How we cut CPA 51% for a national women's fashion brand with a creative-testing system

The challenge

A national women's fashion DTC brand, $28M ARR, $94 AOV, ~28% returns rate. They spent $180K/month on Meta with a 1.9× blended ROAS and a $47 blended CPA. The brand's growth marketer had been running the account in-house for 18 months and had hit a ceiling, same creative concepts cycling, same audiences, same offer.

The founder's brief: "We're running the same playbook everyone else in our category is running. We need a creative engine, not a media buyer."

What we found

The audit confirmed what the team suspected:

• Creative production was the binding constraint. They were shipping ~2 new creatives per week, almost all derivative of the same hero photoshoot. Top-performing ads were 14+ months old and showing fatigue (frequency 9-12 in their core audience).
• No creative-testing system existed. Tests were ad-hoc, never reaching significance, and learnings weren't documented. The same losing concepts were quietly being re-tested every 4-6 months.
• The brand was over-relying on Meta. TikTok had been tested twice with poor results, but the tests had used Meta creative on TikTok, which never works.
• Returns were eating margin. 28% returns vs 18% category benchmark. The creative implied a fit and styling that didn't match the actual product.

Approach

We didn't change the media buyer (the in-house growth marketer kept the seat). We built the operating system around her, a weekly creative cadence, a testing framework that reached significance, and a TikTok program that didn't recycle Meta assets.

Week 1-3

Creator sourcing + brief library

Built a roster of 14 US-based UGC creators and a brief library of 9 angles. Templated weekly creative briefs so the team could ship without rewriting the spec each Monday.

Week 4-6

Testing framework rebuild

Designed a 4-week test calendar with pre-registered hypotheses, sample-size targets, and a loss-memo template. Every test was logged in a shared doc the team referenced before approving new tests.

Week 7-9

Returns-aware creative

Audited 80 prior creatives against the actual return rate they drove. The ads that promised "oversized fit" against products that ran true-to-size were responsible for 38% of returns. Rewrote 6 hero concepts with fit honesty built in.

Week 10-12

TikTok launch

Native TikTok creative, not Meta cuts. Sourced 5 fashion creators, ran 8 Spark Ads concepts over 6 weeks. Two became hero ads that we then re-cut for Meta.

Week 13-16

Scale + cohort

Spend lifted from $180K to $231K/month while CPA dropped to $23. Built a returning-customer creative track separate from acquisition.

Execution

The piece of work that mattered most was the creative-testing framework. The brand had no shared definition of "win", some tests had been declared winners after 3 days of data, others had been killed after a week of underperformance with no statistical basis. We standardized on Bayesian testing with a 95% probability-to-beat-control threshold, fixed-duration tests of 7-14 days, and pre-registered hypotheses.

Of the 38 creative tests we ran over 16 weeks, 14 were winners, 19 were losers, 5 were inconclusive. Each got a loss memo or win memo, a one-page write-up that the next test designer referenced. By month 4, the team was no longer re-running concepts that had been tested 6 months earlier.

The returns-aware creative work deserves its own callout. We pulled return reason codes from Shopify against the ad-creative IDs that drove each purchase and found that one hero concept, a flowy oversized silhouette, was driving 38% of returns single-handedly. Customers who clicked that ad were expecting a different cut than the product delivered. Rewriting the creative with honest fit messaging dropped that ad's return rate from 41% to 14% inside two months.

Results

(See the full headline-results grid at the top of this page.)

What we learned

The creative-testing system is the agency moat for DTC in 2026. The brands that compound are the ones that ship 8-15 new concepts per week with disciplined testing and loss-memo discipline. Brands that ship 2-3 concepts per week plateau within 12 months, almost without exception.

The other lesson, returns-aware creative, is the underleveraged surface for fashion specifically. Most brands optimize creative for first-purchase CPA without measuring downstream return rate. The brands that build the loop find 4-8 points of margin sitting in the same creative pipeline that drives acquisition.