Chatbot for Automated Customer Support
Sales and support chatbot covering FAQs, handoff and booking flows.
Key Details
| Challenge | Agents repeated FAQs while booking intent was lost in chat transcripts. |
|---|---|
| Solution | A production chatbot with retrieval, tools for booking and clean handoff to humans. |
| Technologies | Custom LLM stack, Node.js, CRM APIs |
Technologies used
OpenAI
Node.js
PostgreSQL
Redis
Docker
Client background
A travel and hospitality brand flooded agents with the same FAQs while high-intent booking chats died in transcripts. Guests bounced between bot and human without context, and sales never saw which conversations were ready to convert.
Key challenges
- FAQ volume drowned agents who should have focused on complex bookings.
- Booking intent was buried in free-text chat with no structured handoff.
- Legacy bot answers conflicted with current offers and policies.
- No analytics on deflection vs. revenue-generating conversations.
What we built
- Production chatbot with retrieval over FAQs, offers and booking policies.
- Tooling to start booking flows and pass structured state into the CRM.
- Human handoff that preserves full transcript and guest intent.
- Ops dashboard for deflection, containment and booking conversion.
Project team: 6 engineers across AI/ML, backend and domain specialists — delivery over 12 weeks.
How we delivered
01
Listen
Sampled tickets and chats to rank intents by volume and revenue.
02
Design
Separated FAQ containment from booking assistance and escalation.
03
Integrate
Connected CRM and booking APIs with safe tool boundaries.
04
Tune
Iterated prompts and retrieval using live deflection metrics.
Business impact
DeflectedRoutine tickets
FasterBooking start
FullHandoff context