DTC / Beauty
When attribution broke, we rebuilt for a world without IDFA.
How a DTC skincare brand got blended CAC back under target in one quarter after iOS 14.5 collapsed paid attribution.
CASE / 05
Confidential (Series A DTC skincare)
Q2 2022. The client had a good product and a six-month CAC trend going straight up. iOS 14.5 had been live for a year. Meta's reported ROAS and the bank account had stopped agreeing on the most basic question: is this spend working. Nobody on the team trusted the numbers, and nobody knew what to replace them with.
GEO
US · CA · UK
Setup
Full-funnel rebuild
Duration
16 weeks
Shipped
Q3 2022
What we were looking at
What we were looking at
Reported ROAS sat at 4.1x blended across Meta and TikTok. Actual blended CAC was up 67% year over year. Email revenue had been flat for three quarters. SMS was not in the stack. The team had been layering more creative on top of the same measurement system, and the measurement system had quietly stopped working somewhere around April 2021.
We opened with a financial reconciliation, not a creative audit. Media spend against shipped orders, by week, by channel, with no attribution model applied. The gap between reported and actual CAC was 38%. That number did not appear on any dashboard anyone on the team used.
Marketing mix modelling, not attribution
Marketing mix modelling, not attribution
We stopped trying to fix last-click. Robyn was set up on 104 weeks of spend and revenue data. The first run was ugly and non-converged. The second, after we cleaned the promo calendar and holiday flags, produced usable incrementality estimates by channel. Meta was doing about 60% of what the in-platform number claimed. TikTok was doing roughly what it claimed. Google Shopping was doing more.
We moved to weekly MMM-informed budget allocation. Nobody on the team was optimizing per-conversion inside Ads Manager anymore. It felt slow. It was correct.
First-party data, in the funnel not bolted on
First-party data, in the funnel not bolted on
The email list was growing 8% a month on a passive pop-up. We replaced the pop-up with a six-question routine quiz on the product detail page. The quiz result gated a personalised regimen and a welcome code. List growth went to 24% a month inside four weeks.
SMS went live on Attentive with flows triggered by quiz answer, not by site behaviour. Someone flagging dryness got a different sequence from someone flagging acne. Opt-in rate sat at 31% of quiz takers, which was 3x the industry benchmarks the client had been quoted.
Creative, but honest about what we can measure
Creative, but honest about what we can measure
Production was pumping out 40 variants a month. Six of them were doing the work. The rest were drag. We moved to six concepts a month with a written hypothesis per concept, and a geo-holdout test behind each one. DMAs matched on baseline revenue trend, then split. If a concept could not beat a two-week holdout in one DMA, it did not scale. Three of the first six did. Two were killed. One moved the needle harder than expected.
Launch With Us