Case study · Paid media & client strategy · May – November 2026

The account operating system behind a premium sleepwear relaunch

A premium sleepwear & intimates brand came back online in May 2026 after an acquisition and a dormant year. This piece stitches seven client deliverables into one story: how the relaunch was measured, corrected, run week to week, extended into retention and feed operations, and budgeted into Q4.

Each chapter is something the Client actually received. What it shows is the work of running the account: process, budget management and optimization, and client communication.

Google Ads (PMax, Search) · Meta · Klaviyo · Merchant Center · Shopify7 deliverables · 6 chapters
How to read the numbers. All revenue and spend figures are indexed to one base: May 2026 DTC gross sales = 1,000. Spend uses the same units, so spend ÷ sales, ROAS and every percentage are real. Customer-value figures use a second base: 2024 average first order = 100. Product, campaign and partner names are replaced with generic labels.
Chapter 1 · Executive package, June 9

Relaunch economics: one honest number

Five weeks after the relaunch, the leadership team wanted to know whether the brand was healthy. The answer had to work for finance and for marketing, and it couldn't depend on which attribution model you picked.

Measurement frameworkExecutive communicationUnit economics
Ad spend per sales dollar, May
47%
healthy level ~20%; break-even 50–57%
First-time buyer share
15% → ~50%
relaunch week to early June
Cost to win a customer
−40%
in four weeks, toward 2024 lifetime value
Google sales per ad dollar
0.81 → 1.95
over four weeks, while spend fell ~40%

The yardstick: ad spend ÷ total sales

One blended ratio that nobody can spin. It is the number the Client's team already used.

May ran at 47%. That made money, but it sat well above the ~20% level a business at this margin needs to be sustainable. The plan brings it under 20% by Q4.

Three ways to credit the same advertising

Share of total website sales credited to paid media, May 1 – June 7

Depending on how strictly you credit the first ad, the same spend gets 36% to 50% of sales. Because any single figure is a judgment call, the report gives the range and steers by the blended ratio.

What a customer costs vs. what a customer is worth

Indexed: 2024 average first order = 100. The line is cumulative spend per 2024 customer. The dashed lines are today's cost to acquire one.

2024 cumulative value per customer24-month rangeCost to acquire, start → latest

Acquiring a customer cost about 1.1x their first order (111 vs. 100), and the first order's contribution after product and shipping is about 67. So the first order doesn't pay back the acquisition cost, and the profit comes from repeat purchases. 2024 customers reached 1.5x their first order within a year. Over four weeks acquisition cost fell from 147 to 92, which put it below first-year value. 2024 is the benchmark because 2025 was dormant, and holiday gift buyers are excluded because they rarely return.

Who brings in whom

SourceFirst-timeReturning
Google ads~65%~35%
Meta prospecting~100%–
Meta retargeting–~100%
Email, direct, organic searchlowhigh

About 75% of new-customer sales came through paid media, and about 67% of returning-customer sales came through owned channels. Paid budget goes to growth. Repeat buyers cost almost nothing to bring back, and the retention tools for that were barely switched on (see Chapter 4).

Google: getting more efficient as it learned

Bars show sales per ad dollar by week. Labels show weekly spend indexed to the first week (= 100). Over the same stretch, Meta's cost per new-customer purchase fell 64%.

The recommendation the Client signed off on: stay the course, take spend down from 47 cents per sales dollar to under 20 cents by Q4 along a published monthly path, and switch on the retention tools that raise what each customer is worth. The cost to win a customer and proven repeat behavior then meet in the middle.
Chapter 2 · May → June recap & July pacing

The correction: missing gross, beating net

June was planned as the correction month: cut spend hard and accept a lower top line in exchange for efficiency. Gross sales came in under target. The recap's job was to show the Client why that still counted as a win, using their own plan's line items.

Budget managementPlan-vs-actual reconciliationExpectation setting
Total ad spend
−28%
MoM (469 → 339 indexed)
Blended ROAS
2.13x → 3.33x
target 3.30x · break-even 1.65x
Spend ÷ gross
47% → 30%
plan 30% · inside the 28–32% band
Net sales vs. plan
+1.2%
while gross reached 91% of target

Where the plan over-budgeted leakage

June, % of gross sales

ActualPlan

The plan assumed 12% discounts and 8% returns. Actuals were 5.6% and 5.2%. The leakage avoided was worth more than the gross shortfall, so net sales came in ahead of plan.

