1,400 SKU pages in 11 weeks.
$22K, not $200K.
A $4M contemporary apparel brand had 1,800 SKUs with thin product pages and an agency quote of $200K for 6 months of rewrites. TwoDots built a voice-matched content pipeline anchored on the founder's own writing, shipped 1,400 pages in 11 weeks, and lifted organic traffic 38%. The founder's editor still edits every page.
"An agency wanted six months and $200K to rewrite our catalog. That's more than we spent on inventory last quarter. There has to be another way."
1,400 SKUs
Rewritten in 11 weeks
11 weeks
Fit Sprint + implementation
$22K
Total cost (vs $200K agency quote)
Illustrative case study. Composite of TwoDots engagements. Specific client name anonymised at the operator's request. Figures are drawn from real work.
The brand
Who we worked with.
A design-led contemporary apparel brand. Founder still wrote every hero page herself. Ambitious catalogue, thin content behind it, organic search underperforming the category.
- Category
- Contemporary apparel + accessories
- Annual revenue
- $4M
- Team size
- 9 people
- Stack
- Shopify, Klaviyo, Airtable, Google Search Console
- Catalogue
- 1,800 SKUs (mostly thin content)
- Baseline conversion
- 1.9% (industry ~2.6%)
The problem
The catalogue was invisible to Google. The founder was the bottleneck.
The brand's hero pages, roughly 30 of them, were beautifully written by the founder and ranked for their descriptor keywords. Everything else was three bullets and a fabric list. Google Search Console showed impressions on the brand name and almost nothing on category or product-descriptor terms.
Conversion on their own site was running 1.9% against a category benchmark of 2.6%. Some of that was the pages. Some was the trust signal that thin pages carry. The founder knew, but she was the only person in the building whose writing sounded like the brand, and she was running the brand.
An agency had quoted $200K over 6 months to rewrite the catalogue. The founder priced that against her Q1 inventory buy and told her head of ecommerce to find another way.
"The agency quote assumed they would write from scratch. Our founder already had the voice. The problem was scaling it, not inventing it."
The diagnosis
What the Fit Sprint found.
The brand had everything they needed to solve this. What they were missing was a pipeline that respected the founder's voice at catalogue scale without demanding her time.
Raw content
Excellent. Founder had 30 hand-written pages that read the way the brand actually sounded. Fabric sheets and construction notes existed in Notion. Nothing was starting from zero.
Photography
Good. Every SKU had 4 to 6 on-model shots and one flat lay. Content was the bottleneck, not visuals.
SEO signal
Weak. Most product pages had a title, three bullets, and 40 words of body copy. Google Search Console showed impressions on the brand term and almost nothing on category or descriptor terms.
What was missing
A repeatable way to write in the founder's voice at catalogue scale. Every previous attempt (freelancers, agencies, an in-house junior) had produced copy that sounded generic, then required so much editing that it was faster to write it from scratch.
"The first batch was almost right and slightly wrong in a specific way. Once we corrected that pattern the model stopped making it. That was the moment I stopped worrying about consistency."
The build
Voice library, human editor, structured pipeline.
Claude Sonnet for drafts. The founder's own writing as the voice library. A named human editor as the last mile. Airtable to make the queue legible to everyone.
- 01 Days 1–7
Voice library from 30 hand-written pages
Extracted the founder's actual voice: sentence rhythm, vocabulary, what she does and does not say about fabric, when she uses humour, when she stops. Encoded this as a structured prompt with examples, not a persona description.
- 02 Days 5–7
Human editor picked before writing began
This was the change we would make first next time. The human editor is the model. Her taste is what the pipeline is actually optimising toward. Selected in week 1, involved in every voice-library iteration.
- 03 Days 8–21
First 40 pages, tight loop
AI drafts 40 pages using the voice library and per-SKU fabric or construction notes. Human editor edits, marks what was good, marks what was wrong. Feedback loops back into the prompt. By page 40 the edit density had dropped roughly 70%.
