Case study: Travel

From a dead-end chatbot to an AI agent that books the call.

Secrets du Siam ran a scripted Tidio widget that stalled ready-to-book travellers. We replaced it with a custom conversational agent that qualifies them and schedules the discovery call, at a lower running cost.

Up to 60-70%estimate, to be confirmed

of repetitive enquiries handled without a human

Under 1 minuteestimate, to be confirmed

first response on common questions, down from hours

About 10x lowerestimate, to be confirmed

cost per handled enquiry versus the human channel

Context

Secrets du Siam is a Thailand-focused travel agency selling tailored trips. Like most agencies its size, the website ran a Tidio chatbot: a scripted, button-driven flow that answered a fixed menu and stalled the moment a traveller asked something off-script.

Travel enquiries are messy by nature: dates, budget, party size, what to see and what to skip, season, visas. A decision tree cannot hold that conversation. The bot was a deflection tool that deflected the wrong things: it pushed engaged, ready-to-buy visitors into a dead end instead of into a booking.

Challenge

Three problems, one root cause. The brief was not to add another chatbot. It was to replace the rented script with an owned agent that actually converts, in the languages travellers speak, and that hands the sales team qualified calls instead of transcripts.

1

The scripted bot could not qualify

It captured clicks, not intent, so the sales team received either nothing or noise.

2

It could not act

It could not propose a slot or book a discovery call, the one action that turns a curious visitor into a pipeline lead.

3

It billed for a low ceiling

A recurring SaaS fee for a tool whose limits were baked in by design.

Solution: an agent that qualifies, books and writes back

We replaced Tidio with a custom conversational agent built on a large language model. The difference is structural: a chatbot answers, an agent qualifies, books and writes to the system.

  1. 1

    A real conversation

    The agent holds a natural conversation about the trip in the traveller's language, grounded in the agency's actual offer so it never improvises prices or itineraries.

  2. 2

    Qualification in context

    It captures the constraints that matter (destination focus, dates, budget band, party size, language) and turns them into a structured lead instead of a click log.

  3. 3

    Booking inside the conversation

    When intent is clear, the agent proposes a slot and books the discovery call directly in the chat: the action that turns a curious visitor into pipeline.

  4. 4

    Owned system, controlled costs

    The agency trades a recurring SaaS subscription for a system it owns: an isolated embed that loads without breaking the page, and a model-abstraction layer so there is no vendor lock-in and a lower cost per conversation.

Results

The figures below are market-derived estimates based on comparable conversational deployments. They will be confirmed against the agency's live data before any number is published as a fact.

Up to 60-70% of repetitive enquiries handled end to end by the agentestimate, to be confirmed

First response in under 1 minute where the old bot left travellers waiting for hoursestimate, to be confirmed

Cost per handled enquiry roughly 10x lower than the human channelestimate, to be confirmed

The sales team receives qualified discovery calls instead of click transcripts

The honest framing: the agent does not replace the travel advisor. It protects the advisor's time by absorbing the repetitive front end and surfacing only the travellers worth a human call.

Stack

Custom conversational agent on a large language modelDiscovery-call booking triggered in conversationStructured lead qualification into the agency pipelineMultilingual by designIsolated embed widgetModel-abstraction layer, no vendor lock-in

Your chatbot answers. Should it be booking?

Start with a free AI audit: we will tell you where a real agent converts, and where it does not.