over week with improving CAC
- Second persona seed launched by the agency unprompted within a week
The order-value result deserves its own line: persona-seeded audiences didn't just find cheaper customers, they found customers who spend more from their first order. Cheaper and better is the combination audience testing rarely produces, because it requires knowing which customers are worth resembling.
"The real surprise was the first-purchase order value, especially at a lower cost per acquisition. Customers who spend more upfront give us a strong signal for better long-term customers."
Acquisition is the first chapter. The same personas now power what comes next for the team: retention flows personalized to how each kind of customer actually buys, and margin protection built on knowing who needs a discount and who would have paid full price.
Mizzen+Main's results were measured by their agency in Northbeam (clicks and deterministic view model, new-customer CAC basis) over a fixed multi-week window with both persona-seeded ad sets live, alongside directional Meta in-platform reporting for the same period.