Holiday Inn Córdoba
- 86% of inquiries resolved automatically
- 23% increase in direct bookings
Customer story Google Cloud published a case study on WeSpeak
Google documented how more than 700 hotels across Latin America turn their messaging channels into a direct sales channel. Here are the key points of the case study.
The original article was published by Google Cloud. Read the full case study →
The published results
- 70% of guest inquiries resolved end to end, with no involvement from the hotel team.
- Up to 167% increase in direct booking conversion.
- 0 platform outages since migrating from a private server to Cloud Run.
- 70% fewer support tickets after adopting an agent architecture.
The problem isn't demand. It's the moment it arrives.
A hotel that receives inquiries on WhatsApp, Instagram, Facebook or TikTok already has demand covered. What it lacks is the capacity to respond at the exact moment purchase intent exists.
That intent lasts minutes. A guest who asks about availability on a Sunday at eleven at night won't wait until Monday: they keep searching. And when they find an answer, it often comes from an intermediary that charges a commission on a booking the hotel had already won.
That's the real cost of an unanswered message. It isn't a lost booking: it's a booking the hotel pays for twice, first in sales effort and then in commission.
WeSpeak exists to close that gap. A conversational sales agent that handles every inquiry the moment it arrives, carries the conversation with commercial judgment and takes the guest from the first message to a confirmed payment, without ever leaving the conversation.
Two hotels, by name and by the numbers
The case study published by Google cites results from specific clients:
Howard Johnson
- 89% of inquiries resolved automatically
- 21% increase in direct bookings
These are international chains operating in competitive markets. The difference isn't in receiving more inquiries, but in answering the ones they were already getting.
What's behind those numbers
- 01
A model chosen on evidence, not availability
Before choosing the model that powers the agent, the WeSpeak team ran more than a thousand automated tests on every alternative evaluated, measuring hallucination rates in real hotel scenarios.
That standard isn't a technical detail. A guest asking about a rate, a cancellation policy or whether breakfast is included needs the exact answer. An approximate response isn't a minor annoyance: it's a booking that falls through or a complaint at check-in.
- 02
Agent architecture, not a predefined flow
The case study documents a major design shift: moving from several models, each handling a different part of the conversation, to a single agent capable of handling all of it.
The agent has tools and decides how to use them. It checks availability, creates the booking and generates the payment link within the same conversation. It doesn't follow a decision tree: it responds.
In practice, a guest messages on WhatsApp asking about rooms. They receive options with photos, get their questions about parking or late checkout answered, share their details and receive a payment link. The booking is confirmed in minutes, without anyone on the team stepping in.
- 03
Infrastructure that matches what's at stake
Every minute of platform downtime is a booking left unanswered. That is why the migration to Cloud Run was a business decision before a technical one: containers that scale automatically with inquiry volume, with no fixed server that could drag down the whole system.
On top of that, there is a queue system with automatic retries. If an external service has an intermittent issue, the platform keeps trying until the guest gets a response.
“Hotels that use WeSpeak sell more—that is what the numbers show.”
What this means if you're evaluating AI for your hotel
The question is no longer whether artificial intelligence can hold a conversation with a guest. It's whether it can close a sale unsupervised, with the right information, at any hour, without the hotel taking on the risk of it going down at the worst possible moment.
The case study published by Google Cloud answers those three questions with results measured at hotels operating today.
