From navigation to conversation: the AI ​​infrastructure to compete in the new hotel ecosystem

The user browsed, compared rooms, and finally made a reservation. The website became a stable model upon which the hotel industry built its digital sales project with varying degrees of success.

But this model is beginning to change.

More and more travelers are asking questions instead of browsing pages. We want immediate answers. We've become impatient.

“Do you have family rooms available for this weekend?”
“What’s the price for tonight for one person with breakfast and late checkout?”
“I’d like a room with a king bed and ocean view from June 15th to 18th.”
The interesting thing is that conversation has always been the most natural way to interact. This is demonstrated by the fact that many hotels still generate a significant portion of their sales by phone and email.

To cater to this “new” conversational user, our first reaction was the easiest and most intuitive: adapt our successful web-based model. This was also our initial thought at Mirai. However, what seemed logical and simple soon became anything but. What were once advantages turned into problems and doubts. We paused and reflected: do we continue adapting or start from scratch? We opted for the latter.

The web will lose its prominence… but how much and when?
Since the web wasn't designed to interpret a conversation, we users have learned to interact in other ways: browsing pages, clicking, and filling out forms. And we've made it a habit.

But this is starting to change.

AI assistants, such as ChatGPT or Gemini, thanks to language models (LLMs), are able to understand natural language and hold conversations with users, something that until recently was only possible between people.

And if users can interact in real time through conversation—without needing to make a phone call—the role of the website as an interface will be threatened, as it will face increasing competition from 100% conversational channels that exist outside of a browser, such as WhatsApp, and especially from the rise of AI assistants like ChatGPT or Gemini. The website will remain a very important direct channel, but it will no longer be the only point of interaction with the hotel. How will these new “direct channels” be adopted? Only time will tell.

For the hotel industry, this has a clear implication: its direct channel must be able to understand, respond to, and close a booking within a conversation, regardless of the medium in which it takes place.

AI, the rapid evolution from product to infrastructure.
Most providers—including ourselves at Mirai—initially interpreted AI as new products or new functionalities on top of the existing infrastructure: some still common examples are adding AI to the chatbot, functionality in the web engine that changes behavior according to an AI, making the MCP-ready engine.

After almost two years of innovating, testing, and above all, making mistakes and redoing everything several times, we have reached another conclusion: if AI changes the paradigm and moves users from web to conversation, the systems used by hotels must be natively designed to be able to operate in that environment.

And that can't be solved simply by adding a new piece or layer to the existing system. It requires something more structural: an infrastructure prepared for a conversational and agentic environment.

An AI infrastructure that feeds on the same data.
When you open this new conversational door to your business, an important question quickly arises: where do the data and capabilities used by these assistants come from? The answer is usually: from a proprietary database disconnected from your systems. And this is a double problem: on the one hand, the cost of inconsistency (giving one answer on the website and another via WhatsApp destroys customer trust) and, on the other, the inability to scale. If the content, inventory, and pricing are not the same, the user will perceive that they are talking to two different companies.

All the interfaces in your direct channel—website, contact center, AI conversational assistants—must operate on the same knowledge base and with the same operational capabilities. This allows you to offer a consistent guest experience. A traveler could check availability on your website, book through your booking engine, modify the reservation later via WhatsApp, and cancel it through ChatGPT.

If the underlying data (content, inventory, pricing, and booking information) differs between web and conversational search engines, the user experience will be poor, and they will perceive themselves as interacting with two different systems—an undesirable scenario. Conversational infrastructure should be built as an evolution of the hotel's existing digital infrastructure. The solution should be integrated with your current systems, not implemented as a plugin.

The reality, however, is that many hotels still lack systems prepared to operate in this way. Many rigid and outdated technologies are deeply integrated into hotel operations. In these cases, deploying a conversational AI infrastructure to complement existing systems is a viable option, but it's crucial that it draws on the same content, especially inventory, rates, and offers. The channel manager, CRS, or PMS is the most logical source to ensure this consistency.

The five layers of a conversational and agentic infrastructure
To prepare for this new conversational environment, hotel systems will need to rely on an infrastructure composed of different fundamental layers that interact with each other and form a new operational architecture. 

1. Booking engine ready for conversation

The first key element is the booking engine.

Until now, the booking engine was designed for forms and web browsing. But in a conversational environment, the system must be able to interpret intentions, quote prices, check availability, and close the booking within the conversation.

This implies a significant change: the booking engine ceases to be simply a website and becomes a transactional capability accessible from any conversational interface.

2. Agent-readable structured hotel database

For a conversational system to function correctly, it needs reliable, structured, and up-to-date information.

