LLM with guardrails
Natural conversation in Orbit's tone of voice, fenced by guardrails so it never invents prices or promises what isn't available.
Case study · AI & Automation
An AI assistant that books travel on WhatsApp
Orbit Travel was drowning in repetitive enquiries. We shipped a conversational AI agent — wired straight into their live reservation system — that now answers, quotes and holds bookings around the clock, in seconds.
ClientOrbit Travel
IndustryTravel & Tourism
ServicesAI · Integration
Timeline4 weeks
Hi! Any Maldives packages for 2 in October? 🏝️
Yes — I have 3 options from AED 4,200pp: 5 nights, half-board, return flights from DXB. Want me to hold one?
The 4,200 one please 🙌
Done ✓ Held for 24h under your name. Shall I send a secure payment link?
Booking heldLive reservation sync
92% automatedNo agent needed
0%
Enquiries resolved end-to-end
no human in the loop
0s
Median first response
down from ~3 hours
0h
Agent hours saved each week
redeployed to closing
0%
Customer satisfaction (CSAT)
post-chat survey
01 — Challenge
Enquiries never slept. Neither did the team.
Orbit Travel runs on WhatsApp — it's where their customers actually are. But the same questions arrived at every hour: availability, pricing, baggage rules, date changes. Agents spent the bulk of their day typing the same answers instead of closing high-value bookings, and travellers in other time zones waited hours for a reply.
Canned auto-replies weren't enough — customers expect real answers tied to live inventory. Orbit needed instant, 24/7 responses that spoke to their actual reservation system, without hiring a night shift to do it.
- Peak enquiry volume hit at night and on weekends, when no agents were online.
- Roughly 7 in 10 messages were repetitive FAQs that still needed a human to type.
- Slow first replies meant lost bookings to faster-responding competitors.
02 — Approach
A four-week path from transcript to live agent
We started from real conversations — six months of Orbit's WhatsApp history — so the assistant learned how their customers actually ask, and how their best agents actually reply.
01
Discovery & data
Mined chat history and the reservation API to map every intent, edge case and handoff trigger.
02
Conversation design
Defined a branded tone, guardrails and fallback flows — reviewed with Orbit before a line of code.
03
Build & integrate
Wired the LLM to live inventory over secure webhooks, with a Node.js orchestration layer.
04
Launch & tune
Shipped behind a pilot, watched real chats and tightened prompts until automation held steady.
03 — Solution
What we built
A conversational AI agent on the WhatsApp Business API that understands natural-language enquiries, checks live availability and walks customers through booking — handing off to a human only when it genuinely should.
Live reservation sync
Real-time availability, quoting and 24-hour holds over secure webhooks into the existing booking system — no copy-paste.
Seamless human handoff
Complex or high-value chats escalate to an agent instantly, with the full conversation context already in hand.
Ops analytics dashboard
A live view of volumes, automation rate, peak hours and escalation reasons — so Orbit tunes coverage with real numbers.
Pay-by-link checkout
The assistant sends a secure payment link the moment a customer confirms, turning a chat into a paid booking in-thread.
Safe by design
Rate limiting, PII redaction in logs and a strict scope keep customer data private and the agent on-task.
In context
One dashboard, the whole conversation funnel
Live ops dashboard — automation rate, volume and escalation reasons at a glance
Under the hood
The stack that powers it
Production-grade, observable and built to scale with Orbit's booking volume.
WhatsApp Business APILLM / MindStudioNode.jsWebhooksNext.jsSupabaseVercelReservation API
04 — Results
Faster replies. More bookings. Zero new headcount.
Within weeks of launch, the assistant was handling the overwhelming majority of inbound enquiries on its own — instantly, at any hour. Agents shifted from typing the same answers to closing the bookings that actually move revenue.
- 92% of enquiries resolved without an agent, day or night.
- 1.8× more enquiries converted into confirmed bookings.
- Zero extra agents hired to cover 24/7 — coverage came from automation.
- Happier customers, reflected in a 96% post-chat satisfaction score.
0×
more enquiries converted to confirmed bookings
24/7
instant coverage, including nights and weekends
0h
agent hours freed every week for high-value work
PrimeX built our WhatsApp booking assistant in weeks. It now handles the majority of our inbound queries automatically — our team finally focuses on closing, not replying.
AR
Ahmed R.
Operations Director · Orbit Travel
Your project next
Want results like these?
Tell us about your business and we'll map the fastest path to impact — from a single AI workflow to a full product build.