01 the business
Outdoor gear retailer
30 staff, 6,000 products
Each season brings hundreds of new products with a supplier spreadsheet, a PDF spec sheet and a few photos. Someone types the details into the shop and two marketplaces. Products go live late, and missing attributes mean shoppers cannot filter to find them.
02 before
The problem
New stock arrives with a supplier spreadsheet, a PDF spec sheet and a few photos. Someone copies details by hand into Shopify and every marketplace, so products go live late or with thin, inconsistent descriptions. Missing attributes mean shoppers cannot filter to find them.
- New stock waits days or weeks to be listed.
- Descriptions vary by who wrote them.
- Filter attributes such as size or material are often blank.
03 the build
What the AI does
- Step 1: Reads new or changed SKUs from Shopify, your inventory system or a supplier spreadsheet.
- Step 2: Extracts sizes, materials, dimensions and care details from the supplier PDF and photos.
- Step 3: Drafts a title, description and bullet points in your house style, sized for each channel.
- Step 4: Fills missing attributes and filter tags, and flags any field it could not confirm from the source.
- Step 5: Pushes approved listings to Shopify, Google Merchant Center and Amazon Seller Central.
paused: waiting for a person to approve
How it was put together
- Read new and changed products from the shop and supplier spreadsheets.
- Extracted sizes, materials, dimensions and care details from the PDFs and photos.
- Drafted titles, descriptions and bullet points in house style for each channel's limits.
- Pushed approved listings to the shop, Google Merchant Center and Amazon.
04 human in the loop
Where a person stays in charge
A merchandiser approves each new listing in a review queue before it is published. Any claim about safety, materials or certifications is highlighted for a second check.
- A merchandiser approves every new listing in a review queue.
- Safety, material and certification claims are highlighted for a second check.
- Fields the system could not confirm from the source are flagged, never guessed.
05 rollout
From kickoff to live in 4 weeks
- Week 1Agree house style, required attributes and channel rules.
- Weeks 2–3Draft one category and compare with the team's own listings.
- Week 4Use for all new products; backfill missing attributes on old ones.
06 after
What changes
New products go live sooner with complete, consistent details on every channel. The team edits drafts instead of typing from spec sheets.
Assume 400 new products a season and 20 minutes to list each by hand, against 5 minutes to review a draft.
400 × 20 minutes = about 133 hours of listing. 400 × 5 minutes = about 33 hours of review. Roughly 100 hours saved a season.
What to measure
- Days from stock arrival to live listing.
- Share of products with complete filter attributes.
- Edits per draft listing.
- Returns that cite 'not as described'.
07 evidence
What the research says
US retailers expected 16.9% of 2024 sales to be returned, about $890 billion, and 93% said return fraud and abuse is a significant issue.
In a study of 453 professionals doing writing tasks from their own jobs, access to ChatGPT cut the time taken by 40% and raised quality, as rated by evaluators, by 18%.
08 lessons
Mistakes to avoid
- Inventing details the spec sheet does not support.
- Ignoring channel rules. Amazon and Google have strict title and attribute requirements.
- Backfilling everything at once. Start with your best-selling categories.
09 questions
Common questions
- Can it handle supplier data in different formats?
- Yes. Spreadsheets, PDFs and photos are all read, and anything unclear is flagged.
- Does it work with WooCommerce or a PIM?
- Yes. It can write to WooCommerce, Akeneo and most systems with an API.