This Meta Andromeda e-commerce case study documents how a direct-to-consumer skincare brand reduced its cost per acquisition by 41% in 90 days — not by finding a new audience hack, but by fundamentally rethinking what signals Meta's Andromeda system actually needs to optimize. The brand abandoned years of accumulated audience architecture, stripped its campaign structure back to near-zero, and rebuilt entirely around creative as the primary targeting input. What followed was one of the most significant performance turnarounds the brand had seen in three years of paid social.
The Brand, the Problem, and What Was at Stake in This Meta Andromeda E-Commerce Case Study
The brand in question — a DTC skincare company generating between $4M and $8M annually in revenue — had been running Meta ads continuously since 2020. Over those years, the account had accumulated the hallmarks of a legacy campaign structure: 14 active ad sets, each targeting a different custom audience or interest cluster, with creative rotated in and out based on gut instinct and monthly performance reviews. The team knew every audience segment by name. They had look-alike stacks built off their best buyers, retargeting sequences layered over cold traffic campaigns, and a creative library of over 200 assets categorized by product line.
On paper, the account looked sophisticated. In practice, it was hemorrhaging efficiency. By Q4 2025, CPA had climbed from $38 to $64 over an 18-month period, while return on ad spend (ROAS) had dropped from 3.8x to 2.1x. The brand was spending approximately $85,000 per month on Meta and seeing diminishing returns with each budget increase. The marketing director described the situation plainly: every time they tried to scale, performance collapsed within two weeks.
The timing mattered. Meta's Andromeda ranking system — the AI-driven auction infrastructure that determines ad delivery, reach, and cost — had been progressively de-emphasizing manual audience signals in favor of creative-driven quality signals throughout 2024 and into 2025. Brands still operating under the assumption that audience segmentation was the primary lever for performance were increasingly fighting against the system rather than with it. This brand was a textbook example of that misalignment.
"We had spent three years building audience architecture that Andromeda was quietly rendering obsolete. The more we refined our targeting, the more we were restricting the very system that could find our best customers better than we ever could manually."
What was at stake was not just a 30-day performance blip. If the brand couldn't reverse its CPA trajectory, it faced reducing spend to unprofitable levels or exiting the Meta channel entirely — a significant risk for a company where paid social drove roughly 60% of new customer acquisition.
The Strategic Decision: What They Did — and What They Refused to Do
After an audit in late October 2025, the growth consultancy working with the brand presented a counterintuitive recommendation: do less audience work, not more. Instead of testing new lookalike seeds, introducing interest stacking refinements, or experimenting with Meta's Advantage+ audience expansion controls, the recommendation was to collapse the campaign structure almost entirely and redirect all creative, budget, and analytical attention toward what Andromeda actually optimizes around — ad quality signals derived from creative engagement, predicted click-through rate, and downstream conversion behavior.
The strategy was explicitly built around the principles of a Meta Andromeda ad targeting strategy: treating the algorithm as the targeting layer and treating creative as the primary communication channel between the brand and Meta's ranking system. This meant one Advantage+ Shopping Campaign structure, broad targeting with no manual exclusions beyond basic geographic constraints, and a disciplined creative testing system designed to generate clear signal data — not creative volume for its own sake.
Equally important was what they decided not to do. They did not launch a new retargeting funnel. They did not create audience-specific creative variants. They did not attempt to game the algorithm with engagement-bait formats. They resisted pressure from internal stakeholders to maintain the legacy structure as a "safety net" alongside the new approach — a compromise that would have muddied the signal data and made learning impossible. The team committed to a clean break.
The budget during the transition was reduced temporarily from $85,000/month to $60,000/month — a deliberate choice to protect profitability during the learning phase while the system recalibrated around new signals.
Implementation: 90 Days, Three Phases, One Big Bet
The implementation ran from November 2025 through January 2026, structured across three distinct phases, each with specific objectives and defined exit criteria.
| Phase | Duration | Primary Objective | Key Actions |
|---|---|---|---|
| Phase 1: Structural Reset | Weeks 1–3 | Collapse legacy structure, establish new campaign architecture | Archive 12 of 14 ad sets; launch single ASC campaign; pause all retargeting |
| Phase 2: Creative Signal Generation | Weeks 4–8 | Identify high-signal creative concepts through systematic testing | Launch 18 distinct creative concepts; enforce $1,500 minimum spend per concept before evaluation |
| Phase 3: Scale Around Winners | Weeks 9–13 | Concentrate budget on proven creative signals; restore spend to prior levels | Consolidate to 6 winning concepts; scale budget to $90,000/month; introduce iterative creative variations |
The creative testing protocol in Phase 2 was the operational heart of the strategy. Rather than testing execution variables (color, CTA button text, font size), the team tested concept-level differences: problem-first narratives versus transformation-first narratives, social proof formats versus authority formats, and short-form video (under 15 seconds) versus medium-form video (30–45 seconds). Each concept was built with a single clear hypothesis tied to a specific purchase barrier identified through customer research.
