When this DTC skincare brand came to us, their blended customer acquisition cost had climbed 65% in eight months while revenue stayed flat. The founder's diagnosis was 'iOS killed our targeting.' The real diagnosis was messier and more fixable: a Meta account structured for a platform that no longer exists, creative that had not meaningfully changed in a year, and tracking that fed the algorithm noise instead of signal.
Ninety days later, blended CAC was down 40% and monthly revenue was up 32% on roughly the same spend. Here is exactly what we did, in the order we did it, including the two weeks where things got worse before they got better.
Weeks 1 and 2: Fix the Signal Before Touching the Spend
The account was optimising toward a pixel that double-counted purchases and missed roughly 30% of conversions entirely. Before restructuring anything, we implemented Meta's Conversions API with proper event deduplication, rebuilt the event taxonomy, and reconciled the numbers against the store's actual order data.
This step is unglamorous and it is also the reason everything after it worked. An algorithm learning from wrong data optimises confidently toward the wrong customers. Two weeks of tracking work bought every later dollar of spend a better teacher.
Weeks 3 to 5: Collapse the Account, Rebuild the Audiences
The account had 47 active ad sets, most spending too little to exit the learning phase, many bidding against each other for the same users. We collapsed it to a structure with three prospecting pools, a two-stage retargeting ladder, and strict exclusions so no user could be hit from two campaigns at once.
- Broad prospecting with creative doing the targeting work
- Interest-stacked prospecting for the two highest-LTV customer segments
- A lookalike pool seeded from 180-day high-value purchasers, not all purchasers
- Retargeting split by engagement depth, with frequency caps and a 14-day exit
Weeks 4 to 12: The Creative Testing Engine
This is where the CAC actually fell. The brand had been running four ads, all polished studio work, all fatigued. We committed to shipping eight to twelve new creative variants per week: UGC testimonials, founder-story videos, before-and-after statics, problem-agitation hooks, and price-anchoring angles.
Every variant launched with a hypothesis and a kill threshold. Anything below the account's median cost-per-purchase after a statistically meaningful sample got cut without sentiment. The winners revealed something the brand did not expect: raw, phone-shot UGC from customers in their 40s outperformed the studio work by nearly 3 to 1, because that is who the actual buyer was, not the 25-year-old in the brand deck.
The account did not need more budget. It needed more shots on goal and the discipline to count them honestly.
What Moved the Number
Attribution for the 40% is roughly: a third from signal quality and structure, letting the algorithm find buyers it was previously blind to; half from creative velocity, because fresh angles kept CPMs and hook rates healthy; and the rest from budget discipline, moving spend daily toward marginal winners instead of weekly toward last week's averages.
None of this is secret. It is a system, applied consistently, measured against blended numbers instead of platform-flattered ones. That is the entire trick.