Telstra AVA
Voice Interface Design
Brief
How might we make the users’ process of interacting with the Google Home Assistant for Telstra service more concise and effortless?
Role
Researcher / UI Designer
Year
2019
Tools
Illustrator
Adobe XD
Photoshop
Premiere Pro
Dialogflow
PROCESS
1. Research
Undertook secondary research on the market of voice technology.
2. Exploration
Used a range of methodologies came up with some initials ideas.
3. Design
Iterated on an prototype voice interface and set up process.
4. Test
Took participants through a range of scenarios to test the prototype.
5. Outcome
Produced a range of insights and recommendations to implement.
BACKGROUND
Telstra AVA Voice Assistant for Google Home aims to assist customers with any queries they might have regarding their account, such as billing information and data usage. The aim of this project was to keep the interactions between AVA and the user as simple as possible
This was a 12 week industry project with RMIT University and Telstra, where myself and two of my classmates from my digital media course teamed up to answer the brief given. I was responsible for the research and usability aspects of the project, as well as creating the Adobe XD prototype.
RESEARCH
We started the project by conducting secondary research and user interviews to establish the views of the market and the current state of voice technology. As no one on the team had worked on a voice interface project before, it was important to initially understand what voice technology is, how it is currently being used and the benefits and concerns of the technology.
Secondary Research
Benefits of Voice Assistants:
Assistance for visually or mobility impaired
Automating peoples’ lives
AVA's ability to remember (e.g. user’s shopping habits, can simply re-order items with a voice command).
Hand free control
Gives users the freedom to do multiple things at a time.
Concerns of Voice Assistants:
Privacy
There is a misconception that devices are always listening and recording what is being said. Smart speakers only start recording once their wake word is said.
Accuracy
VA’s do not always understand what the user is asking for, which may be due to the way we speak or sometimes it is simply that the command has not yet been learnt.
Security
61% of Australians being concerned about their private information being exposed online.
User Interview
User interviews were conducted to gain a deeper understanding of current users’ thoughts and views on voice technology. The purpose of the interviews was to obtain two types of information based around mobile phone accounts and the use of voice assistants (VA) such as Siri or Google voice assistant.
Affinity Map
Key Findings
Phone Accounts
Participants currently look up account information on the Telstra app or website.
Look up data usage and billing information.
Some participants stated that they do not bother checking their billing and data information and others usually rely on the company to notify them when bills are due.
Voice Assistants
Positives - Quick, convenient for simple tasks (making phone calls or instant information on weather) and hands free.
Concerns - Security, privacy, socially awkward to use in public spaces, inaccuracy of information VA gives due to accents or mispronunciations and the VA's lack of flow.
Despite concerns, participants would use VA technology if it were more “socially acceptable”.
ANALYSE
Using the information from the research stage we were able to create storyboards and user journey maps and user flows to better help us understand the problem. This is so that we could work towards creating solutions to the concerns highlighted and also capitalise on the positives.
Ideal Target Audience
Middle aged Telstra customers (45-55)
Own a Google Home device and use it actively
Tech-savvy professionals
Busy and time poor
Have a demanding job and lots of responsibilities
Want answers in a quick and convenient way
Storyboards
Journey Map - Data Usage
User Flow - Verbal Notifications
Key Findings
Found that the three main things that need to be improved or implemented are;
1.
Changing the name of ‘Telstra Voice Assistant’ to be more concise.
2.
Adding a four-digit verification code for security purposes, which will also eliminate the lengthy process of stating and remembering long account numbers.
3.
Adding verbal notifications on prompt to automatically remind customers when bills are due.
NEW NAME, WHO DIS?
A problem highlighted in the client brief was to find an easier way of engaging with the voice assistant. Currently the phrase required for engagement is “Hey Google, talk to Telstra Voice Assistant”. This needs to be replaced with a shorter more captivating phrase that better resonates with users. Brainstorming and dot voting was used to identify a new name.
Alongside developing a new name, it was also important to develop a persona to match this. Creating a persona makes the voice assistant distinguishable from others and provides an emotional connection and response to the brand. Without the use of visual cues, this is achieved through certain language, phasing, timbre, pace and style.
Brainstorm
AVA's Description
“AVA is a teacher who is an expert in Information Technology (Telstra services). AVA responds instantly and is always available to help out with any problems, simple or complicated. AVA is calm and relaxed when addressing students, no matter how irritable or agitated they may be.”
AVA's Qualities
Reliable | Approachable | Informative | Friendly | Calm
SETUP PROCESS
The setup process involves linking the user’s Telstra ID account to their Google Assistant and making a four-digit code to be used when engaging with AVA.
Research into security methods used in current technology indicated that four-digit codes were commonly used. This code is convention for many banking apps, allowing for quick access to private confidential banking information.
