Online retail · Hong Kong

AI product description writing for a Hong Kong homeware store

800 products, fully described in one weekend.

-95%listing-copy time
+22%product-page conversion
HK$29,000saved / month

Hundreds of products sat with blank or one-line descriptions because nobody had time to write them. We built an AI content pipeline that writes a full, on-brand description for every SKU — and keeps writing them for every new arrival.

The challenge

Writing one good product page took 20–30 minutes. With an 800-SKU backlog, around 150 new arrivals a month, and English and Chinese copy plus social captions for each, 1.5 writers worked on it full-time — and the backlog still grew. Thin pages hurt both conversion and search ranking.

The solution

A pipeline that reads the product sheet, drafts title, description and social caption in the shop's voice with Claude, routes everything through a one-click review step, and publishes approved copy straight to the store.

The workflow we built

  1. Learn the voice. We tuned the prompts on the shop's best existing pages so every draft sounds like them.
  2. Draft in bulk. Title, description and caption for every SKU, generated from the product sheet.
  3. Human review. The owner approves or tweaks each draft in a simple review screen — nothing publishes itself unseen.
  4. Publish & repeat. Approved copy goes live automatically; new arrivals enter the same pipeline every week.

A week before, a week after

TopicBeforeAfter
WritingEvery product page was written by hand, in English and in Chinese — 20–30 minutes a page.Social captions for each product were written separately again.Title, description and caption are drafted from the product sheet, in the shop's own voice.The owner approves or tweaks each draft on one review screen.
New arrivalsAround 150 new arrivals a month took up the writers' time, so the 800-SKU backlog never shrank.New arrivals go through the same pipeline every week, so a backlog no longer builds up.
Product pagesHundreds of products sat live with a blank or one-line description.Thin pages sold less and ranked lower in search.Nothing goes live without being seen.Approved copy goes to the store automatically.

How the saving adds up

  • Each product needed an English page and a Chinese page at 20–30 minutes each, plus social captions and loading into the store: about 1¾ hours per product.
  • About 150 new arrivals a month × 1¾ hours ≈ 263 hours a month — all of the writers' time, before the backlog was touched.
  • 263 hours × HK$110 an hour ≈ HK$29,000 a month.
  • 263 ÷ 173 hours (one full-time month) ≈ 1.5 full-time staff.
  • HK$110 an hour is about HK$19,000 a month for one full-time admin staff, MPF included.

These are the working assumptions behind the figure, not an audited result.

How the project ran

  1. Free audit. We looked at how pages were written, which pages sold best, and how new arrivals reached the store.
  2. Built on their own files. Prompts were tuned on the shop's best existing pages and tested on a sample of real products from their own product sheet.
  3. Side by side. For a period the writers kept writing as normal while drafts were produced for the same products, and the owner compared the two before trusting the drafts.
  4. Live, then monitored. The backlog went through the pipeline first, then each week's new arrivals. We monitor it and fix anything that comes up after go-live.

Questions owners usually ask

  • Did anyone lose their job?
    The hours went back into running the shop instead of typing copy. What to do with freed-up time is always the owner's decision, not ours.
  • What happens when it gets something wrong?
    Nothing publishes unseen. The owner approves or tweaks every draft on the review screen, so a weak or wrong draft is caught before it reaches the store.
  • Where does our data live?
    The product sheet stays where the shop already keeps it, and approved copy goes into their own store. Who has access to what is agreed with you during the free audit, before anything is built.

Tools we used

ClaudeShopifyGoogle Sheets

The results

  • The full 800-SKU backlog was described in one weekend instead of months of writing.
  • Product-page conversion rose about 22% once every page had real, persuasive copy.
  • Roughly HK$29,000 a month of writing time saved — the work of about 1.5 full-time staff.

Representative example — figures illustrative, client anonymised.

Have an operation that looks like this?

Tell us the process and we'll show you exactly how to automate it — free, and no obligation.

Get a Free Automation Audit →