Tourism data in the age of AI – a decisive factor AI & Data Quality

Interview with Sandy Rischette, Head of Digital & Data Management
 

1. What does ‘data quality’ mean exactly in the context of LFT – and why has it become a strategic topic?   

Sandy Rischette: The quality of the data in our national tourism database is defined by four aspects. Data needs to be...

Complete: Is all the key information covered – opening times, prices, descriptions, contacts, photos? And is it available in all four languages (DE/FR/EN/NL)?
Coherent: Is the data consistent across languages and adapted to the various distribution channels – the website, Visit Luxembourg app and digital partner terminals?
Up-to-date: Is the information reflective of the current situation and not a year out-of-date?
And, finally, correct: Does the information paint a picture that corresponds to a visitor’s actual experience?

Data quality has always been a core aspect of our work. Providing a national tourism database as a single source of truth for the whole country avoids a situation where visitors find contradictory information across different channels.

It’s the scope that’s changing right now. Lots of AI systems have direct access to this data, including search engines, chat bots and travel planning agents. AI cannot compensate for data that’s incomplete or outdated. “Garbage in, garbage out” applies more than ever when generative AI is involved.

2. What is the current status of your data? What are the biggest challenges surrounding that data being complete, coherent and up-to-date?

Sandy Rischette: Without referring to specific figures that belong to an internal audit, it’s straightforward enough to summarise the structural challenges. And they’re the same for most destination marketing organisations (DMOs). The completeness of the data varies, with some partners taking care to provide a full picture and others not even inputting their opening hours. There’s a sizeable gap between large establishments and small, independent providers.

When it comes to data being up-to-date, we find that seasonal updates – summer and winter opening times, temporary closures and new offers – arrive late or not at all when we leave this entirely in the hands of our partners. Often, the problem isn’t that partners don’t have good intentions. Lots of partners just don’t understand how important it is to keep their data up-to-date. They don’t naturally make the link between outdated information and disappointed visitors.

Turning to the different languages, content is usually available in DE, FR and EN, while the NL version is often incomplete, out-of-date or missing altogether – even though the Dutch-speaking market is well represented by our visitors.

These are structural challenges facing our sector and not Luxembourg-specific issues.

3. How are you using AI in your processes now – from capturing the data to using it?

Sandy Rischette: We’re still at the early stages of our AI strategy, which covers 24 months. This phase is deliberately focused on governance, solid foundations and initial experiments, so we’re not going live before the framework is in place.

But we are already using AI to support our internal workflows, drawing on it as a tool for writing, summaries, translation, research and initial quality checks within our database. The AI governance framework is being created, covering a system register, risk assessments and compliance with the EU AI Act. This is the requirement for any live deployment. We’re not going live before we’ve laid these foundations.

In the medium term, we want to integrate advanced tools that make it easier for us to keep the data in our database up-to-date, with some automation, as well as an internal agent that draws on this data to streamline our team’s working day. Looking ahead to the longer term, end users should be benefiting from the data – via a conversational assistant on the website and in the Visit Luxembourg app that can answer specific questions about travel planning.

4. Is it possible to automate the improvement of data quality with AI or is human input always required? Where do you draw the line – and what form might proactive AI support for partners take exactly?

Sandy Rischette: There is a clear line between AI as an assistant and AI as a decision-maker – at least whenever it comes to published data.

We can use AI to automate the detection of anomalies without ongoing manual approval being required. For example, AI can identify unusual or missing opening hours, photos that don’t have a high enough resolution or that are in the wrong format, and inconsistencies between the information provided by partners in our database and reliable third-party sources. Partners can also use AI as a tool that will work through the information they have entered and flag up any missing information.

Equally, certain steps absolutely do need human input – such as any actual change to a published entry, editorial content, pricing and accessibility information, issuing of quality seals and distribution of content to partner channels.

And the reasoning behind it is simple – it comes down to responsibility and trust. If AI automatically changes the opening times of a restaurant and gets it wrong, visitors and partners will hold LFT responsible. AI can speed up and improve processes, but it can’t replace them.

5. You have hundreds of tourism partners in your database. How do you organise regular updates to all those entries – and what is the biggest barrier you face when you ask hotels, restaurants and attractions to manage their own data?

