An AI that helps sell more and build loyalty
An independent platform that, with the help of AI, enriches the catalogue and the customers of an e-commerce without having touched its systems architecture.
0
changes needed in their architecture
2
use cases in production
much+
we are turning it into a product
Context
An online shop for fresh, seasonal food. Their store runs on a standard e-commerce (WooCommerce) which is itself an extension of the WordPress that holds the website and the company's search work. A common setup, one that works, and one nobody wants to touch because everything depends on it.
What it looked like
“We want to bring AI into our day-to-day to give more value to customers.” The most frequent request of 2026 and the hardest to turn into something useful.
What it actually was
The real problem of a shop selling perishables: the catalogue changes every day (stock, price, season) and the customer who wants to repeat an order finds that half of what they bought is gone. The store only knew about its products and its customers what the e-commerce itself stored, which is little. And nothing could be done inside the e-commerce without putting the website and its search rankings at risk.
What we did
A separate platform. It replicates the product and customer databases, strips out everything the AI doesn't need, and works on that copy. The client's e-commerce changed nothing.
An enriched catalogue. For every product, alternatives when there is a stock-out, a price rise or the end of a season, plus combinations that make sense together for various reasons and are more appealing to the buyer.
Customers with context. With techniques like RAG and MCP we widen the context of each customer's activity to know which products sit closest to their behaviour and their wishes.
Two use cases to prove the value. Repeat purchase, which used to fail because of season or stock and now proposes substitutes; and a virtual nutritionist that offers food alternatives according to the user's needs and turns them into baskets bought in the client's store.
Where we are now
Widening what the vector database knows with tables from other standard client systems, so the embeddings work with more objective data and better context. The platform doesn't depend on the type of goods: it works for any catalogue and any customer base. We are considering turning it into a product.
What we took away
Useful AI in a shop is not a chat: it is knowing what to offer when what the customer wanted isn't there. And the fastest way to get AI into production in a small company is not to touch what already works.
