Hotel chatbot: from question box to sales conversation

Most hotel websites still ask guests to do the work: pick dates, tick filters, compare room types and hope the right offer turns up. A hotel chatbot promises the opposite, a conversation that asks what the guest wants and guides them to it. Yet many first attempts were little more than a question box in the corner of the screen. So what does a conversational assistant that actually sells and serves look like?

Adrien Mathieu, co-founder of Kleio.ai, builds conversational AI for travel and hospitality brands. In his view the problem is a matching problem: guests start with vague intentions, and websites answer with static filters and keywords that reflect how brands think, not how guests think.

Why filters and keywords fail the modern guest

A guest rarely starts with a room code. They start with an idea: a quiet weekend, a family trip, a stay with something to do nearby. In Mathieu's experience, consumers now begin that search online, more and more often exclusively, and they get lost because the web experience relies on filters and keywords chosen by the brand. Conversational AI bridges that gap by turning a vague wish into matching options, step by step.

He describes an early project with a leading travel company in France that sold mainly through close to 400 local stores. The company doubted its web channel and expected its customers, many of them fifty, sixty or seventy years old, to dislike a conversational experience. The opposite happened. Usage was intense from the first deployment, and the customers they interviewed said it felt like talking to a human, but online.

Does a hotel chatbot remove the human touch?

Losing the human touch is an objection Mathieu hears a lot, in hospitality as much as anywhere. First, guests are already used to this kind of experience, because a growing share of traffic arrives through AI search engines such as ChatGPT and Perplexity. Second, the assistant does not replace people. With complex products there is often a human at the end who converts the prospect. The point is to let guests find what they are looking for online and talk to a person when they want to, without forcing them either way.

Service during the stay matters just as much. Mathieu thinks hospitality can go further here than travel: guests have all kinds of questions during their stay, and they would rather send a text and get an answer instantly, 24/7, than pick up the phone every time. For calls that still come in, a hotel answering service makes sure none are missed.

From question box to sales conversation

The first generation of chatbots sat in a corner of the website and answered simple questions. Mathieu argues that the real value appears when conversational AI is present at every step of the online journey, for example as a central search bar, integrated naturally into the site. When brands do that, he says, engagement jumps from a couple of percent of visitors to a double digit percentage.

“That's really the key milestone. If there is usage, there is value.”

Adrien Mathieu, 17:29

Selling is harder than answering. A sales conversation can go in many directions: qualification questions, specific questions about a product, generic questions, data pulled from a product catalogue. Mathieu's team handles this with roughly 15 to 20 AI agents, each specialised in one task, coordinated by an orchestration layer built on a reasoning engine. The goal is a conversation with the quality of a human sales interaction.

Tone differs between service and sales.

“It's not a customer service conversation where you are stuck with the brand.”

Adrien Mathieu, 19:44

A prospect who is not engaged simply leaves. His answer is to bring added value content into the conversation, for instance by showing products visually. The more useful the exchange, the more the guest shares about their plans and personal details, and the further they move towards a booking.

What makes a hotel chatbot feel personal

Personalisation is often a buzzword, but in a conversation it has a concrete meaning. Mathieu's approach comes down to three elements that any hotel can use as a checklist:

• Recommendations that are refined during a multistep conversation, converging on the right product instead of showing everything at once.
• Integration with a lot of data, because personal offers are impossible without it.
• A tone of voice that fits the brand and reacts to what the guest says, including qualification questions that change with each answer, so that no two conversations are the same.

In his experience, more personalisation means more engagement. The guest feels the brand cares about their plans and keeps the conversation going until the booking is complete. The same idea drives upselling in a hotel guest app during the stay.

Agents talking to agents: where booking is heading

Mathieu expects the next step to be agent to agent interaction, whether that arrives around 2028 or 2030. A traveller asks the agent on their smartphone for options. That agent talks to the agents of different brands, retrieves the solutions, summarises them and can even pay.

“There might be a scenario where you don't even go over any website at all.”

Adrien Mathieu, 11:43

For independent and smaller hotels, he sees a real opportunity here. A hotel with its own agent that is good at promoting its brand and experience could, to a certain extent, bypass the OTAs, where it was often barely visible. He also believes OTAs are at risk if they ignore this trend. OTAs that connect to hotels should know what a hotel booking API must offer. Hotels that serve a very specific segment of the market and are the best at it stand to gain most.

Visibility in AI search is part of the same shift. Mathieu advises brands to think about how guests search for and think about their product, rather than how Google Search thinks about them, and to build real human content that no one else could write, including video. A hotel chatbot can only recommend well when the content behind it is rich and distinctive.

How to start with conversational AI in your hotel

Many companies end up with a list of ten, twenty or fifty AI use cases from consultants. Mathieu's advice is to start small in order to scale quickly:

• Pick one use case: the one you expect to bring the most value.
• Check feasibility: the data it needs must be available.
• Solve it fully, even on a small scope, before you look at other use cases.
• Aim for real impact on the top line or bottom line, not a pilot that leaves everyone half convinced.

He adds that AI is largely technology, so treat it as an engineering problem to solve. To hear his full reasoning on agents, AI search and direct booking, watch the full conversation.

Frequently asked questions

What is a hotel chatbot?

It is a conversational assistant on a hotel website or messaging channel that asks guests what they are looking for and guides them to the right offer. According to Adrien Mathieu, the strongest versions go beyond the first generation of chatbots: they sit at every step of the online journey, recommend personal offers and also answer questions instantly during the stay.

Will guests accept talking to an AI instead of a person?

In Mathieu's experience, yes. A travel company that expected its older customers to dislike it saw intense usage from the first deployment, and customers said it felt like talking to a human, but online. He adds that the AI does not replace staff: guests can still talk to a person whenever they want to.

How can a small hotel compete with OTAs using AI?

Mathieu sees an opportunity for independent and smaller hotels to run their own AI agent. In a future where travellers' agents talk to brand agents, a hotel whose agent presents its brand and experience well could, to a certain extent, bypass the OTAs. Hotels that serve a specific segment and are the best at it stand to gain most.

Where should a hotel start with conversational AI?

Start small and scale quickly. Pick the single use case you expect to bring the most value, check that the data it needs is available, and solve it fully even on a small scope. Only then move to the next use case. Mathieu advises aiming for real impact on revenue or costs rather than an endless pilot.