Weekly Performance Report

Point of Sale is excluded from all reported revenue by the account's saved sales-channel policy. The "matrixify app" channel carries bulk-imported historical orders and did not contribute this week.

Net merchandise revenue, refund-correct, incl. tax and excl. shipping (shipping cannot be defensibly allocated to a product line).

Reading the trading week

• Basket mix did the work, not traffic. Orders +12% but revenue +36% - the delta is entirely AOV. Midnight Ritual Set (€2,330, one unit), Midnight Renewal Serum (5 units, +€1,468 w/w) and Polar Eye Concentrate (10 units, +€1,638 w/w) drove it. Worth checking whether the July offer is pushing customers up the ladder, and whether that survives when the offer ends.
• Arctic Renewal Mask remains the volume engine but softened - 35 units, -15% w/w, €4,585. Its Limited Edition sibling added 16 units / €2,239 (+23%). The LE is the highest-ROAS creative subject on Meta and is under-supported everywhere else.
• Hong Kong AOV is €836 across 4 orders. Small volume, extraordinary basket - this is the VIC profile the ICP describes. Four orders is not a trend, but if HK/SG holds this basket it deserves a dedicated retention and concierge motion rather than broad acquisition.
• €5,545 across 20 orders has no country attached ("ZZ") in the POS-inclusive operational country view. This view does not reconcile to the saved store basis above, so use it for directional market mix only until the country field is fixed at source.
• Refunds at 1.9% of revenue (from 8.5% last week) is healthy for luxury skincare and suggests last week's spike was a one-off batch rather than a product or fulfilment issue. Worth confirming the cause so it does not recur.

Only three campaigns delivered at all, and only on 20 July. Revenue is Meta last-click attributed value. The single €420 purchase booked to Retention · Cross-Sell on 23 July is attribution catching up on a click from before the blackout.

Account structure - what is waiting to switch back on

The ad account carries 147 campaigns, of which only 3 are ACTIVE. The paused estate still holds live daily budgets - BFCM 2025 (€5.27k/day across 6 campaigns), Holiday Peak 2025 (€2.7k/day across 4), Christmas Essentials (€1.04k/day), Spring Launch 2026 (€498/day), Awareness Copenhagen 2026 (€1.7k/day). None of them should ever be re-enabled by accident.

• Before anything restarts, pause the two active “kill” ads. “Clinical Proof” and “Cloudberry Cream Hero” have spent €1,316 in the last 30 days for zero purchases at 0.04% and 0.05% CTR. They are still ACTIVE and will start burning the moment delivery resumes. See the Creative tab.
• Restart at a controlled level, not at €1,494/day. Last week Meta spent €10.3k for €9.5k attributed at 0.93 ROAS - below the 1.5 target and roughly half Google's efficiency. Going straight back to the old run-rate reinstates a known-unprofitable spend line.
• Use the blackout as a measurement asset. Store revenue rose 36% with Meta at zero. One week is not proof, and last week was itself a weak comparison - but it is enough signal to justify a proper geo-holdout (e.g. Meta on in US/AU, off in UK/EU for 3 weeks) before committing Q4 budget. This is the single highest-value analysis available right now, and it is free.
• Warm audiences are 4.6× more efficient than cold. Over the last 30 days, warm-audience creative returned 4.97 ROAS at €62 CPA on €3.4k of spend; cold returned 1.09 at €271 CPA on €27.9k. The restart should be weighted far more heavily toward retention, DPA and cross-sell than the current split implies.

Conversions can be fractional because Google attributes fractionally across data-driven paths. Targets: ROAS 1.5, CPA €208.

Two distinct problems show up in this week's terms, and they need opposite fixes.

1. Brand intent bought at Shopping prices - move to Brand, negative in Shopping

These are people looking for Northlight, misspelling it. Shopping_2026 is paying €3.39–€31.75 a click for them. BRAND_US buys the correctly-spelled version of the same intent at €1.10.

2. Generic and competitor terms with no return - negative these

All in the Shopping "Hero Products" ad group, all at zero conversions. Competitor names (Competitor Epsilon, Competitor Zeta, Competitor Eta, Competitor Mu, Competitor Beta, Competitor Delta, Competitor Theta, Competitor Alpha, Competitor Iota) plus broad generics. The brand's own positioning is explicit that competitors are never named - buying their traffic at €3.32–34 a click is neither on-brand nor profitable.

