Client identifiers removed. Metrics directionally preserved.
This is a real BSD audit with the brand and exact figures anonymised. The pattern, the diagnostic logic, and the directional numbers are reported as they were found.
The headline finding is simple. Roughly 90% of the ad spend was in dynamic product ads and catalogue carousels, not in real creative. That single fact explains almost everything else that was wrong with the account, because a brand that puts nearly all of its budget into demand harvesting has no engine producing the demand in the first place, and no testing layer telling it what to build next.
Fast fashion, trading across two major markets, on a custom platform rather than Shopify.
Spread across Meta, Google and TikTok in two markets. Serious budget, structurally misallocated.
Almost the entire account sat in catalogue ads. Real creative was getting close to nothing.
Acquisition had tipped into losing money on the first order, and retention was too weak to fund it.
The pattern.
DPAs work on cold audiences in fashion. That is not in dispute. You can spend on catalogue ads to people who have never heard of you and it will return. The problem is that it is not scalable on its own, because it depends entirely on something further up the funnel creating the demand that the catalogue then converts. Here, that something barely existed.
The reason brands overspend on DPAs is that the ROAS on them looks good, so the temptation is to keep pushing budget in. It looks good because the format sits right at the bottom of the funnel. It takes last-click credit for purchases that other creative actually drove. The strongest asset in this account was a video that rotated through a range of products and pulled a click-through rate near 7%. That ad was doing the work. The DPA was standing at the checkout claiming the sale.
Why it was happening.
Underneath the DPA problem was a structural one. The creative testing campaigns that did exist were running on CBO, so budget concentrated into one or two ad sets and starved everything else. The result was extreme. The top 1% of ads held around 80% of spend. Every other ad the team built got a dollar a day and never got a real read, which means the account could not tell you why anything underperformed. Not hook rate, not hold rate, not click-through, not landing page. The ads never spent, so there was nothing to diagnose.
Worth being precise here, because it is the part most teams get wrong. The brand was launching enough creative. It was pushing close to 200 ads a month, which is above what the volume model says it needs at this revenue. The constraint was never velocity. The constraint was that the velocity led nowhere, because the budget never reached the new creative and the catalogue absorbed it instead.
You do not scale an account on one or two ads. You scale it by building a wide base of ads that can each hold real spend, and this account had no base.
What it was costing.
The commercial read was already negative. CAC had been roughly flat for a year, but gross profit on the first order was falling, driven by average unit retail compressing around 25% and dragging AOV down with it. So the same customer was costing the same to acquire while delivering less margin. Run those two together and the 30-day LTGP:CAC had compressed from comfortably above 1 to under it, which means the brand was losing money on acquisition. That is survivable if retention is strong enough to fund it. Here the 90-day and 12-month repeat rates were running roughly 40% to 50% below the fashion portfolio average, so there was nothing behind the loss to make it back.
Google made it worse, and in an unusual way. On this account Google was under-attributing, reporting close to a 0.9 against a P&L position that should have read higher. That happens on a tiny share of accounts. It did not make Google profitable. Whether the read was a tracking fault or real, once you discount Google for its usual lack of incrementality the channel was unprofitable either way, and a large slice of its budget was being burned on branded terms the brand owned with almost no competition.
A flattering ROAS on a bottom-of-funnel format is the most expensive number in the account.
If you scale spend behind an ad that is harvesting demand other creative created, you dilute the return with no incremental gain, and you feed the creative team a false signal that the wrong ad is the winner. The feedback loop breaks in both directions at once.
The fix.
- Move 30% to 40% of catalogue spend into creative testing. The DPA layer was oversized for the demand being created above it. Reallocating it funds a real testing engine without raising total spend.
- Break the CBO concentration. Stop letting one or two ads hold the account. Build a wide base of ads that each get enough spend to produce a clean read, so the creative team gets learnings back instead of silence.
- Fix the measurement before scaling anything. Strip 1-day view from decisioning, double-track against incremental attribution, exclude existing customers from top-of-funnel, and pull spend out of branded search and cross-bidding PMax that was never incremental.
The point underneath all three is the same. Get profitable on the first order again before scaling, because retention is not strong enough to subsidise a negative acquisition position, and a brand carrying this much concentration risk is one creative fatigue cycle away from the account collapsing with nothing benched to replace it.
What changed.
The diagnosis reframed the brief entirely. The team had assumed they had a creative volume problem and were about to spend into it. They did not. They had a structure and measurement problem wearing a creative problem's clothes. The real constraint was that almost the entire budget sat in a format that harvested demand it never created, with no testing layer learning anything, on top of a P&L where the first order had quietly stopped making money. Fix the allocation, fix the read, and the volume they were already producing finally has somewhere to go.