Grounded answers
Retrieval over the catalogue, ingredient guides and policies, so every product answer quotes the brand's own material rather than a model's general knowledge.
( Cookies )
We would rather ask than assume.
This site sets nothing optional unless you say so. Analytics would tell us which pages are worth improving. Say no and the site works exactly the same. The cookie policy has the detail.
Strictly necessaryAlways on
Remembers this choice and keeps the site secure and working. Nothing here identifies you, and it cannot be switched off.
Analytics
Counts page views so we can see which pages are worth improving. Off unless you turn it on, and we do not use it to follow you around the web.
Marketing
Would let us measure whether an advert brought you here. We run none today, so leaving this off changes nothing you can see.
An AI support assistant on WhatsApp and web chat that answers from the product catalogue and live order data.
Solenne · Support assistant
Live( The challenge )
Solenne sells skincare online across three countries, and almost every order produces a question. Where is my parcel, can I change the size, which serum works with retinol. Most of those questions arrived on WhatsApp, where a team of four answered them by hand.
At peak the first reply took more than five hours. Answers varied depending on who picked up the chat, and the team spent its day looking up the same order statuses in Shopify instead of handling the conversations that actually needed a person.
( Our approach )
We built an assistant that answers from two sources it can trust: the product catalogue and ingredient guides, and the live order record in Shopify. It never guesses. If a question falls outside what those sources cover, it says so and hands the chat to a person with the full context attached.
Order changes, address updates and returns run through the same checks a human agent would apply, so the assistant can complete them rather than just describe them. Every conversation is logged, scored and reviewable, which is how the team kept improving the answers after launch.
( What we built )
Delivered in 5 weeks, with the source code, accounts and documentation handed over in full.
Retrieval over the catalogue, ingredient guides and policies, so every product answer quotes the brand's own material rather than a model's general knowledge.
Order tracking, size swaps, address changes and return requests completed directly against Shopify, inside the rules the business already had.
One assistant across the WhatsApp Business API and the website chat widget, with conversation history shared between them.
Anything sensitive or uncertain goes to an agent with a summary, the order details and the conversation so far. The customer never repeats themselves.
Topic limits, refusal rules for medical claims and a confidence threshold that decides when the assistant answers and when it asks for help.
Resolution rate, handoff reasons and flagged answers in one view, so the team can see exactly where the assistant needs better material.
( The outcome )
( How it was delivered )
We read three months of support chats and grouped them by intent, which showed that eight question types made up most of the volume.
Retrieval, the Shopify integration and the order actions, tested against real historical questions before any customer saw it.
The assistant drafted replies that agents approved or corrected, which gave us a measured accuracy figure before switching it on.
Live on WhatsApp and web chat, with daily reviews of flagged answers through the first weeks.
( Services used )
( Built with )
Mainstream tools with deep talent pools, so the client is never dependent on us to maintain what we built.
( Read next / Freight and logistics )
An operations dashboard, Python data pipelines and a customer tracking portal that replaced a freight team's spreadsheets.
Send a brief and you get a written scope, a timeline and a fixed price back within 48 hours. It costs nothing and commits you to nothing.