June: plan vs. actual (indexed)

LinePlanActualVar.
Gross sales1,2411,127−9%
Discounts−149−63+86
Returns−100−59+41
Net sales9931,005+1.2%
Ad spend375339−10%

Same net definition on both sides (gross less discounts and returns, before fees) so the comparison is like for like. Orders rose 16.6% MoM with AOV flat, so the growth came from volume and not from markdowns.

Holding the spend-to-gross path

The recalibrated plan set a monthly spend ÷ gross target. Points are actuals as reported to the Client.

Plan ratioActualTarget zone ≤ 24%

The plan has pre-agreed triggers. If June lands above 32%, adjust mid-month. Below 28%, lean back in. If August runs above 25%, do a structural review. With the triggers written down, a budget change is a decision the Client already agreed to, not a surprise.

Mid-July pacing check

  • Blended ROAS was 4.05x with spend at 25% of gross. That matched the reinvestment plan and ran 11 points leaner than the same days in June.
  • Demand ran +19% ahead of June on a same-day basis.
  • Spend was pacing to only 76% of plan, and gross to 78% of target.
  • Returns hit 7.6% against the 8% earmark. That's in line with the expected 30-day lag, so the guidance was to watch it, not react to it.

The client-communication point

"The plan-vs-actual gap is drift, not strategy. That decision should be made explicitly."

New collections were arriving in late summer, so some July dollars might have been better held for those launches. The recap put that trade-off in front of the Client as a decision for them to make, instead of letting under-pacing quietly decide it.

Chapter 3 · Weekly report, Sep 30 – Oct 6

The weekly operating rhythm

Every week the Client got a one-page read: what happened and why, which numbers to trust, what we changed, and who owns the next step. This is a representative week. It was soft, and it was the week the account made two hard calls.

Day-to-day optimizationData hygieneOwnership & follow-through
Gross, week on week
−24%
one free-shipping Sunday explains most of it
Net, week on week
−13%
returns eased: 4.5% vs. 15.7%
Paid spend ÷ gross
29.8%
over the 25% ceiling
vs. unpromoted baseline
2.07x
late-August daily rate

Five weeks against the unpromoted baseline

Daily gross ÷ the late-August no-promotion daily rate

The best days were email days. The loyalty launch and the collection-promo email ran at 2.1x and 2.4x the average of the four days with no email. The week's planning lesson: the store hits its targets on promotion or email days, so the calendar has to be built out ahead of time.

Correcting Meta before reading it

ProblemBrowser and server purchase events weren't deduplicated, so Meta counted about 1.45x the real purchases.
CorrectionEvery Meta ROAS in the report is multiplied by 0.69.
ResultMeta ran about 0.51x this week, not the ~0.74x it reported. Break-even is 1.65x.

Source of record is Shopify. GA4 was set aside because its order capture had run at about half of Shopify's since a consent banner went live. Klaviyo-credited revenue is shown as a ceiling and never added to paid.

Campaign read (platform ROAS, Meta deduplicated)

CampaignChannelShare of wk spendPrior wk ROASThis wk ROASCall
PMax – Feed-OnlyGoogle38%1.72x1.24xHold. Under break-even this week, so the paused budget does not move here
PMax – Catch-AllGoogle17%1.00x0.23xPause It failed the Sept 30 break-even test we had set for it
Meta – RemarketingMeta21%1.62x0.98xHold
Meta – Prospecting A+ (retired Oct 2)Meta9%0.51x0.00xReplaced
Meta – Prospecting ASC (new collections)Meta9%–0.33xIn learning. Judge after it exits, not before
Meta – Promo, warm audiencesMeta7%–0.00xToo early and too thin to read

Total paid spend was held flat (+4% WoW). Pausing the catch-all campaign removes about 17% of weekly paid spend. October spend was pacing to about 63% of plan, which was the right call while ROAS sat below break-even.

Collection promotion, first three days

  • Promo collections sold at 3.11x their full-price daily rate.
  • Collection value per order rose 41% and pieces per order went from 1.57 to 2.25. Customers bought more pieces at the same price per piece.
  • The promotion added sales instead of pulling them from the rest of the store, which ran +12% per day.
  • Collection A drove nearly all of the lift and Collection B didn't move. The guidance: if B is still flat by Thursday, give it more space in the second email.

