Lena Concierge: answers that fit the question and the season

Lena Concierge is working to give guests in Saas-Fee more relevant answers: to the current question, about the right place and for the right season. Between 2 and 8 October 2026, conversation rules, supplementary hints and tourism data were revised. The latest code changes were deployed to the public website on 9 October; their effect in guest operations has not yet been demonstrated. Reviewed on 9 October 2026.

Why accurate information can still be the wrong answer

Someone asking for directions to their hotel is not necessarily looking for another room. Someone mentioning the Aqua Allalin pool is not necessarily asking about the similarly named mountain restaurant. A summer activity answers an autumn question only if it fits the travel dates.

An AI concierge therefore needs more than accurate local facts. It must understand what the guest means now, which facility they mean and how current the information is. These connections are the focus of the latest work on Lena. They should reduce unnecessary hints and misleading recommendations; the changes alone do not establish a measured improvement.

What Lena worked on between 2 and 8 October

The following changes entered the development version on 7 October 2026 and were deployed to the public website on 9 October. This establishes the released code version, but neither a complete update of tourism data nor a measured improvement in answers.

A greeting should not repeat an old restaurant list

A bounded rule was added for standalone greetings, thanks and short acknowledgements in the five chat languages. For recognised messages, the language model receives no previous conversation window. A simple “Hello” after a restaurant recommendation should consequently be handled as a greeting.

Real follow-ups such as “And tomorrow?” retain the earlier context. This is deliberately limited recognition, not a guarantee for every change of topic. Another mechanism can flag similar answers. Repetition is not automatically an error: a guest asking the same question again may need the same information.

Directions to your own hotel should not start an accommodation search

Another change distinguishes known phrases about a guest’s current stay from a search for accommodation. Questions about their own hotel should therefore receive no additional Booking.com search link. The appropriate hint remains possible for a genuine accommodation search.

The overlapping names Aqua Allalin and Allalin mountain restaurant were also considered: mentioning the pool should not on its own add the restaurant’s opening hours. Recognition uses fixed phrases and known names. Other ways of expressing a question still need checking.

Recommendations need seasonal conditions and traceable sources

On 7 October, selected restaurant and activity information was revised in the development version. The changes remove unsupported preferences for individual businesses, add sources and correct selected season and opening-hours information. Without review scores in the data, a fixed recommendation should not appear as an objective ranking.

A publicly verifiable example explains why this matters: on its marmot information page, Saastal Tourismus says marmots enter hibernation around mid-October and become active again from May. Source checked on 9 October 2026. This does not guarantee seeing an animal on a particular day. A concierge must treat seasonal information as a condition.

These changes are not a complete check of every business. After deployment, the records actually updated and their availability in search need checking. The changes do not establish automatic, comprehensive monitoring of all opening hours.

What resorts can learn from current digitalisation projects

On 20 August 2026, the Valais Tourism Observatory Tourobs published a Digital Transformation Canvas for destinations. The tool connects guest needs, data architecture, processes and responsibilities. It calls for structured, trustworthy and continuously maintained data. Another tool does not replace that work.

The AI Local Guides provide a practical industry example. Basel Tourismus reported on 9 February 2026 on digital travel companions based on local knowledge maintained by tourism organisations. This shows how curated information can be combined with a conversational interface. It is an independent project, not a Lena partnership or a measurement of Lena’s performance.

For Lena, this suggests a practical task: every recommendation needs suitable text, an accountable source and a traceable indication of freshness. This is a development priority, not a function already implemented comprehensively.

The potential benefit for guests, accommodation and destinations

Guests: fewer irrelevant additions

A guest should receive directions in response to a directions question and a connected answer to a follow-up. Seasonal information should help with planning without promising current availability. Whether these changes reduce unnecessary journeys or follow-up questions must be measured in operation.

Hotels and holiday apartments: coherent information

Accommodation providers could support guests better when the concierge distinguishes their existing stay from a new booking search. Maintained information about activities and businesses could also reduce the need for hosts to correct replies. No demonstrated time savings or reduction in reception workload is available for these changes.

Destinations: data maintenance as a shared task

A destination management organisation can define which information comes from businesses and who verifies it. In a mountain region, travel dates and operating seasons belong to that work. Lena can provide an interface for the information; a shared checking and updating process remains an organisational task. Such a process is not established as operational for Lena here.

Next steps: checking data and comparing answers

Following deployment, the same questions should be compared before and after the changes: a greeting after a restaurant conversation, directions to the guest’s own hotel, a question about Aqua Allalin and a seasonal activity. Checks must cover relevant answers as well as real follow-ups that need to retain context. This comparison is a next step, not a published result.

Outstanding data work includes verifying the applied changes, search availability and other questionable profiles. Responsibilities and deadlines for seasonal updates should then be defined. The number of irrelevant hints and factually wrong recommendations would be useful operational measures; no values are claimed here.

On 8 October 2026, an optional connection for the IBM Granite model family entered the development version and was deployed with the website’s code on 9 October. Use requires a configured model host and an explicitly selected model. Whether Granite suits particular Lena questions must be compared using the same specialist questions, languages and knowledge sources. Another model replaces neither current tourism data nor checking its answers.

Related reading

• •Why tourism AI needs to understand the resort: place context and separate knowledge areas.
• •What must happen after a guest request is forwarded: the business’s reply and confirmation.