AI Customer Enquiry Assistant
Customer enquiries arrive in different shapes and priorities — most businesses sort them by hand, and the good ones go cold. This project demonstrates an assistant that classifies, prioritises, and drafts follow-up — with a human approving anything before it is sent.
Enquiries don't sort themselves
A typical business receives messages that mix sales questions, support requests, billing issues, partnerships, and spam — through web forms, email, WhatsApp, and social. Sorting this by hand is slow, inconsistent, and depends on whoever happens to check first. By the time a high-value enquiry gets attention, it has often gone cold.
Classify first, then route and draft
The assistant reads each incoming enquiry and produces a structured triage:
- Category — sales, support, billing, spam, or other
- Priority — based on intent signals and urgency markers
- Suggested routing — who on the team should handle it
- Draft reply — a starting response a human reviews and sends
Nothing customer-facing is sent automatically. The human-in-the-loop step is the product, not an afterthought — automation that loses trust is worse than no automation.
Try it — classify an enquiry
Why it is built this way
Human approval is default
Enquiries are how revenue arrives. A wrong automated reply costs more than the minutes saved. The system drafts; a person approves. Autonomy can be added per-category later once trust is earned.
Transparent classification
Every classification shows its signals — which phrases triggered which category — so your team can audit and correct the system instead of blindly trusting it.
Grounded in your business
Production versions answer from your approved documentation: services, pricing rules, booking policies. The assistant does not invent answers — it quotes what you have approved.
What this demo is and isn't
It is: a working, functional demonstration of the triage workflow — classification, prioritisation, routing, and draft generation — running client-side so you can verify it yourself.
It isn't: a production LLM deployment. The browser demo uses keyword/pattern rules, not a language model, so it handles common phrasings well but will not match an LLM's nuance. A production build connects to an LLM API with your business context, adds delivery integrations (email, WhatsApp, CRM), and includes monitoring.
What you can learn from this
- Enquiry triage is a strong first automation target — high volume, clear categories, measurable improvement.
- Human-in-the-loop designs get accepted by teams faster than fully autonomous ones.
- Showing the "why" behind each classification builds more trust than a black box.
Want this adapted to your enquiry flow? Discuss your project →