The month, stated plainly

  • September closed at 1,854 indexed against a 1,534 target (+21%), with paid spend at 16% of gross and 37% under its spend plan.
  • October needed 1.67x its run-rate per day to make plan, and it was tracking to about 65% of plan.
  • That's why the weekly asked the Client for a committed date on the next promotion.

Next steps: what we will do this week

  1. Send the second promo email Thursday with "through Saturday" framing. Read the full week on Shopify product exports (orders per tier, whether Collection B moves), not on email-credited revenue.AgencyOct 8
  2. Pause PMax – Catch-All and let Google's total come down by that amount. The feed-only campaign is under break-even too, so the budget doesn't move across.Agencythis week
  3. Bring paid back under 25% of gross. Hold total spend flat and judge the new Meta prospecting campaign after its learning period.Agencyongoing
  4. Put the next October promotion on the calendar. The store only reaches the required daily rate on promotion or email days.Client, with Agencyby Oct 14
  5. Fix Meta's duplicate purchase counting. The platform support ticket for the missing deduplication key comes first. When it lands, reported purchases drop about 30%. The report already allows for that.Client, with Agencyopen
Chapter 4 · Klaviyo review, Sep 1 – Oct 7

Retention: the lever that raises what a customer is worth

Chapter 1 ended on "turn on the tools that bring customers back." By autumn, new flows, a new sign-up popup and a three-tier loyalty program were live. This review checked whether each one was doing its job, and it was measured on clicks, orders and activations. Open rates were set aside because Apple Mail inflates them.

Lifecycle / CRMTesting disciplineHonest attribution
Email's ceiling share
≤ 30%
of tracked revenue, and overlaps paid
Labor Day sends
75%
of campaign revenue in the period
New popup
+29%
sign-ups/day, not yet proven
SMS unsubscribe rate
~6x
email's, so text sparingly

Revenue per recipient decays with every send

Indexed: Labor Day launch send = 100

Labor Day sale (discount)Free-shipping weekend (no discount)Small VIP audience

Each send earned less per recipient than the one before. The last-chance send still paid for itself. The third free-shipping send earned the least and had the highest unsubscribe rate (0.26%). New rule: two sends per promotion unless the offer is a discount. The VIP send went to a small engaged audience and earned 4.7x the launch send per recipient.

Loyalty launch: activation rate by tier

Share of recipients who activated a loyalty account

The entry tier was 88% of the launch audience and activated 0.04%. Between 9% and 16% of clickers activated in every tier, so the sign-up page works. The entry tier loses people before they click. Next send: narrow it to people engaged in the last 90 days and exclude existing members. Site-wide activations ran about 2.9x normal for three days, which roughly matches the email-attributed count, so the attribution looks sound.

What the review found, ranked

AreaFinding
Fix first
Welcome flow
The new first email opens at 21% vs. 65% and converts 2.9% vs. 11.4%. A drop that size is too big to be copy. Test two causes: the popup shows the code on screen, or the email lands in spam. Run a seed-list inbox test and count code redemptions.
High
Abandoned checkout
Reaches about 51% of abandoners and has only one step, with a 9.2% click rate. Check the filters and add a 24-hour second email.
Watch
Sign-up popup
+29% sign-ups a day, but the old form beat it in its one clean, no-sale week (7.3% vs. 6.3%). Needs 3–4 sale-free weeks to settle.
Before BFCM
Deliverability
Bounces passed 1% on whole-list sends. Build a sunset segment and exclude it before Black Friday volume.
Chapter 5 · Merchant Center feed audit, October

Feed operations: the plumbing under PMax

After PMax – Catch-All was paused, Google ran on a feed-only campaign, so the product feed became the targeting. The audit split the findings into what a Merchant Center rule can fix today and what has to change in the store, and it listed the Client decisions each fix depends on.

OperationsShopping / PMaxPrioritization
Products with dead pages
47%
of the feed, but only 12% of clicks
Clicks on unreadable sizes
85%
brand's 0–5 scale, misread by Google
Titles off-format
99%
of products
Rules proposed
8
ordered, 6 marked safe

Findings by severity

Percent of all products / all variants affected, and share of total feed clicks on live items