The evaluation framework followed the Meta Ads test and scale strategy 2026 playbook: concepts were evaluated on a composite score combining hook rate (3-second video views divided by impressions), hold rate (ThruPlay divided by 3-second views), and cost per initiated checkout — not just surface-level CTR. Any concept that failed to hit threshold benchmarks on hook rate and hold rate was paused regardless of its downstream CPA, because weak upstream signals were understood to generate poor Andromeda quality scores that would inflate costs over time.
By the end of Phase 2, 6 of the 18 concepts had cleared all three evaluation thresholds. These became the foundation of Phase 3 scaling.
Results: The Numbers Before and After
The headline result — a 41% reduction in CPA — was measured by comparing the average CPA across the 90-day period preceding the restructure (August through October 2025) with the average CPA across the final four weeks of Phase 3 (January 2026). The comparison was intentional: the team explicitly excluded the Phase 1 and early Phase 2 data from the headline metric because performance during structural transitions is not representative of steady-state results.
| Metric | Pre-Restructure (Avg, Aug–Oct 2025) | Post-Restructure (Jan 2026) | Change |
|---|---|---|---|
| Cost Per Acquisition (CPA) | $64 | $37.76 | –41% |
| Return on Ad Spend (ROAS) | 2.1x | 3.6x | +71% |
| Hook Rate (avg across active creatives) | 18% | 34% | +89% |
| Monthly Ad Spend | $85,000 | $90,000 | +6% |
| New Customer Acquisitions (monthly) | ~1,328 | ~2,384 | +79% |
| Active Ad Sets | 14 | 1 (ASC) | –93% |
The ROAS improvement of 71% was considered the more strategically significant result by the brand's leadership, because it validated the thesis that the legacy structure had been actively suppressing auction efficiency. Spending $5,000 more per month and acquiring 79% more customers represented a fundamental shift in unit economics, not an incremental gain.
Notably, the improvement in hook rate — from 18% to 34% — preceded and predicted the CPA improvement by approximately three weeks, reinforcing the team's belief that creative quality signals function as leading indicators of auction cost. When creative earns attention, Andromeda reduces the cost to compete for delivery.
"The hook rate improvement showed up in our data first. We knew CPA was going to fall before it fell — because we could see Andromeda rewarding the creative with cheaper delivery even before the conversion numbers moved."
Key Learnings: What Worked, What Failed, What Surprised Everyone
What worked: Concept-level creative testing — not execution-level — was the highest-leverage activity in the entire 90-day period. The six winning concepts were not more polished or higher-production than the losers. They were more strategically differentiated in terms of the specific customer belief or purchase barrier they addressed. Andromeda rewarded relevance and engagement depth, not production value.
The single ASC campaign structure eliminated audience overlap and budget fragmentation that had been silently inflating CPMs across the legacy structure. Industry practitioners widely observe that overlapping ad sets cause brands to compete against themselves in Meta's auction — a problem invisible in standard reporting but materially costly.
What failed: Two of the 18 creative concepts in Phase 2 were built around influencer-style testimonial formats that the team had expected to perform strongly based on historical account data. Both failed to clear the hook rate threshold (below 20% versus the 28% target), and their downstream CPAs were 60–80% above the campaign average. Historical creative performance data from a legacy account structure proved to be a poor predictor of performance in the new system — the audience context had changed too significantly.
What surprised everyone: The single most surprising finding was the speed at which Andromeda responded to creative quality signals. The team had budgeted for a 6–8 week learning phase before meaningful performance data would emerge. In practice, concepts that were going to work became identifiable within 7–10 days and $1,200–$1,500 in spend. The algorithm's ability to rapidly classify and route creative quality was faster than anyone on the team had anticipated — which compressed the testing timeline and allowed Phase 3 scaling to begin ahead of schedule.
How to Replicate This: An Actionable Checklist
The core mechanics of this case study are transferable to most DTC e-commerce accounts running Meta with $30,000 or more in monthly spend. The sequence matters as much as the individual steps — do not attempt to jump to Phase 3 without completing the signal-generation work in Phase 2.
