.svg){kind=logo}
Always accurately and politely addressing the user's need to know.
An AI voice chatbot that answers user questions 24/7, 365 days a year—that is the "AI Minarai" we developed. AI Minarai is not a product specialized for a specific industry, but a general-purpose AI chatbot that can be utilized in a wide range of business types and services. This time, as one of its use cases, we will introduce utilization scenarios in the hospitality industry, such as hotels and ryokans.
|
【Table of Contents】 |
Reasons for our commitment to testing
For example, in a hotel, it can accurately answer questions such as "Is there a parking lot?" or "What are the hours for the large public bath?" and if an answer cannot be found, it can directly connect the user to a staff member upon request.
In order for users to truly feel at ease using this service, the most important thing is that "the correct answer is always returned regardless of when, who, or how they ask." Therefore, we focus on "testing" as much as, or even more than, development.
Why are we so committed to testing? There are three main reasons.
1. Because user questions are always beyond expectations
Some users ask simply, "Where is the nearest convenience store?" while others ask in seemingly unusual ways, such as "I'm a bit hungry in the middle of the night, is there anything within walking distance from the hotel?" Users ask questions in many different ways. That is precisely why we anticipate every possible pattern when testing, so that the intent can be correctly understood regardless of how it is phrased.
2. Because ambiguous answers and mistakes disappoint users
Chatbots can sometimes return "plausible but incorrect answers" (hallucinations). Answering "Yes, we have one" to the question "Do you have a pool?" when there actually isn't one... such incorrect information betrays the user's expectations and causes trouble. We thoroughly verify AI Minarai's dialogues to ensure such mistakes do not occur.
3. To always maintain a polite and courteous response
Accommodation facilities are places where users should feel truly relaxed. Therefore, we believe it is essential to always maintain a polite and warm response. Whether we can face the user with unchanging politeness even when questions are ambiguous, phrased roughly, or sound like complaints. We repeatedly confirm the quality of these responses as well.
The reason we put so much effort into testing is not just for functional confirmation. It is all for the sense of security and satisfaction that makes users feel, "I'm glad I asked AI Minarai."
How does AI Minarai work?
To help you better understand our commitment to "testing," first we will introduce the basic mechanism of how "AI Minarai" operates.
AI Minarai adopts a technology called Retrieval-Augmented Generation (RAG), which combines a Large Language Model (LLM) with a search function. It may sound difficult, but this mechanism is very similar to the flow of how a person looks for information.
So, what would you do when you want to look something up about a hotel?
- First, you organize what you want to know in your mind.
- Next, you look for the hotel's website and check the FAQ page or service guide.
- Then, you look for an answer that fits your question from within the page.
- Finally, you reorganize the found answer into a form that is easy for you to understand.
AI Minarai performs search and response in almost exactly the same flow.
- Understand what the user wants to know
It receives questions such as "Is there a parking lot?" or "Is there a convenience store nearby?" and first organizes the intent. - Referencing the hotel's official information.
AI Minarai possesses official Q&As and service guides provided by hotel staff in advance as knowledge. - Find the best answer quickly
It accurately reads the user's intent and instantly searches for the most appropriate answer from within that knowledge. - Carefully summarize the answer
It communicates the found information in polite and easy-to-understand language, as if a staff member were answering directly.
As an even more convenient point, even if an answer is not found in the official information, it can offer to "connect to a human staff member" and connect you immediately if desired.
Behind the scenes of testing
Here, we introduce the specific testing initiatives that support the quality of AI Minarai. In order to realize an AI chatbot that is trusted in actual operations, we will explain the behind-the-scenes look at what perspectives we use for verification and what efforts we have accumulated.
1. Assuming various user situations
Situations where inquiries to a hotel occur can be classified into the following three categories:
- When considering a stay in the future
- When already booked and wanting to confirm details
- When information is needed immediately while already staying
The information sought by users varies greatly for each of these situations. For example, "parking availability" may be important if planning to come by car, "facilities for children" for a family trip, and "Wi-Fi stability" for a business trip.
Taking these differences into account, we begin testing by setting diverse user profiles (personas). We define typical usage patterns such as "families traveling by car," "short-term business use," and "individuals staying long-term for sightseeing," assuming specific user profiles.
Next, we prepare a set of questions for each assumed user and use them to repeat conversation simulations, verifying whether AI Minarai can return appropriate responses in various situations. Through such tests, we confirm whether consistent answers can be provided for different prerequisites and needs.
2. Supporting diverse question expressions
After preparing the question sets based on the aforementioned persona settings, the next point of focus is that "even for the same question, the way people say it is completely different." For example, the intent "Is there a convenience store nearby?" actually changes into various phrasings like those below:
- Straightforward expression: "Is there a convenience store near the hotel?"
- Expression adding context: "Is there a convenience store within walking distance for when I get a bit hungry late at night?"
- Elliptical expression: "Excuse me, are there any places selling food within walking distance?"
- Compound question: "Is there a parking lot, and also a convenience store within walking distance?"
All of these seek the same information (presence of a convenience store around the hotel), but the sentence structure and vocabulary differ significantly. For example, in internet search scenarios, the choice of keywords and expressions for the same purpose differs from person to person, and if an answer is not found, users may change to a different phrasing.
In the testing of AI Minarai, we prepare every possible question pattern for a single answer and verify whether it can respond correctly to any expression. This confirms that it can handle the variations in phrasing and ambiguity that occur in actual usage environments.
3. Verification of conversation continuity and context retention
Inquiries to a hotel are not necessarily completed in a single question-and-answer session. In actual usage environments, there are many cases where multiple interactions occur in succession. For example, after first asking "Is there a parking lot?", additional utterances such as the following may follow:
- Confirm details of the previous answer: "Is that free of charge?"
- Correcting a misunderstanding or rephrasing the question: "No, that's not what I meant..."
- Moving to a new topic: "Never mind" -> Next question / Tone becomes harsher
To handle such actual conversation flows, not only single question-and-answer capabilities but also context retention (management of long-term dependencies) and appropriate responses to emotional tone changes are required.
In AI Minarai testing, in order to confirm whether these conditions are met, we conduct simulations that reproduce multi-turn dialogue scenarios. Each scenario reproduces diverse conversational developments, such as the user asking follow-up questions based on the previous answer, rephrasing, proceeding to the next step if satisfied, or showing dissatisfaction and requesting escalation to human staff.
Through this process, we verify whether AI Minarai can respond accurately and politely in a stable manner even within complex dialogues close to the actual operational environment.
Summary
AI Minarai continues to undergo testing and improvement to establish reliability that can withstand actual operational settings while possessing the flexibility to be applied to a wide range of industries. We will continue to incorporate user feedback to improve accuracy and response quality, aiming to be a presence that can be used with peace of mind in even more workplaces.
Latest Articles