FindingSeverityProductsVariantsLive clicksFix
Landing page returns 404 (retired lines, most variants still "in stock")Critical47%25%12% (all)R1 stopgap, then store
Size on the brand's own 0–5 scale ("3" read as US 3; actually L)High54%60%85%R4, R5
Gender missingHigh32%23%19%R2
Promotional text in title ("FINAL SALE", "Best Seller")High29%15%14%R5
Colour differs from store dataHigh2%1%1%Store
Main image belongs to another productHigh9%5%1%Store
No GTIN, identifier_exists not setHigh2%1%0%R3
Title not in agreed formatMedium99%99%88%R5
google_product_category blank (incl. #1 product by clicks)Medium65%63%28%R6
custom_label_0 missing or wrongMedium24%16%<1%R7
Description under 150 charactersMedium20%17%16%Store
Same product listed twice; gift cards eligible for adsMedium3%3%0%Store, R1
Material blank though description names itLow82%86%79%R8
Sale price with no end date (fine for permanent markdowns)Low47%25%7%No action

Sorting by click exposure as well as severity changes the order: the size scale touches 85% of live clicks and the dead pages only 12%. Every live variant's gender, colour and size were checked against the store's data.

Attribute rules, in the order to enter them

  1. R1 excluded_destination. Exclude 404 products and gift cards from Shopping, free listings and display.Safe
  2. R2 gender. Blank → female. "Men's" titles → male. Leave unisex robes as they are.Safe
  3. R3 identifier_exists. Set to "no" where there's no GTIN or MPN.Safe
  4. R4 size. Map 0→XS, 1→S, 2→M, 3→L, 4→XL, 5→XXL.Safe
  5. R5 title. Rebuild as Product name | Colour | Size | Brand and strip promo text.Preview first
  6. R6 google_product_category. Fill blanks from keywords and never overwrite an existing value.Safe
  7. R7 custom_label_0. Fill and correct category labels.Confirm first
  8. R8 material. Pull from the description.Safe

Rules run after each store sync, so a sync doesn't overwrite them. Title example: "Core sleep set | Best Seller | 100% Organic Cotton Lounge Pants 1 / Black" → "Core sleep set | Black | S | Brand".

Decisions handed to the Client

  • Do the PMax listing groups split on custom_label_0? If not, R7 is housekeeping, not performance.
  • One-piece garments: add them to bottoms, or create a new label (which needs a matching listing group)?
  • Title descriptor: keep the clean four-part title, or add a material/category segment for keyword reach on a feed-only campaign?
  • Unisex robes: leave them, or set them to female?

Store-side fixes

  • Unpublish or republish dead products. Then the R1 stopgap can come out.
  • Correct stale colour overrides, replace shared images, merge the duplicate listing, lengthen short descriptions.

Known limits: the export had no disapproval column, and every product link carried an organic UTM tag, which could make Shopify report paid PMax clicks as organic. That was flagged for verification.

Chapter 6 · Q4 budget & projection, prepared Nov 2

Q4: a forecast that tells the Client the truth about the goal

The Client set a November goal and asked for a Q4 budget. The answer was a working model: day-by-day spend and sales for the quarter, three boost tiers to compare, and a plain statement that no tier reaches the November goal, plus what it would take to get there.

ForecastingBudget scenariosSaying no with data
Boosted-post tier
Core paid budget: Oct 341 · Nov 579 · Dec 491 (indexed). Boosts are added on top in Nov–Dec.

Day by day: projected gross sales

Indexed daily gross. Oct 1–25 are actuals from the model, and the dots are the Nov 2 snapshot of actuals (static).

Actual (model input)Projected, full priceProjected, on saleFrom boosted postsSnapshot actual

Daily paid spend

Core budgetBoosted posts

Spend ramps into each sale and peaks over Thanksgiving to Cyber Monday. Each month's daily rates are scaled by one factor to land on its budget, so the shape holds at any total.

The plan by month

ROAS is Shopify gross ÷ all paid spend, not platform-attributed. Boosted posts are priced at 2.26x on gross. That's Meta's Labor Day return after deduplication. The range covers the core plan. The Nov 2 snapshot shows October landed at 98% of its projection, at 4.62x blended on 93% of budget.

What closing November would take

  • At the planned budget, November demand would have to run 1.14x the model's base.
  • Or spend would have to rise to about 1,016 indexed, 1.75x the November budget, because added spend returns only 1.03x at the margin.
  • The goal implies November at 2.52x October. The old business never went above 1.84x in three holiday seasons.

Telling the Client this in writing in early November was more useful than projecting to the goal and missing it in December.

Which holiday season the plan assumes

The brand has no 2025 to compare against, so each prior season's shape is scaled to the current run-rate.