- Audit your current structure for audience fragmentation. If you have more than 4–5 active ad sets targeting different segments, you are almost certainly experiencing auction overlap that inflates costs. Map all active audiences and identify overlap before restructuring.
- Identify your actual purchase barriers through customer research. Survey recent buyers, review negative reviews, analyze abandoned cart behavior. Each creative concept you test should map to a specific, documented barrier — not a creative preference or aesthetic choice.
- Collapse to one ASC campaign with broad targeting. Remove manual interest targeting. Remove lookalike stack structures. Allow Andromeda to function as the targeting layer. This feels counterintuitive but is the foundational prerequisite for everything that follows.
- Design concept-level creative tests, not execution-level tests. Test fundamentally different narratives, value propositions, and emotional angles — not different fonts or CTA colors. Aim for 12–20 distinct concepts in Phase 2, budgeting a minimum of $1,200–$1,500 per concept before evaluation.
- Define your evaluation criteria before launching. Set hook rate, hold rate, and cost-per-key-event thresholds in advance. Do not move goalposts mid-test. Kill concepts that miss threshold even if their surface-level CPA looks acceptable — poor upstream signals will inflate costs at scale.
- Monitor hook rate as a leading indicator. When hook rate improves, expect CPA improvement 2–4 weeks later. Use this as an early warning system to identify winning concepts before spending budget to statistical confidence on downstream conversion data alone.
- Scale budget only after identifying at least 4 proven creative concepts. Scaling budget into an unproven creative pool forces the algorithm into inefficient exploration. Concentration of budget around proven signal-generating creative is what unlocks efficient scaling.
- Build an iterative creative pipeline, not a replacement cycle. Once winners are identified, develop close variations (different hooks, different opening frames, different value proposition sequences) rather than replacing winners with entirely new concepts. Andromeda favors account-level consistency in creative signal quality over time.
The single most important mindset shift this case study represents is treating creative as infrastructure rather than content. Each ad is not a piece of communication — it is a signal sent to an AI ranking system that determines whether your brand competes efficiently or expensively in Meta's auction. Build creative to generate the right signals, and the system will deliver customers at a cost that manual audience architecture never could.
Frequently Asked Questions
What is Meta Andromeda and why does it matter for e-commerce advertisers?
Meta Andromeda is the AI-driven ranking and delivery system that powers Meta's advertising auction, determining which ads get shown to which users, at what cost, and with what frequency. For e-commerce advertisers, it matters because Andromeda uses creative quality signals — hook rate, engagement depth, predicted conversion probability — as primary inputs for determining ad delivery cost and reach. Advertisers who optimize creative to generate strong Andromeda signals consistently achieve lower CPMs and CPAs than those who rely primarily on manual audience targeting to drive performance.
How long does it take to see results after restructuring a Meta account around Andromeda signals?
Most practitioners report that initial signal data becomes meaningful within 7–14 days of launching a restructured campaign, provided each creative concept receives at least $1,000–$1,500 in spend before evaluation. Full steady-state performance — where the algorithm has stabilized delivery patterns around proven creative — typically emerges within 6–10 weeks. During the first 3–4 weeks, expect performance variability as Andromeda explores delivery patterns for the new creative set.
Should I completely eliminate retargeting when switching to an Andromeda-first Meta strategy?
Not necessarily permanently, but during the restructuring phase, pausing retargeting allows the ASC campaign to operate without interference and generates cleaner signal data. Once the primary campaign structure is stable and performing, a lightweight retargeting layer — typically 10–20% of total budget — can be reintroduced without meaningfully disrupting the core campaign's signals. The key is avoiding the legacy approach of treating retargeting as a separate, heavily-segmented funnel that competes with prospecting for budget and signal clarity.
What creative formats perform best in Meta's Andromeda system for DTC brands?
Industry observations consistently point to short-to-medium video (10–45 seconds) as the format that generates the strongest Andromeda quality signals for DTC e-commerce, primarily because video provides more engagement data points (hook rate, hold rate, completion) for the system to evaluate. Static images remain effective for brands with strong visual identity and clear value propositions, but they generate fewer quality signals per impression than video. The format matters less than the strategic clarity of the concept — a mediocre video concept will underperform a strong static concept regardless of format.
How many creatives do I need to run an effective Andromeda-focused Meta strategy?
For testing phases, 12–20 distinct concepts is an effective range — enough to identify statistically meaningful winners without creating a signal fragmentation problem within a single campaign. In steady-state scaling, concentrating budget on 4–8 proven creative concepts (with iterative variations being introduced monthly) tends to produce more consistent results than maintaining large creative libraries. The quality and conceptual diversity of creatives matters significantly more than raw volume.