Decisions for Google

• Reach US (video) - €1,743 this week, €3,537 over 14 days, zero conversions and zero attributed revenue at a 0.08% CTR. Either it is an upper-funnel play with an agreed brand-lift measurement and a fixed budget cap, or it is €7.5k a month with no answer attached. This needs an explicit client decision, not another week of drift.
• Shopping_2026 is the weakest converting line - ROAS 1.33 (below the 1.5 target, down from 1.57) and CPA €337 (63% above target) on €2,023. Roughly €291 of this week's spend went to terms that cannot convert. Cleaning the negatives alone should move it back above target without touching bids.
• PMAX_Feed_Only is the improvement story - ROAS 1.08 → 2.44 on 8% less spend, conversions +43%. This is where the marginal euro should go if Meta stays dark, not into Reach US.
• BRAND_US at 42.8 ROAS and €17 CPA is doing exactly what brand search should do - but note that half of Google's reported revenue this week sits in a campaign that harvests demand created elsewhere. As Meta comes back or awareness spend changes, this number will move for reasons that have nothing to do with the Search campaign itself.
• BRAND_HKSG CPA jumped from €19 to €175 - flagged automatically as critical, but it is one conversion on €175 of spend. Watch it, don't act on it.
• Feed hygiene: several top-spending Shopping items carry Chinese or untranslated titles while serving English-language markets, and brand values are inconsistent ("northlight" vs "northlight"). That suppresses relevance and query matching. A feed title and brand-field pass is a cheap, high-leverage fix.

What to do with email

• Segment by market before anything else. "July Offer one week left" sent to 25,649 on the main list returned €0.067 per recipient. The same offer sent to 3,233 US contacts returned €0.314 - 4.7× - with a 41% open rate and nearly double the click rate. That is the cheapest available win in the account: split the main list by market and language and re-run the next send.
• Click rate, not open rate, is the constraint. 10,791 people opened the Extreme prolongation email and 244 clicked (2.3% of openers). At luxury-skincare benchmarks that should be 8–15% of openers. The offer and the CTA hierarchy inside the email are what is failing, not the subject line or the send reputation.
• Browse Abandonment is broken. 122 recipients, 78 unique opens, 9 unique clicks - and zero conversions and zero revenue. The Polar variant is receiving traffic and the legacy variant (43 recipients) is also at zero. Check that the flow's conversion tracking and product links are intact before assuming it is a creative problem.
• Abandoned Checkout returned €0 on 50 recipients while Abandoned Cart returned €877 on 143. Checkout abandoners are further down the funnel and should convert better, not worse. Same diagnostic: tracking first, then content.
• Post-Purchase is quietly excellent - €1,023 from 173 recipients on a single click. That is subscription and reorder revenue landing through the flow. Worth expanding into a proper Saho-ritual replenishment sequence given the routine-adoption strategy.
• Review and Loyalty-Points flows returned zero revenue on 194 combined recipients. They are not revenue flows by design - but the Loyalty reminder at 5.9% click rate and no conversion suggests the redemption path is worth testing end-to-end.
• Unsubscribe rate on campaigns (0.16–0.22%) is safe, but the main list is absorbing three promotional sends a fortnight against a "no discount language, exclusivity-first" brand voice. Two of the three sends this fortnight were offer-deadline emails. Worth watching list health over the next month.

The same creative appears more than once when it runs in multiple ad sets - rows are ad-level, matching Ads Manager. CTR trend compares the second half of the window with the first.

Creative brief - what to produce next

• Three Arctic Renewal Mask Limited-Edition statics in the overhead / texture family that produced 7.1 and 7.9 ROAS. Exclusive-gift framing, no discount language, on-brand vocabulary ("a treasure", "Northlight Circle", "limited"). This is the account's proven angle and its two best assets are both fatiguing.
• Two "stat / proof" statics - stat_number returns 2.13 ROAS at €165 CPA on 16 ads, second-best hook by volume. Lead with the clinical claim in the visual, not in the body copy where nobody reads it.
• One testimonial static in the "hotel 5-stars" layout - 1.97 ROAS at €150 CPA and the highest CTR of any hook with real spend (1.14%). The existing pair is fatiguing.
• Stop commissioning long-form UGC video for cold prospecting until the format proves out on a controlled budget. Three UGC videos consumed €14,794 at a blended 1.03. The €2,442 "UGC Nora – Tired-Skin Reset" cut produced one purchase at 0.06% CTR.
• Retarget with what already works. Warm audiences at 4.97 ROAS / €62 CPA are barely being served - €3,432 of €31,352. Arctic Renewal Mask is the stated gateway product in the ICP; a cross-sell sequence from Arctic Renewal Mask buyers into Cloudberry Barrier Cream and Nordic Hydration Serum is the obvious next build.
• Kill the "Clinical Proof" and "Cloudberry Cream Hero" concepts entirely. €1,316 for zero purchases at 0.04–0.05% CTR is not a fatigue problem or a budget problem - the creative is not being looked at.

Not covered in this report

TikTok and any other paid channel, organic social, affiliate and influencer, retail and spa sell-through, inventory and stock cover (the inventory table is empty in the data source), subscription churn and cohort LTV, gross margin and contribution profit, logistics and fulfilment cost, and fraud/chargebacks. If any of these should appear in the weekly rhythm, they need a data source connected first.