TASK
The Problem:
Due to the stresses in people's lives, especially during a global pandemic, people are finding it harder to sleep.
The Solution:
Design a voice assistant for people to speak to in order to calm down, relax and enjoy better sleep.
My Role:
Conversation Designer, UX Researcher.
Tools:
Whimsical, Voiceflow, Logic Pro X, Amazon Developer Console, Dashbot.
Application:
Alexa Skill
RESEARCH
After identifying the problem, I conducted primary and secondary research to analyze and further understand the problems and pain points that people currently face.
Through the secondary research, I began to understand the wider sleep problem, by gathering some hard data that backed up the project.
"What effect does sleep have on the body?”
By conducting user interviews with a variety of different participants, who have had, or currently have problems sleeping, I developed a deeper understanding of their problems. It allowed me to connect with potential users emotionally, by hearing their struggles and how they were trying to eradicate them.
I then produced a competitive analysis, to identify the current market and the potential gaps in which to position my product.
Research Report
DEFINE
Prioritise Features
Target Audience & Psychographics
By starting with a broad range of features, I narrowed them down and established a project timeline to define and prioritise the most important features. The 4/6 Thinking Hat method, allowed me to holistically visualize and assess each feature with a different set of eyes, identifying any concerns or emotions that could affect the project timeline.
By researching and identifying the target audience through their demographics and psychographics, I gained a deeper understanding of what drives and influences them, ensuring I follow the user-centric approach throughout the design.
HOW?
-
Crafted a user persona from the research (customer profile) to establish the target audience.
-
Created a customer mood board to help visualise the project with the team or stakeholders.
Defining the assistant persona, allowed me to decide on the personality of the voice assistant, and how it would interact with the user. I took some keywords that were mentioned by users themselves in the interviews to establish the bot's personality.
"Trusting, Supportive, Calm, Attentive, and Warming."
I also identified the functions and limitations of the assistant during this phase, to further cement what the assistant could and couldn’t do.
Click images to enlarge
Assistant Persona
DESIGN
Conversational Components
Sample Dialogs
The initial phase of the design process started with the creation of the Conversational Components.
These helped to define the standard vocabulary of the assistant by resonating with the personality and keeping consistency throughout each stage of the design.
- Introductions, confirmations, acknowledgements, error handling and endings.
I then drafted out sample dialogs for each use case and scenario, including first-time users vs returning users, happy path vs error or edge cases. A simplified Wizard-of-Oz method was used to test and improve the copywriting, and to identify other conversational flows.
Click images to enlarge
Flow Diagram
To improve the dialog for speech, I wanted to test how it would sound. Using Google’s online TTS engine to play out the dialog allowed me to alter the copy for a more natural-sounding speech.
The examples below show the very first versions of the welcome message.
-
Intro - Demo Version 1
Welcome to Imagine Sleep Sounds. I can help to relax your
mind and fall asleep by playing breathing exercises and
ambient sounds. Which would you like to hear?
-
Intro - Demo Version 2 - (after listening on TTS engine)
Welcome to Imagine Sleep Sounds. I'm here to help you to
fall asleep and relax your mind. I have a number of breathing
exercises and ambient sounds for you. Which would you like
to hear?
Following that, I produced a flow diagram of the entire conversational flow to understand the overall structure of the design. Only once all the previous steps had been completed, I began the design using Voiceflow prototyping tool.
Imagine Sounds Flow Diagram
Natural Sounds
Earcons
The natural sounds of: Ocean, Rain, Birds and Jungle/Rainforest were chosen, due to my research showing they were popular for relaxing and for improving sleep. I used my sound design skills to create and mix different sounds from my own personal sound library that I had recorded or collected myself.
I used sound designing techniques to create the earcons, aiming to match the calm and warming assistant persona. Various earcons were created with subtle changes within them, designed to showcase the different user journeys whilst keeping consistency throughout the experience.
Earcon 1
Earcon 2
Earcon 3
Google's Conversation Design Guidelines
Earcons, or audio icons, are a type of non-verbal audio (NVA) designed to convey a specific meaning. Your persona can use earcons to communicate with users in the briefest way possible.
TESTING AND ITERATIONS
Usability Testing
Step count
New vs Returning Users
Further Improvements
Once the first version of the design was complete, I conducted usability tests to identify issues and highlight areas of improvement. By structuring the problems and errors using a severity scale ranging from minor to critical, I was able to analyse and recommend the areas that needed urgent attention.
One main user concern was highlighted due to the taken for the user to reach the end goal being too long. This was a difficulty I had throughout this project. I needed to somehow present the required information, in smaller chunks, with time efficiently a priority.
This problem was solved by streamlining the experience. This was done by reducing the number of steps it could take to reach the end goal. Parts of the copywriting style were changed, allowing the user to take more control over the experience and giving them more opportunities to decide on their own path.
Users also showed concern with the overall structure in relation to time efficiency. It became clear that I needed to alter how the responses were structured for new vs returning users.
To solve this I increased the session number every time the user opens the skill, and created separate welcome responses, depending on how many times they used the skill. This allowed for a much more personal experience, where each welcome message contained several variations to keep them varied and less repetitive.
Other noticeable improvements included changing the synthetic voice from a British male voice to a US male voice that sounded more natural and flowed better. I also changed the names of the breathing exercises to ease confusion among the users.
The full usability test report can be viewed in the PDF link below.
Video: Step Reduction
Screenshot Voiceflow: New vs Returning User Variations
Data, Analytics and Looking Ahead
With the design complete and ready for deployment, I implemented Dashbot's API's into the Voiceflow project, at each stage where I wanted to collect data. This data will be analysed periodically, to identify common issues or areas of improvement. Further design iterations and adaptations will be continually processed to ensure the success of the skill.
