Option A
Off-the-shelf AI tools
Tested Jasper, Copy.ai, and peers. Output was generic enough that brand-voice consistency would require so much human editing that the time savings vanished. Not a product; a faster typewriter.
All work
1,200 SKUs. 6 markets. 90 days.
A growing DTC fashion brand was launching in six new markets but couldn't write product copy fast enough to hit the dates. We built them a full custom AI content engine: admin panel, brand voice tuning, multi-language generation, SEO scoring, human review queue. They launched on time. Their content team now reviews instead of writes.
$40M ARR, board-approved expansion, no content path that worked.
The brand had hit $40M ARR in their home market and the board approved a six-market international expansion. Each market needed localized product descriptions in the local language (not just translated), region-specific size guides and care instructions, SEO-optimized titles and meta-descriptions, and brand-voice consistency across ~2,000 SKUs.
The content team (4 writers and an editorial lead) was already maxed running the home market. The COO had sketched a plan to hire 12 more writers. The CFO had killed it. The Head of Content had pitched contracting to a translation agency, but a sample test showed the output read like a translation, not like a brand voice, and the brand voice was the entire point.
"We could either launch on time and sound generic, or sound like us and miss the launch dates.
Both options were going to cost us a year of growth."
Mark N. Head of Content
By the time we got involved, they were 14 weeks from the first market launch with no content path that worked.
We pitched a product, not a workflow.
Option BChosen
Custom internal product
A focused product their content team would use as their daily workspace. Higher upfront cost, but full control over brand-voice tuning, review workflow, and PIM integration. Built to feel like Notion or Linear, not a Python script with a UI.
Option C
Human writers + AI drafts
Lower upfront investment, but once we modeled the timeline the human review was still the bottleneck. We couldn't hit the launch date. The slower path was also the riskier one.
We took Option B, and we were explicit with the client: this was a real product build, not a prompt-engineering project with a thin interface. The people using it every day would be editors, not engineers. It needed to feel like Notion or Linear. That framing changed what we scoped, what we tested, and how we measured success.
A standalone product deployed under their own subdomain.
Six core components. One workflow the editorial team owns end to end.
Brand-voice profile builder
Editorial lead trained the engine by tagging "on-brand" and "off-brand" examples from the existing catalog. The system encoded those patterns into a per-language style profile that every generated draft must match.
SKU-to-copy generator
Given a SKU's structured data, the engine generated drafts in the requested language: product title, short description, long description, SEO meta-description, size guide, care instructions.
Multi-language workflow
Six languages launched: EN, FR, DE, ES, IT, SE. Each market's content goes through a per-language brand-voice profile, then a localization-review step before approval.
SEO scoring + PIM sync
Every draft gets a per-keyword SEO score against the brand's target term list. Approved copy syncs to their PIM, then Shopify, then the storefront. Zero manual copy-paste, end to end.
Editors review. They don't write from scratch anymore.
The review queue puts humans exactly where they add value: judgment calls, not keystrokes.
Twelve months post-launch, the numbers held.
340
SKUs per month per market (was ~80)
9
Minutes per SKU (was 47)
6/6
Markets launched on time
0
Net-new hires (12 planned)
Four metrics. Every one moved.
| Metric | Before | After |
|---|---|---|
| SKUs launchable per month per market | ~80 | ~340 |
| Avg time per SKU (write + review) | 47 minutes | 9 minutes |
| Content team headcount (planned vs actual) | +12 hires planned | 0 net new hires |
| Markets launched on time | n/a | 6 of 6 |
Twelve months post-launch: 1,200 SKUs launched in the first 90 days; $1.2M incremental revenue from on-time market launches; 37% YoY organic SEO traffic growth.
"Most agencies would have built us a workflow. They built us a product. The difference is that the editorial team actually wants to use it."
Mark N. Head of Content, DTC fashion brand
A real product on a real stack, not a black box.
- Frontend
- Next.js, React, Tailwind, shadcn/ui
- Backend
- Node.js, TypeScript end-to-end, tRPC
- Models
- Anthropic Claude (primary copy generation), GPT-4 (SEO scoring)
- Database
- Postgres, Pinecone
- Auth
- Clerk
- Storefront
- Shopify Plus
- Infra
- Vercel, Render, client AWS for data storage
14 wks
Timeline · 3-wk discovery + 11-wk build
5
Team · engineers, AI specialist, designer, delivery
$85K
Budget cap · fixed scope
$79.5K
Actual spend · 6% under cap
12 hrs
Training program + 90-day support included
If you need to launch faster than your content team can write
We built this for that.
We built this for exactly that. Tell us your timeline, your markets, and your brand voice. We'll scope a product that ships on time.
No deck · No demo · No sales pressure