• travel
  • Travel & hospitality

A travel concierge that quotes in the chat, then hands over to a person

A travel agency

A concierge that lives in Telegram and WhatsApp, so a traveller never leaves the app they were already messaging from. Ask it for a week in Bali and it comes back with an actual itinerary — day by day, priced from live supplier data across 50+ countries, as cards rather than a wall of text. The interesting decision is where it stops. Once someone is ready to pay, the bot writes a quote, works out which agent is online and knows that country, and hands the whole conversation over. Start in Telegram and finish in WhatsApp and the itinerary, the history and the payment progress all follow you across.

Challenge

The people ready to book were queued behind the window shoppers

A major travel agency was losing 40% of potential leads due to delayed responses on WhatsApp and Telegram. During peak holiday seasons, human agents were overwhelmed with repetitive 'price check' queries — spending 70% of their time on non-qualifying window shoppers while high-intent travellers who were ready to book waited 4-6 hours for a quote and went to competitors.

Someone messaging a travel agency at eleven at night has usually messaged two others as well, so a reply the next morning arrives into a decision already made. But a booking is flights, hotels, visas, insurance and transfers, and the money moves at the end of it — which is work that still wants a person on it. So the question was never whether to automate the agent away. It was where to put the line, and how to get a traveller across it while they are still interested.

Solution

A real quote in the chat, then a person once they are ready to pay

We built a 'Discovery-to-Handover' pipeline where the AI concierge serves as the first point of contact. Using Retrieval-Augmented Generation (RAG), it pulls real-time travel data, pricing, and availability from supplier APIs across 50+ countries. When a user signals booking intent, the bot generates a PDF quote and instantly notifies the best-matched available agent via a custom dashboard.

To lift conversion, we implemented 'Visual Itineraries' — the bot doesn't send walls of text. It generates interactive cards with destination images, day-by-day breakdowns, and transparent pricing. We also built 'Availability-Based Routing', which checks who is actually online and knows the destination country before it routes anything — that is what keeps the handoff quick rather than merely intended, and when nobody is free the bot says so instead of queueing the traveller silently.

Calls we would still defend

  • Real-Time Pricing Engine — Built a microservice aggregating flight, hotel, and transfer costs from 15+ supplier APIs with tax calculation for 50+ countries — providing accurate quotes in under 3 seconds.

  • Cross-Platform Session Sync — Unified backend ensures a traveller who starts in Telegram can carry on in WhatsApp with the conversation history, itinerary state, and payment progress intact.

  • Dynamic Lead Scoring — AI reads urgency signals (travel dates within 2 weeks), budget indicators, and engagement patterns to prioritize 'Hot Leads' for immediate agent intervention — preventing high-value bookings from going cold.

Results

Agents stopped answering price checks

Before

Customers waiting 4-6 hours for a quote. Agents spending 70% of time on non-qualifying inquiries. 5% conversion rate. Zero automated lead qualification. Peak season bottlenecks causing 40% lead loss.

After

Itineraries with live pricing, in the chat. Agents pick up conversations that have already got past price-checking, with the dates and the budget attached. 12% conversion rate. 200+ daily bookings. 3x agent productivity without additional hiring.

30%

Conversion Rate

More enquiries turned into bookings by the concierge

2s

Avg Response Time

Traveller questions answered in about two seconds

3x

Agent Productivity

Each agent handles roughly three times the earlier volume

FAQ

The questions this one raises

What was actually built, the constraints it had to meet, and what we would do differently.

  • Yes. The AI understands multi-leg journeys and generates routes considering flight connections, visa requirements, local transfer times, and seasonal pricing. A user can request 'Tokyo → Kyoto → Seoul → Bali' and receive a complete itinerary with per-segment pricing within seconds.

  • If all agents are occupied, the bot collects the user's contact details, captures their itinerary preferences and budget, offers the callback slots that are actually free and books the one they pick. The lead and full conversation context are synced to the CRM so the agent can pick up exactly where the AI left off.

  • Yes. The bot auto-detects the user's region and language preference, then provides quotes in local currency using real-time exchange rate APIs. It supports 15+ languages for conversation and 135+ currencies for payment processing.

  • Quotes are generated by aggregating live pricing from 15+ supplier APIs (airlines, hotel chains, transfer services) with real-time tax and surcharge calculation. Prices are cached for 15-minute windows to balance accuracy with response speed, and users are notified of any price changes before payment.

  • Through REST APIs and webhooks, which is what the booking platforms expose. We build the connector a deployment needs rather than shipping a library of them — and a GDS is a bigger piece of work than an in-house platform, so which system you are on changes the size of the job rather than whether it is possible. Once it is wired, a confirmed booking syncs across with the customer record, the payment and the conversation attached.

Keep reading

Other builds, taken apart the same way

Want to talk about yours?