A UK supplier of specialist plumbing fittings wanted trade customers to find the right part by describing the job, not the part number. In 14 days we built conversational search and an AI shopping assistant that answers from the site's own knowledge and can add products to the basket.
This UK business supplies specialist plumbing fittings to the trades. The range is large and technical, and most products come in several variants. Customers often arrive knowing the problem rather than the exact part: a leaking shower, a joint that won't seal, two pipes that need adapting.
The client wanted those customers to find the right product on the site without having to ring and ask. We built two things in 14 days: a conversational search bar and an AI shopping assistant.
A customer can type "I have a problem with my shower, it's leaking, and I think it's from the pipes" into the search bar, and the results show the fittings that suit that problem. Search doesn't write an answer. It just puts the right products in front of the customer.
Behind it sits a scoring model rather than a large language model (LLM). It classifies each query and ranks products against it, which keeps it fast and cheap to run. In live testing it classified queries correctly 98 to 100% of the time. A standard fast LLM tested on the same task managed 56%. An average search costs £0.006.
For customers who'd rather talk the problem through, there's a chat assistant. It asks about the job, suggests products that fit and can add them to the basket. The customer confirms before anything goes in.
The assistant works from two search indexes, one for the product catalogue and one for the site's knowledge base, and searches both by meaning rather than exact keywords. It only answers from what's in them. When a question needs more help than the knowledge base can give, it opens a short contact form inside the chat and raises a support ticket instead of guessing. A full conversation costs at most around £0.70.
We built the search engine itself for the client, set up to understand how products and their variants relate, so both the search bar and the assistant suggest parts that actually fit together.
The build drew on our AI chatbots and customer support work and our ecommerce practice.
An assistant like this is only as good as the product information and knowledge base behind it. This client already had its product data in good order and ready to index, which is a big reason the build took 14 days.
That isn't always the case. When it isn't, we help build the knowledge base first, because no amount of clever search fixes missing or messy product data.
Every question we get asked on first calls. Answered in writing — decide before you book.
We take five new clients per quarter. Apply for a free audit and we'll tell you what we'd do first.