Today, in many hotels, that information is scattered among:

Web pages,
PDFs,
channel descriptions,
manual team responses.
An AI assistant, however, needs a clear knowledge base where it can find answers to questions such as:

Schedules,
Services and Facilities,
Room Types,
Policies,
Points of Interest:
This database becomes the single source of truth for machines and assistants. Without it, any AI system will end up giving incomplete or incorrect answers. Almost 60% of the questions travelers ask are about the hotel, not the reservation. To secure a booking, we must first answer users correctly. Otherwise, the user may abandon your direct channel in search of that answer.

Therefore, the first step for any hotel that wants to prepare for AI is to organize and structure its knowledge. The master or canonical database is a critical starting point.

3. Conversational Assistant

Building upon this operational foundation—the conversational engine and knowledge base—comes the third element: the AI ​​assistant. This component, leveraging the power of LLMs, provides the understanding of the guest's natural language and manages the interaction. This includes interpreting questions in different languages, maintaining the conversation's context, guiding the user, and connecting to the booking engine when appropriate. The assistant should be able to automate conversations across multiple channels, such as web chatbots, WhatsApp, Instagram, email, and phone (conversational but with voice).

What many hotels currently do manually can now be largely automated, improving operational and cost efficiency and guest satisfaction. This doesn't mean a complete replacement of human teams, but rather freeing them from repetitive questions and actions so they can focus on higher-value tasks: pursuing leads, closing complex bookings, or intervening in conversations when truly necessary.

4. Connection with external agents via MCP

Once you have automated your own conversational channels, the next step is to open the hotel to the new intermediaries of the AI ​​era: intelligent agents like ChatGPT, Gemini or Claude.

For this to work, the hotel's systems must be able to securely and systematically present their availability, pricing, and booking capacity to these agents. This is what's called Agent AI.

That's where interfaces like MCP -Model Context Protocol- come in, allowing the hotel to provide its own controlled and structured content to AI assistants, as well as providing agentic booking capabilities.

It is important, however, to clarify the current scope of MCP. Having an MCP server does not currently imply greater automatic visibility for AI assistants. The discovery layer—how assistants find and prioritize hotels—remains an area still under development and will largely depend on the assistants themselves.

5. Governance and control

Finally, if you're going to allow your guests to chat with you, you need a layer of governance and control. You can't go from a web channel where you measure everything to a conversational channel where you don't know what's happening.

Before opening conversational channels, you should make sure you can:

See what conversations are taking place
, monitor responses,
intervene in conversations when necessary,
analyze user performance and satisfaction,
and continuously improve the system to respond better each time.
The Mirai Omnichannel team can also manage this part for you.

AI managing conversations cannot be a black box. It must be a transparent and governable tool.

How do we begin to build this infrastructure?
The transformation to a conversational environment doesn't happen overnight and must be planned carefully but decisively.

The starting point is your current providers, primarily the booking engine, CRS/channel manager, and PMS. They are the best candidates to offer you this infrastructure and integrate it with your existing systems.

The order in which to assemble the pieces should be this.

Step 1 — structure your hotel's knowledge base

Create a clear and up-to-date database of your hotel's information. In a format that is structured and readable by machines or agents.

Step 2 — ensure that the booking engine can operate in conversation

Expand the functions of your booking engine so that it can quote and, very importantly, also close bookings based on an intent, not just a form.

Step 3 — deploy a conversational assistant

It incorporates an AI assistant that gives you the understanding and automates the conversation with your guests on the largest number of channels.

Step 4 — Open the infrastructure to external agents

Open your infrastructure to AI assistants like ChatGPT or Gemini through a system like MCP, where agents can natively and directly query and book, skipping any interaction with your website.

Step 5 — Establish mechanisms for continuous monitoring and improvement

Finally, add a tool that gives you visibility and traceability of what's happening. It should allow you to intervene if the conversation requires it and provide advice on how to improve your knowledge based on the questions customers ask.

At Mirai, we already offer our clients our AI infrastructure natively integrated with the web engine and the rest of the stack.

Conversational booking engine: to respond to your customer in conversations.
Sarai: the AI ​​assistant that understands and responds with your customer.
Knowledge: the canonical or master database.
MCP: the interface that connects you with AI assistants like ChatGPT.
Lobby: the tool that provides governance, transparency, traceability, and intervention capabilities.

An opportunity to strengthen direct sales.
The arrival of conversational AI represents more than just a technological shift. It can also redefine the relationship between hotel and guest.

If hotels build the right infrastructure, they can interact directly with travelers on any channel where the conversation arises, maintaining control of the relationship and the sale. This will give them a competitive advantage over OTAs.

Instead of simply reacting to market changes, they have the opportunity to lead this new stage.

Source: Mirai.