CRAIC — AI-Powered Band Recommendation Engine
We built a conversational AI tool that guides customers through creating a personalised medical ID bracelet, from engraving suggestions to product selection.
The Challenge
ID Band Company sells medical identification bracelets — products where getting the details right genuinely matters. Customers need to choose the correct band style, size, material, and engraving content, and mistakes can have real consequences in an emergency. The existing product pages presented dozens of options with no guidance, leading to high cart abandonment and a significant volume of pre-sale support queries.
The client wanted something more than a standard product configurator. They needed a tool that could understand a customer’s medical condition, suggest appropriate engraving text based on clinical best practice, and guide them toward the right product — all without requiring the customer to already know what they wanted. Essentially, they needed the expertise of their best customer service rep, available 24/7.
The additional complexity was that this had to integrate directly into their existing BigCommerce storefront and add products to the cart programmatically, not redirect users to a separate flow.
Our Approach
We built CRAIC (Complex Recommendation AI Chat) as a custom chat interface embedded directly into the BigCommerce theme. The front end is a lightweight widget built in vanilla JavaScript to avoid framework bloat, communicating with a serverless back end running on Cloudflare Workers.
The AI layer uses a fine-tuned prompt chain rather than a single monolithic prompt. The first stage identifies the customer’s medical condition and requirements. The second stage maps those requirements against a structured product catalogue we maintain as a JSON dataset, pulling in engraving templates that follow medical ID conventions (e.g., condition first, then allergies, then emergency contact format). The third stage handles product selection, presenting filtered options with real-time pricing pulled from the BigCommerce Storefront API.
We implemented guardrails to prevent the AI from offering medical advice — the system suggests engraving content based on what the customer tells it, but explicitly defers to their healthcare provider for clinical decisions. Session state is managed client-side with fallback to a KV store, so conversations persist across page navigations without requiring user accounts.
The Results
TBC: We are still in the early testing phase with this tool. We are measuring several key data points to assess three main questions;
- Does it improve conversion rates
- Does it improve average order values
- Does it reduce support queries and requests
Need similar results?
Every project starts with understanding your specific challenges. Let's talk about what you need.