Boosted posts: where the tier budget goes

Share of each month's boost pool by post. Shaded rows are December slots the Client still has to fill before anything can be boosted.

PostWindowTier 1Tier 2Why
Still: gift pairing (Collections D + E)11/3–11/912%12%Gift guide launch week, full price. Stills were the old account's best format
Reel: hero robe11/5–11/118%8%Robe as the gift hero, at full price, while PJs are on sale
Carousel: core sleep set11/10–11/11–5%Full-price gap between two category sales
Still: hero wrap robe11/17–11/2314%13%Last day of robes 25%, running into a partner livestream event
Reel: Collection B + short set11/19–11/236%5%VIP 30% is email-only, so the boost promotes the gift guide, not the code
Carousel: robe + PJ set11/24–11/2714%13%BFCM 25% opens
Still: signature slip11/25–11/3026%24%The only still in the BFCM run, so it carries the most
Reel: hero robe11/26–11/3010%10%Thanksgiving to Cyber Monday
Carousel: Collection C11/27–11/3010%10%Black Friday to Cyber Monday
Needed: matching PJs still12/2–12/410%10%Matches the 12/2 homepage
Needed: gift set still12/5–12/1330%30%Free Shipping Week 20%. Ends before the shipping-deadline banner
Needed: second creative12/8–12/1420%20%So one post doesn't carry ten days
Needed: full-price robe still12/15–12/2010%15%Full price only. Do not boost Deal of the Day
Needed: gift card post12/21–12/2410%10%Past the shipping deadline, gift cards still deliver
Needed: intimates still12/26–12/3120%15%Year-end buy-more-save-more on intimates

Boosts are priced at 2.26x on gross, and break-even on the boosted days is 2.23x, so they roughly pay for themselves (about +1% of boost spend). If they return only the account's 1.03x marginal rate, Tier 2 loses about 54% of what it spends. If they match the old account's best stills (2.90x), it makes about +30%.

Boost rules: do

  • Run boosts inside the live ad sets, using Ads Manager with "use existing post."
  • Put boost money behind offers and gifting moments, weighted to BFCM and Free Shipping Week, mostly stills.
  • Judge boosts on Shopify orders in each flight window.

Boost rules: don't

  • Don't use the in-app Boost button. It creates its own engagement campaign. That repeats an August traffic campaign that got zero purchases, and it splits an account that's already short of purchase signal.
  • Don't boost Deal of the Day. At 41–50% off, break-even on gross runs above 4x.
  • Don't read boosts on Meta-reported purchases until deduplication is fixed, because Meta overcounts about 1.45x.

How the number is built, and why it might be wrong

  • Level: the brand's no-sale run-rate (Sep 10 – Oct 21) vs. the same days in 2024. Today's business is 0.47x the old one's size.
  • Shape: each 2026 day takes the 2024 day the same distance from Thanksgiving, so Black Friday and Cyber Monday line up.
  • Promos: measured lifts are 1.92x for a 20% sitewide sale, 1.57x for VIP early access and 1.05x for category sales. Days with no 2026 sale reset to 2024 full price.
  • Spend: dollars above the 2024-implied level add gross at 1.03x, and dollars below it remove gross at the same rate.
  • May run high: fitted on September alone, the 2024 shape over-predicted October's full-price days by 14%.
  • Back-to-back promos: eight sales in ten weeks. Free Shipping Week uses the Halloween sale's lift, not Labor Day's 3.86x.
  • Break-even rises with the discount: 1.92x at full price, 2.37x at 20% off, 2.66x at 25% off. Every month clears these on a blended basis.
  • Under-pacing: at 85% of planned spend, Q4 gross lands about 2.5% lower.
Daily plan table (92 days, indexed)
Across the six chapters

How the account was run

Operations & process

The same cadence every month: an executive yardstick, a monthly plan with written triggers, a weekly one-pager with owners and dates, and periodic audits of email and feed. The Shopify record comes first, and platform numbers are corrected before anyone reads them.

Budget management

Spend went from 47% to 30% of gross in one month, and the account then held the path to 16–25%. Campaigns were cut when they failed a stated test. Spend was held flat while ROAS was below break-even. Q4 was forecast at the margin (1.03x), not on averages.

Client communication

Each report leads with the answer, gives attribution as a range, and frames budget gaps as explicit decisions. Each action has an owner and a date. In November the Client heard in writing that no tier reaches the goal, along with what would.