Skip to content
Fona

Solution

Applied AI for tourism

Tourism has volume, several languages and seasonal peaks: three things that make well-applied AI show up in operations and in margin.

Let's talkSee case studies

The problem

A lot of repetitive text, in four languages, with peaks

Answering enquiries, translating listings, classifying reviews and updating content eats hours from a small team, and in high season it cannot keep up.

Signs this is you

  • Repeat enquiries in several languages that overwhelm the team in season
  • Listings and descriptions translated by hand every season
  • Reviews left unanalysed: nobody knows what fails systematically
  • New content takes weeks to exist in all four languages

What we do

What we do

Assisted multilingual replies

Draft replies to common enquiries in the customer's language, with human review and a clear limit on what it must not answer on its own.

Review analysis

Automatic classification of reviews by topic and sentiment to see what recurs and prioritise real improvements.

Content generation and translation

First drafts of descriptions and listings in several languages, from your tone and your data, so the team edits rather than writes from scratch.

Operations support

Extracting data from booking emails, reconciling with the booking system and flagging inconsistencies.

Measured cost
per operation, before scaling
4 languages
without multiplying the team
With limits
what it does not answer alone, defined from the start

How we do it

How we do it

  1. Pick the case with a return

    Of everything that could be automated, we start with what saves measurable hours and carries low risk if it is wrong.

    DeliverablePrioritised case with an estimated saving

  2. Pilot with evaluation

    A set of real cases with the correct answer annotated, to measure whether it works before scaling.

    DeliverableMeasured pilot and cost per operation

  3. Production with limits

    Rollout with an escalation threshold to a person, a decision log and a monthly review of failures.

    DeliverableSystem in production and review plan

Technologies

  • Claude
  • OpenAI
  • Open models
  • Python
  • Job queues
  • PostgreSQL
  • pgvector

FAQ

What people ask us

Is the AI going to answer customers on its own?

Only where the risk is low and with review. For the rest, it generates a draft that a person approves. The limit of what it does not answer alone is defined at the start.

What about customers' personal data?

What is sent to the provider is minimised and the transfer is assessed if it is outside the EU. It is part of the design, not an add-on.

How do I know it works and not just that it impresses in a demo?

With a set of real cases evaluated and the cost per operation calculated. Without that number, it is a prototype with users.

Next step

Half an hour well spent

Walk us through the problem on a short call. You leave with a first read on how we'd approach it and what it would involve — no commitment, no sales deck.

Let's talkHow we work