Meridian Glow — Skincare brand reduced wasted Meta spend 34%

The situation

Meridian Glow is a three-year-old DTC skincare brand running a clinical-grade serum line with three hero SKUs (a vitamin-C serum, a peptide night cream, and a retinol booster). They do roughly $140k/mo in revenue, split across a Shopify storefront and a Klaviyo-driven email program with 62k subscribers and a 38% flow open rate. At the time of onboarding, their paid-media mix was about 70% Meta, 20% Google, and 10% TikTok organic amplified with a small paid spend. Their stack was standard for brands at this scale: Shopify, Klaviyo, a Meta Pixel installed via the Facebook & Instagram app, Google Analytics 4, and a couple of spreadsheets stitched together every Monday morning.

The founder handled strategy; a part-time media buyer ran Meta day-to-day at about 15 hours a week. Decisions were made weekly off two dashboards: Meta Ads Manager ROAS at the ad-set level, and a handmade Google Sheet that pulled Shopify revenue next to Meta spend. The buyer trusted Meta's numbers enough to scale ad sets that appeared to hit 3.5x+ ROAS and pause ones under 2.0x. This is the default operator workflow on DTC Twitter and in the skincare subs — and it is exactly the workflow that has been quietly broken since iOS 14 tightened the screws on signal, and broken more severely since Meta's January 2026 attribution window reduction cut view-through attribution from 28 days to 1 day on the Ads Insights API.

Industry-wide, the pattern is well-documented: "a Meta Ads Manager dashboard showing 50 conversions from a campaign while the Shopify backend shows 30 actual orders is now a common discrepancy," with attribution accuracy reported to have dropped 40-60% through 2026. Skincare specifically tends to sit at the worse end of that distribution because the category leans hard on video creative, which Meta's old engage-through window used to credit generously — post-reclassification, a lot of that claimed revenue simply vanished from the reporting surface without the underlying conversions actually changing.

The problem

Two things were wrong, and they compounded. First, the Meta Pixel was installed but Meta Conversions API (CAPI) was not. Meta had been filling signal gaps with modeling, and the operator had no way to see how much of the reported conversion volume was modeled vs. observed. When Admaxxer connected and ran the first CAPI match rate audit, the number came back at 41% — meaning for 59% of Meta-reported purchases, Meta was guessing which ad set, which creative, and which audience to credit. The optimizer was training on that guesswork, which pushed budget toward ad sets that happened to coincide with organic lift rather than ad sets that actually drove purchases. Running Conversions API alongside Pixel — with proper event deduplication — has effectively become table stakes in 2026; pixel-only tracking is unreliable at best and actively misleading at worst.

Second, there was a 28% gap between Meta-reported revenue and actual Shopify revenue for the paid-media window. Meta's dashboard showed $82k in attributed revenue on $24k of spend — a reported 3.4x ROAS — while Shopify showed only $58k in orders that could reasonably be matched to any paid touch. Blended MER (total revenue / total paid spend) was 2.1x, not 3.4x. Worse, when sorted by ad-set-level true contribution, three of the five "best" ad sets in Meta Ads Manager were actually the worst five once modeled-conversion inflation was removed. The buyer had been scaling those three ad sets weekly — adding $300–$500 in daily budget each — based on reported ROAS that was chasing ghost revenue. The brand was paying for Meta's uncertainty twice: once in the wasted spend, and once in the opportunity cost of starving the ad sets that were actually working.