Converting this idea into voice technology would reduce users’ concerns about security whilst alleviating the need for saying a cumbersome and long account number to access account information. The goal of the setup process is to make people confident in using this new and unfamiliar technology.
High Fidelity Wireframes
AVA PROTOTYPE
The main query Telstra customers had about their account was billing information. In creating the prototype, the focus was to develop questions around billing information. However, for future iterations, the customer would be able to ask more questions about their account.
Dialogflow was used to prototype the voice assistant. Dialogflow is a Google backed, web-based software that helps create artificial intelligence products. We created a sample conversation flow that we transferred into the Dialogflow software.
Sample Conversation Flow
Dialogflow Prototype
TESTING
5
participants (Telstra customers)
40-60
age range of participants
1
hour sessions
Methodology
-
Asked background questions to gather demographic data.
-
Used the prototype developed in Adobe XD, tested the setup process of AVA on a phone.
-
Used Telstra AVA Dialogflow prototype on a Google Home device to complete six scenarios. Discussion questions were asked at the completion of all scenarios. Metrics were recorded and notes were made of any general observations.
-
Ended with participants completing a System Usability Scale (SUS) Questionnaire.
Objectives
-
Observe users speech patterns (how they asked questions).
-
Navigation flow of the users conversation (how easy or difficult it was to converse with AVA)
-
Determine users satisfaction and thoughts on engaging with AVA.
Tasks
Setup Process
Participants were given a Telstra ID account and 4-digit code and asked to go through the process of linking their Google Assistant with their account. Testing was done on a phone using the Adobe XD prototype. Participants were asked to speak through what they were doing, thinking and feeling during the process.
Scenarios
Participants were asked to engage with Telstra AVA using a Google Home device. They were asked to complete six scenarios.
-
See how much your bill is for this month
-
See when your bill is due
-
Set a reminder for 6th June 7pm
-
See what is on your bill
-
See how much your bill was for May
-
Check notifications
Results
Key Findings
“Liked the amount of detail given, it did not feel too overwhelming,”
“It is practical and useful to access information hands free.”
The Good
-
Most participants could complete scenarios without any critical errors
-
Universally participants commented that setting reminders to pay bills is useful
The Bad
-
AVA not understanding commands – either due to the phasing of the question or not being able to understand a participant’s accent
-
Inaccurate responses to some of the account information given
RECOMMENDATIONS
AVA misunderstood how participants phrased questions. Resulting in AVA providing responses that were incorrect or not given at all due to the different ways that each participant worded their questions.
Recommendation
Adding variations of how questions can be asked to give the right responses.
Finding 2
Three out of five participants encountered the issue of AVA not understanding “7pm” as a time when setting reminders. The prototype would only respond to a time if they said, “Set for 7pm”.
Recommendation
Adding variations to how to ask to set reminders.
Finding 3
When AVA did not understand what the participants were saying, it asked them to repeat the question two times before shutting down communication. This left participants feeling frustrated.
Recommendation
Developing the ability for AVA to identify the keywords of a question and then give suggestions on how to phrase the question better so that the conversation can continue smoothly.
Finding 4
What is on my bill? – Participant 4 mentioned that when asking about their bill break down they were not given information about how much their mobile plan was for the month.
Recommendation
Adding mobile plan cost to billing information.
Finding 5
When is bill due? – In one of the testings of this question, AVA responded with when a reminder was set. A further investigation into why this error occurred, found that this was an issue in how Dialogflow was programmed.
Recommendation
Separating questions relating to bill due dates and setting reminders to get the appropriate response for these questions.
Finding 6
Previous billing information– It was highlighted by participants that the phrasing of the information could be improved to more accurately communicate the information.
Recommendation
Giving the amount for the previous month, then comparing the difference to this month. Include the ability to get a breakdown of the reason why the difference occurred.
Finding 1
OUTCOME
The final outcome of the project was the prototype and a case study documenting all our findings. If we were to continue this project we would have implemented the recommendations into the prototype then test again until a minimal viable product is established.
REFLECTION
Working on this project was an enriching experience, as it broadened my knowledge of new technologies in areas where I had no previous experience. It enhanced my ability to work harmoniously in a team with defined roles. Although these roles were well-defined, we did allow room for cross-over responsibilities when one role was more prevalent in certain parts of the project.
Key Takeaways
Research is key
With the team not being familiar with the market or the field of voice technology, it was extremely important to collect accurate research that would help us conduct this project.
Defining an audience can be hard
It was difficult for the team to find a clear target audience for this product. This was the case as Telstra customers are a diverse group and voice technology is still in its infancy in Australia.
Presenting user testing data
We found that the best way to present this data in a clear and concise way was to create graphs and tables for testing results and draw diagrams to show findings and recommendations.