How Novatix supported EHL in deploying an intelligent assistant that reinvents the digital experience for its future students.

Context

EHL (École hôtelière de Lausanne) is one of the world's most prestigious institutions in hospitality and management. Its website, visited by several million people each year, is the first point of contact for prospective students, their parents, and industry professionals interested in its programmes.

In the summer of 2024, EHL and Novatix began a joint reflection: how to concretely integrate artificial intelligence into the school's processes—both to gain efficiency and to affirm its positioning as a resolutely innovative institution?

Several use cases were explored. The public chatbot, capable of instantly answering visitors' questions, quickly emerged as the top priority. You can try it today directly on ehl.edu.

The challenge

The EHL website is rich, comprehensive, and complex. It spans several hundred pages across many categories:

• Academic programmes
• Campus life
• Admissions
• Tuition fees
• Blog
• And much more

Visitors can find all the information they need, but it is not always easy to locate.

As a result, EHL teams received a significant volume of calls and emails every day about recurring questions—tuition fees, programme content, application processes, student life—whose answers were already available on the website.

At night, on weekends, or during holidays, these requests remained without an immediate response, even though EHL attracts an international audience spread across all time zones.

The objective was therefore clear:

• Offer a smooth experience to visitors
• Provide precise, reliable answers grounded in official content
• Guarantee 24/7 availability
• Reflect the institution's standards of excellence

Our approach

Start from the business problem

At Novatix, every AI project begins with the same question:

What is the business problem to solve, and what concrete value are we trying to create?

For EHL, the challenge came down to two indicators:

• Reduce lead qualification time
• Limit the number of prospects lost due to a lack of response at the right moment

All subsequent technical choices were guided by these objectives, starting with foundational work on data.

Map the data

The EHL website contains hundreds of pages across multiple domains, supplemented by numerous PDFs and supporting documents.

Before even developing the chatbot, we had to:

• Map all content
• Categorise it
• Define priority scopes for the first version

In parallel, we worked with marketing and admissions teams to identify the questions most frequently asked by visitors.

The goal: build a use-case repository based on real-world needs rather than assumptions.

An enriched RAG architecture

On this foundation, we built a Retrieval-Augmented Generation (RAG) architecture.

The goal was not simply to index content, but to enrich it.

Each information fragment is associated with metadata:

• Title
• Description
• Page context

These enrichments significantly improve the relevance of answers.

The system is automatically re-indexed every month to stay synchronised with the website's ongoing changes.

Align tone and secure responses

Answering correctly is not enough.

For an institution like EHL, the chatbot also had to adopt the right tone.

We therefore integrated the school's values and communication codes into the model's behaviour.

A second layer of security was then added:

• A lightweight model checks each response before it is sent
• It verifies compliance with the school's policies
• It protects the system against hijacking attempts

An essential requirement for an institution of this scale.

Agentic capabilities to capture value

The chatbot does more than inform.

When a visitor shows strong interest in EHL, it can:

• Collect their contact details
• Automatically forward this information to the school's CRM
• Attach the full conversation context

Recruitment teams thus receive qualified leads without breaking the user experience.

Infrastructure designed to scale

The solution relies on an architecture capable of supporting hundreds, even thousands of simultaneous users without performance degradation.

It specifically anticipates peaks related to:

• Marketing campaigns
• Admissions openings
• Announcements of new programmes

Everything is hosted on EHL's Swiss Azure tenant, in compliance with:

• GDPR
• Swiss nLPD
• The institution's privacy policy

Validation: human and automated testing

Before going live, dozens of EHL collaborators tested the chatbot.

These human tests were complemented by an automated campaign using the LLM-as-a-Judge method to measure:

• Answer accuracy
• Tone compliance
• Absence of hallucinations

Every response systematically cites its sources to guarantee full traceability.

The result: an integrated, high-performing chatbot under continuous monitoring

Deployed on ehl.edu, the chatbot is now fully integrated into EHL's visual identity and accessible from the site's main pages.

The project did not stop at launch.

A comprehensive reporting setup now tracks the two main business indicators defined with EHL:

• Capture prospects who were previously lost
• Accelerate their qualification

Key figures

• 63% of questions are asked outside the school's opening hours.
• 5 to 10% of users voluntarily share their email address to be contacted again.
• Approximately 80% of questions are resolved directly by the chatbot.
• When the chatbot cannot answer, it redirects to the right department in 70 to 80% of cases.
• Average latency dropped from 7–8 seconds to 3–4 seconds thanks to successive optimisations.
• Most conversations take place in English, with users primarily located in Europe and Asia.

Reporting also makes it possible to analyse:

• Activity peaks
• User profiles (prospective students, parents, professionals)
• Most consulted topics

This information continuously feeds improvements to the chatbot, the website, and EHL's broader digital strategy.

What's next

For the chatbot

Upcoming developments include:

• Even more performant models
• New agentic capabilities
• Automatic brochure delivery
• Integration with registration forms
• A connection to Google Analytics to better understand the user journey

In the medium term, the chatbot will also cover:

• Graduate Studies
• The Passugg campus
• The Singapore campus

For the collaboration

This project marked the beginning of a lasting collaboration between EHL and Novatix.

Today, Novatix supports the institution as a strategic partner on several AI initiatives.

Future projects will go beyond the public experience to also support teaching and administrative teams, automating low value-added tasks.

What we learned

Like any ambitious project, this one brought us several lessons.

Data is the real subject

Even before choosing a model, content quality, structure, and prioritisation determine the performance of a conversational assistant.

We devoted a significant share of the project to this foundational step.

Over-provision to last longer

We deliberately designed infrastructure capable of absorbing very high volumes of simultaneous users.

Even if these peaks are not yet reached daily, EHL now has a platform ready to support its growth.

Guardrails are essential

For an international institution, robustness against misuse is as important as answer quality.

The dual-model architecture proved to be a decisive choice.

Co-build rather than deliver

The project's success also rests on the continuous involvement of EHL's marketing, admissions, IT, and leadership teams.

This close collaboration made it possible to move quickly while guaranteeing a high level of quality.

Testimonials

"Novatix perfectly understood our challenges from the very first conversations. Beyond the technical quality of the chatbot, it was their structured, rigorous, and attentive approach that made the difference. We found in them a true trusted expert, with whom we are now building our long-term AI strategy."

Fabiano Ruggeri, Tech Solutions Director, EHL

"The chatbot has become a genuine acquisition lever for our teams. It allows us to capture prospects at the moment when their interest is strongest, including outside opening hours, and to turn these interactions into qualified leads that can be directly actioned. Thanks to the conversation context transmitted, our recruitment teams can engage in more relevant and personalised exchanges from the first contact, which significantly improves follow-up efficiency and conversion rates."

Marjo Jarvinen, Marketing and Recruitment Director, EHL

"This project embodies what we want to do at Novatix: transform a real business challenge into a structured, measurable, and sustainable AI solution. The trust EHL places in us pushes us every day to aim higher."

Guillaume van Lierde, Director, Novatix

Offer this experience to your visitors too

The EHL chatbot was built on Nova Bot, our solution dedicated to designing public conversational AI assistants, grounded in your institution's content and aligned with your identity.

Discover Nova Bot →