What they did with Admaxxer

• Step 1 — Install the pixel. Dropped the Admaxxer pixel on the Shopify theme in 10 minutes via a one-line snippet injection. First-party events started firing to Tinybird within the hour and revenue started reconciling against Shopify orders. The pixel sends PageView, ViewContent, AddToCart, InitiateCheckout, and Purchase with a shared event_id so server-side events can deduplicate downstream.
• Step 2 — Paste the Meta long-lived token. Used the paste-token flow (no app review required, no waiting on Meta's business verification queue), connected Google Ads via OAuth with a refresh token, and let the initial 90-day spend backfill run overnight. Tokens landed in the database encrypted AES-256-GCM at rest, never visible in logs.
• Step 3 — Run the CAPI match rate audit. Clicked the Conversions API tab in the dashboard, saw 41% match rate, and followed the guided server-side setup: enabled Meta CAPI events via Admaxxer's pass-through, added the missing event_id deduplication so pixel and server events reconciled to one conversion each, and hashed the customer email + phone with SHA-256 per Meta's documented parameter requirements. Match rate climbed to 72% in 72 hours as the new events populated, then 88% after a second pass a week later where the team added fbp and fbc cookies to the server-side payload.
• Step 4 — Ask the Claude agent for the real winners. Opened the analytics chat panel and asked: "Show me every Meta ad set running in the last 30 days with >$500 spend, sorted by blended MER contribution minus modeled-conversion inflation." The agent pulled from the query_metrics tool against the ads_ltv and capi_match_rate Tinybird pipes, returned a ranked list, and flagged three ad sets at the top of the Meta-reported leaderboard that were actually below-breakeven on real revenue. The chat UI rendered the table inline and let the buyer click through to the ad-set detail view.
• Step 5 — Pause the losers, scale the hidden winners. Used the agent's update_campaign tool with explicit confirmation to pause four ad sets consuming roughly $12k/mo at a true MER of 1.1x–1.4x. The confirmation screen showed each ad set's current daily budget, the impact of pausing, and a one-click approve. In the same conversation, the agent surfaced three ad sets that Meta had been starving (ROAS looked flat at 2.2x in Ads Manager but true MER was 3.8x) — scaled those to absorb the freed budget in 25% daily-budget increments across 10 days to avoid tripping Meta's learning-phase reset.

The results

After 60 days on the Pro plan:

The headline: they cut 34% of Meta spend without losing revenue, because the spend they cut was genuinely unprofitable and Meta's optimizer had been misallocating against modeled signal.

Why this worked

Three Admaxxer capabilities stacked. First, the CAPI match rate monitor turned an invisible problem into a number on a dashboard — once the operator saw 41%, the next step was obvious. Prior to Admaxxer, there was no single surface in their stack that even computed match rate; it was a number that lived deep inside Meta Events Manager and required a buyer who knew to look for it. Second, blended MER computed from first-party pixel revenue against Meta + Google spend cut through platform over-attribution. Blended MER is one number you can trust precisely because both its inputs are observed — total revenue from your own pixel, total spend from the ad-platform APIs — not modeled. Third, the Claude agent turned "which ad sets are secretly losing money" from a two-hour spreadsheet exercise into a 30-second question, and the destructive-gated update_campaign tool let the operator pause with confirmation in the same flow.

This is the pattern Admaxxer was built for. Platforms optimize against the signal they can see, and since Apple's ATT rollout (plus the January 2026 Meta attribution window reduction), most DTC brands have been flying on 40-50% observed data plus modeled guesses. Fixing match rate is the single highest-leverage intervention available to a paid-media operator in 2026 — it typically unlocks 15-30% better optimizer performance without changing a single creative. The fact that Meridian Glow's revenue held steady at $141k while spend dropped 34% is the signal: they were not revenue-limited by spend, they were revenue-limited by how poorly that spend was being allocated.

The secondary unlock is the operator workflow change. Before Admaxxer, the buyer spent four hours every Monday rebuilding the same Google Sheet. After Admaxxer, that sheet is rendered automatically, and the buyer spends those four hours testing new creative instead. Reallocating buyer attention from reporting to execution is the kind of compounding return that shows up three quarters later in the testing velocity, not the immediate reporting line.

What other DTC skincare brands can learn

• Install CAPI properly from day one. The Meta Pixel alone is no longer enough — aim for 75%+ match rate before trusting the optimizer, and ideally 85%+ if you want the optimizer training signal to be directionally correct.
• Never scale off Meta-reported ROAS alone. Compare it to blended MER weekly; a >20% gap means Meta is over-claiming and your budget decisions are off.
• Watch the "invisible winners" — ad sets Meta starved because its model couldn't attribute them cleanly. Those are usually where the real growth lives.
• Use the Claude agent for ad-set audits instead of building a spreadsheet. One well-written prompt replaces a recurring weekly task.
• Treat modeled-conversion inflation as a margin leak, not a reporting quirk. Every modeled purchase that never happened is budget allocated against a ghost.
• Skincare in particular benefits — high-AOV SKUs mean one misattributed $120 purchase in a modeled window moves the ad-set scoring meaningfully, and the category's heavy reliance on video creative was disproportionately hit by Meta's engage-through reclassification.
• Revisit CAPI match rate monthly, not once. Server-side event infrastructure drifts — a Shopify theme update, a GTM container change, a Cloudflare rule tweak can all silently break match rate by 10-20 points in a week.