Sandy Rischette: imx.Platform is based on a partner self-service model. Tourism providers – accommodation providers, restaurants and attractions – are responsible for updating their own entries. Meanwhile, municipalities and syndicats d’initiative manage the attractions in their area. Theoretically, LFT and the regional tourism offices (ORTs) have a role to play in quality management and support. But, in practice, it currently tends to be the ORTs that are most actively managing data. The model is scalable but fully dependent on partner engagement.

The greatest obstacle isn’t technical in its nature; it’s a question of priorities. A hotel or restaurant manager manages day-to-day operations and isn't likely to view updating an entry on a national platform as a matter of urgency – until the day a visitor arrives at their establishment’s door with the wrong information.

Plenty of other factors exacerbate the problem. Let’s start with fragmented platforms. Our partners are juggling Google Business, Booking.com and imx.Platform. That takes a lot of time and effort. Then there’s the lack of direct feedback. Our partners can’t directly see the impact an incomplete entry has on their visibility or booking figures. Finally, there’s a broad spectrum of digital literacy. The gap between a large hotel with its own marketing department and an independent accommodation provider can often be huge when it comes to using digital tools.

Our response to this challenge is twofold – keeping the entry fields as simple as possible and demonstrating the value of providing complete entries so partners can understand the tangible benefits. This is an ongoing process.

6. What direct impact does poor-quality data have on the visitor experience – whether that’s on the website, in the app or at the tourist information offices?

Sandy Rischette: The direct impact manifests itself on three levels – on websites, in the app and at digital terminals. When a visitor spots incorrect opening times, a wrong address or an inaccurate photo, trust is lost – and it’s really difficult to win trust back. Staff at the tourist information offices refer to this data. That means that incorrect information is not restricted to digital contexts. It’s also infiltrating in-person conversations – and impacting the lived visitor experience on the ground.

But the most critical risk of all is now emerging – AI systems. Google and ChatGPT access and distribute our data – the same will be true of our own assistant in future. This means that an error in the database is not confined to one channel. It is multiplied across many channels, often without any way to correct it in real time.

The reputational risks associated with incorrect information automatically increase every time another system starts using us a source. 

7. What are the next priorities when it comes to using AI to manage tourism data?

Sandy Rischette: Once we’ve finished laying the governance foundations, which we’re working hard to achieve at the moment, we’ll be focusing on four key priorities.

Semantic search: The shift from keyword searches to intention-based searches on visitluxembourg.com and in the Visit Luxembourg app. The idea is that a user can search “things to do with kids on a rainy weekend” and receive relevant suggestions rather than zero search results or a list of randomly unrelated results.

Anomaly detection and automated quality checks: This will allow us to proactively identify out-of-date or inconsistent data in imx.Platform.

Visitor-focused AI assistant: A multilingual conversational assistant that can answer specific questions relating to travel planning, drawing data from our national database, with the go-live date planned for 2027.

And, last but not least, internal agents: AI tools that streamline the working day for LFT teams – supporting editorial work, reporting, data entry, vibe coding and so on.

8. How will the role of data quality in an organisation like LFT change in the next three to five years, particularly in light of AI assistants drawing directly on your data?

Sandy Rischette: Data quality will shift from an operational topic to a strategic asset. Until now, incorrect information on our website could only have a limited impact. Visitors could cross-check on Google, call the tourist information office or visit the partner’s website directly. We had safety nets in place. In future, when an AI assistant tells a visitor that a museum is open on Mondays – on the basis of our data – the visitor will believe that to be true without verifying the information. The scope of LFT’s responsibility as a reference source is shifting significantly.

We’re expecting three main developments. Firstly, structured data quality will drive visibility. LLMs and AI agents favour sources that are well structured, consistent and updated regularly. DMOs that invest in the quality of their database now will be indexed more effectively and cited more frequently. Otherwise, they risk becoming invisible in the AI ecosystem.

Secondly, data governance will become more formal. There will be a need for clear regulations setting out who can change what, what approval is required and what process has to be followed. This won’t just be an internal workflow issue anymore – it will be necessary for compliance purposes.

Thirdly, collaboration with tourism service providers needs to evolve. Data quality can’t be based solely on partners’ good intentions anymore. It needs clear incentives, tools that are as intuitive as possible, and probably also minimum standards requiring data to be complete to a certain extent before it is shared across certain channels.

In five years’ time, we won’t be asking “Is the data on our website up-to-date?” Instead, the question will be “Is our data trustworthy in the global AI ecosystem?” These are two very different challenges – and we’re already working hard to lay the foundations to overcome them.