KLM runs promotional landing pages with sharp fares for a selection of destinations. A list of cities and prices is clear, but it doesn't help a visitor who hasn't decided where to go. Together we built an AI chatbot that thinks along on the campaign page: it asks what the visitor is looking for, suggests matching destinations from the offer and takes them to booking in one click.
From offer to destination
A campaign page such as "Where will you go in 2027?" lists dozens of destinations with a fare. Anyone who already knows they want Lisbon finds it quickly. But many visitors arrive with a question like "where can I go in May for sun and a good city?". A list doesn't answer that.
A chatbot that inspires first
The chatbot is built to inspire before it asks for dates. It asks about the visitor's wishes and suggests destinations that fit. Every suggestion appears as a card with a photo, a short description of the place and the offer fare. Below it are follow-up prompts such as "Tell me about Curaçao", and from the card the visitor goes straight to the booking page.
The chatbot only talks about the destinations the campaign is currently promoting. The page sends that list with every message, and it is the only source for names, fares and discounts. So the chatbot never mentions an offer that doesn't exist.
Safe at KLM's scale
A chatbot on a KLM campaign page gets a lot of visitors, and not everyone comes for inspiration. That is why it is surrounded by a series of checks:
- Transparent. The travel assistant states at the top that it is an AI assistant and that answers may contain mistakes, with a link to the disclaimer.
- Every message is screened first by a separate model before the conversation continues. Abuse and attempts to manipulate the chatbot are refused.
- Fixed answers for fixed situations. A question about an existing booking or baggage always gets the same answer: KLM Customer Service on klm.com can help. Anyone sharing personal data is asked not to.
- Every reply is checked for invented offers and answers that drift off topic. Links are only allowed to KLM, and the booking link always comes from our own integration, never from the model.
- Data stays in the EU. All model calls run in Europe, every conversation is logged for review, and transcripts are deleted automatically after a fixed retention period.
What it shows
This project shows how a front-office agent works in a high-traffic place for a brand that can't afford mistakes: it helps the visitor along, stays within the limits of the campaign and never says anything KLM wouldn't say.
