Doubled chat-to-cart conversion with a product assistant that never invents a detail, for a home-goods store
Client snapshot
A direct-to-consumer home-goods brand launching a range of 30-plus products on Shopify: cookware, storage and small appliances, several of them close variants of each other. Five people, one of them on customer support.
The problem
Customers asked the same questions before buying. Which pan works on induction, what's the difference between the two blenders, what can I get for under $100. The support inbox filled up during launches, and replies came hours later, after the customer had left the site.
The owner had tried an off-the-shelf chatbot. It answered quickly and sometimes wrongly: it gave one product's dimensions for another and promised a warranty the brand doesn't offer. They switched it off after a week. The requirement for anything new was simple. It has to be right, or it has to say it doesn't know.
What we built
A shopping assistant that only answers from the store's own data.
n8n syncs the Shopify catalog every hour: titles, descriptions, specs, variants, prices, stock and the FAQ and policy pages. Each product is stored with its fields kept separate, so the assistant looks up a dimension instead of guessing it from a paragraph.
When a customer asks something, Claude finds the relevant products and writes the answer using only those records. Recommendations respect a stated budget and only include items in stock. Comparisons pull the same fields for each product side by side. Every product it mentions appears as a clickable card with the image, price and an add-to-cart button. If the data doesn't cover the question, the assistant says so and offers to pass it to support, with the conversation attached in Gorgias.
Before launch we tested it against 20 real customer questions from the support inbox and checked every answer against the catalog. The code lives in the brand's own GitHub repository, with deployment notes and a walkthrough video.
Results, 60 days in
- Chat sessions that added something to the cart: about 9% → about 19%.
- Around 70% of pre-sale questions answered without a person.
- Zero answers with a wrong product detail in launch testing, and none reported since.
- Support now handles the questions that need judgement: damaged deliveries, custom orders, returns.
Quote
"The old bot was fast and made things up. This one tells people when it doesn't know, which is why I trust it with the rest." Founder
Build snapshot
- Timeline: 21 business days, including a written design doc signed off before the build
- Stack: Shopify, n8n, Claude, Postgres + pgvector, Gorgias
- Investment: $7,500
- Who owned it: the founder approved the test answers; the support lead reviews the handoff log weekly.
The pattern
This is the kind of problem we take on in a Custom System Build: a specific requirement, here "never invent a detail", that off-the-shelf tools don't meet. If your catalog questions are eating your support time, the approach is the same, with your store swapped in.
This is an illustrative composite drawn from operator engagements. Names, identifying details, and exact metrics are anonymised. The workflow, stack, and approach are real.
Looks like your situation?
A 30-minute call tells us whether the same pattern fits your operation, and what it would cost to ship it.