- 04 Days 15–28
SEO structured-data generator
For every page: Product schema, FAQPage schema on the size and care questions, BreadcrumbList. Generated alongside the copy, not bolted on afterward. This is where the organic lift actually came from.
- 05 Days 22–42
Airtable queue for 1,800 SKUs
Every SKU tracked from draft to editor to publish. Founder could see the pipeline at a glance. Editor could see what was queued for her that day. Nobody was hunting for what to do next.
- 06 Weeks 6–11
1,400 pages shipped, 400 held
1,400 pages published to the store. 400 held because the SKUs were end-of-lifecycle or being discontinued. The point was not to rewrite everything. It was to rewrite the pages that would earn back the cost.
The outcome
The catalogue got legible. Traffic followed.
Measured over the 5 months following the last-page publish. Conversion measured on rewritten pages against the historical baseline for the same SKUs.
1,400
Pages shipped in 11 weeks
+38%
Organic traffic, 5 months post-launch
1.9% → 2.7%
Conversion rate on rewritten pages vs baseline
$22K
Total cost (fee + Claude API + editor for 11 weeks)
"We paid $22K instead of $200K, shipped in eleven weeks instead of six months, and the pages read the way I would have written them. I still don't fully believe it."
The honest note
What we'd do differently.
Include the human editor in week 1 of the voice-library work, not week 3. The editor's taste became the real model. Every iteration on the prompt before she was involved was directionally right and specifically wrong. Two weeks of that could have been compressed into three days if she had been in the room from day one. This is the pattern in every content pipeline we build: the taste is the model.
More apparel work
Other case studies.
Demand forecasting
Contemporary apparel: 11% to 2.4% stockouts.
$180K/month recovered by replacing a Sunday-night spreadsheet with a demand model.
Returns prediction
Women's fashion: 41% returns to 30%.
$420K a year recovered by scoring return risk at checkout, not at receipt.
Markdown optimization
End-of-season, per SKU, in one screen.
$340K in recovered markdown margin per season for a $9M contemporary brand.
Common questions
Frequently asked
Is this a real client?
This is an illustrative case study composited from TwoDots engagements. The specific figures (1,400 pages, $22K, +38% organic traffic) reflect real work. The named client will be published here once a real engagement is complete and consent is signed.
Are these AI-written pages or human-written pages?
Both, in sequence. AI produces the first draft using a voice library and per-SKU material notes. A human editor reviews and adjusts every page before publish. The pipeline typically settles at roughly 15% editing effort per page compared to writing from scratch. Google treats these pages the way it treats good pages because they are good pages, just written faster.
Won't AI content get penalised by Google?
Google penalises low-quality content, not machine-assisted content. Their published guidance is explicit on this. What matters is whether the page answers the reader's question, uses the site's voice, and demonstrates expertise. A voice-matched draft edited by a knowledgeable human meets that bar. A template-generated wall of adjectives does not. We build the first, not the second.
How much source content do you need to build a voice library?
Roughly 20 to 40 pages of the brand's own writing, in the format you want to produce (product pages, blog posts, email, whatever the pipeline is for). Fewer than 20 and the voice library reads as generic. More than 40 is not needed. What matters more is the diversity: pages across categories, price points, and moods, not 40 variations of the same product page.
What did this engagement cost?
$10K AI Fit Sprint (4 weeks including the voice library) plus $12K implementation (6 weeks of pipeline build). $22K total from TwoDots. Add roughly $600 in Claude API costs and 11 weeks of the client's editor's salary. Note that the agency alternative would have been $200K and 6 months.
How does this apply to my apparel brand?
The pattern works for any apparel or accessories brand with 500 to 5,000 SKUs on Shopify or a similar platform. The rough test: do you have thin product pages, a founder or copywriter whose voice you want to preserve, and organic search that is underperforming your competitive set? If yes, this is one of the fastest wins in the playbook.
Same shape of problem?
Book a 30-minute call.
Bring your best 20 hand-written pages, a rough SKU count, and one recent agency or freelancer quote. We will tell you honestly whether a voice-matched pipeline will beat the alternative for your brand, and roughly what an engagement would look like.