Highlights

1.  Bélo can comprehend intent 

from guests’ utterances based on previous conversations and social context. It also has a robust error recovery system when 

corrections are needed.

2. When Bélo doesn’t have enough information to process the guest’s query, it prompts for clarification instead of throwing an error message denying to process the request.

3. Bélo can detect environmental cues like closing and opening of the door and use them as a prompt to start a conversation.

Dialog Flow Diagram

Process

Research

01 Competitive Analysis

We investigated existing CUI solutions in the general hotel domain to gain an understanding of the existing use of this technology in this domain and the challenges it poses.

We observed a clear disconnect between the bot’s capability and user’s needs. Since many chatbots focused on the search and booking stages of a hotel stay, once reservation is made, the bots become useless. Thus users are forced to default back to traditional interfaces for managing or changing their reservation. 

 

The CUI deployed in hotels were direct adaptations of Siri and Amazon echo capabilities. Thus they are already mapped to the obvious ideas we had about the use of voice in the domain: users directly control room and entertainment settings.

Feature Comparision Matrix

02 User Research

For this part of the research, we wanted to find out what are the current pain points in Airbnb housing. We also wanted to find out what people value in such experience so that we can make sure that are not compromising those values in any way. To collect user data, we used the following methods:

1. Interview previous Airbnb guests and hosts

2. Collect feedback from online discussion forums

"Communication with the host is hard when I'm traveling overseas, especially when they don't speak English."

 

"I got locked out once because my host forgot to tell me that the main entrance automatically locks after midnight."

 

"I chose Airbnb instead of hotels because I always like to meet new people."

"It's hard to match the host's schedule sometimes."

“I don't know what to do when my host is not responding to messages or calls.”

 

"I was very hesitant to approach my host with questions."​

 

"I wish the host is there more often so I can get to know more about the local culture, restaurants that I should try out and stuff like that."

03 Synthesis + Opportunities 

1. The use of conversational user interfaces in the context of a setting which is a combination of a “home” and a “hotel” (ex. Airbnb) is still largely unexplored.

 

2. Value in creating a more connected experience during the stay using strengths of a conversational interface (ex. hands-free interaction, system omnipresence, etc.). 

 

3. CUI as the middle person that facilitate communication between the guest and the hosts without taking away the human-human interaction that many Airbnb guests value.

 

 

Ideation

We brainstormed 20 scenarios with a technique called round robin. Because our research identified that host unavailability is a major concern for both Airbnb guests and hosts. We specifically explored three ways where CUI can benefit Airbnb users under such situation.  

 

 

01 Airbnb concierge

Offer personalized hotel functionalities to people staying in Airbnb through CUI

 

02 Airbnb walkthrough in the absence of the host

Allow hosts to create a house tour, with facility instructions. Questions usually directed to hosts upon reception can now be answered by CUI agent.

 

03 Host personal assistant

Help hosts in renting out the place, manage the listings and take care of maintenance tasks.

 

 

System Persona

Why the name Bélo?​

 

Bélo is the name of Airbnb’s logo, a symbol that represents the company’s desire for travelers to have a sense of belonging anywhere in the world. This message resonated strongly with us, so we extended this concept into a personification in the form of a voice assistant. And hence, the name Bélo.

 

Key Personality Traits 

warm and welcoming

Patient 

Friendly 

Knowledge about local experiences 

Personification Through Voice 

Young female 

Calm and soothing

Friendly tone

Conversation Tree

Unavailable Host

James arrives at his Airbnb late at night. His fight was delayed, and he’s worried that his host will be unavailable for check-in. As he nears the apartment, Bélo, the Airbnb voice assistant, senses his presence, reassures him of the directions to the apartment, greets him in the absence of the host and answers to his questions to the best of its capabilities.

 

Experience Prototyping 

Experience Prototyping

To understand whether our scenario and our conversation design create an integrated experience, we asked our colleagues to enact our scenarios. Experience Prototyping helped us in pointing out awkward silences, abrupt conversation transitions and the importance of correct voice tone that a CUI should use.

 

Moreover, it also helped us realize the importance of creating a natural shift between changing points of attention. For example, in some situations, is it more intuitive for the user to talk at Bélo directly than talking to the void.

 

Conversation Modeling 

We created a conversational model to outline in detail how Bélo would interact with the guests. The model showcases Bélo’s pre-attentive and attentive states at various points in the conversation. Bélo is mindful of the environmental triggers and the guests’ utterances and uses them as a prompt to either initiate a conversation with the guests or respond to their queries. This model also showcases Bélo’s robust error recovery feature.

Dialog Flow 

Takeaway 

This project gave me a lot of insights into the emerging field of voice interfaces. Although no dominant design pattern has been developed for this area yet, there are things that a designer should keep in mind.

 

The most important one is probably: How to balance the system's generalizability and context specificity. Under different circumstances, users will prefer the CUI to response in distinct ways. For example, when would a fully context-aware utterance benefit the user without creeping them out, and when would a more general response add value to the user's experience? These are all important question that designers should ask themselves.

Final Scenario: