Google Assistant App POC for Globo Esporte
I had the challenge of defining user needs and conversation flows for a Proof of Concept able to access internal structured data and deliver it using the Google Assistant Platform.
About the project
During an internal hackathon, I worked together with developers to propose a POC of a Google Assistant Application to deliver Globo Esporte content. Observing the supported devices and communication solutions that Google Assistant apps could communicate to, and analyzing the Globo.com challenges and strategies, we presented our project propose to receive 20% of our work time during three months to work on this challenge.
And we won it. Our project was selected by Globo.com executive board on 2018 March and we received 20% of our work time during the next three months to deliver a POC able to access internal structured data and deliver it using the Google Assistant Platform.
As the only UX Designer in the group, I was responsible for understanding user scenarios, the information available and defining input/output texts, conversation flows and testing the outputs. I will detail the activities I performed below:
Understanding Use Cases
The first step to guarantee a great experience was to understand the scenarios that would it make sense to use the help from an assistant related to the information we already have available.
We used internal research about the users of Globo Esporte web and mobile platform, their needs, habits and social use of the content we already provided as a platform. This scenarios of use allowed us to understand when and how an Assistant could be helpful. We noticed the needing of checking scores from the day before, before starting a working day or even the interesting in getting news about a football team or about a specific kind of sport as a pastime during the commute.
After understanding these scenarios, we applied a group activity to understand how people use to talk about the subject. This activity, suggested by Google Design Guidelines, consisted of asking a group of individuals to talk about the subject freely and taking notes about how they ask and answer about the subject, how they construct the phrases and so on. It was useful to have ideas about how to construct the dialogs.
Defining the Assistant Persona
To guide the conversation and the Assistant voice tone, it was necessary to think about the Assistant's Persona, the character that would be noticed by the user according to the way the Assistant answers the questions.
A very interesting study about brands voice conducted to TechTudo, a news portal about technology from Globo.com, gave me the insight of using the Jung's archetypes, described at the book The Hero and the Outlaw: Building Extraordinary Brands Through the Power of Archetypes from Margaret Mark and Carol S. Pearson. Based on this, I selected 4 archetypes that could match our brand and the news already produced, and with the team, we choose two: Caregiver and Everyman.
Creating the Intents and Entities
A key point of designing for a specific technology is understanding how it works, so you can better communicate with the tech team and guarantee an output that makes sense for the final user. And the Intents and Entities from Google Assistant are the base for understanding how this technology behaves.
Intents are about the user intention, they are about what kind of information the user is expecting when builds a phrase or use an expression. For each Intention you expect from your user when talking with your Assistant, you provide a list of expressions and phrases that reflect the user's intention.
The Entities are about related subjects, that may change and, in our case, represented parameters from our Information Database, such as all the kind of Sports Globo Esporte covers or Soccer teams, that are very popular in Brazil.
Documenting specific logic flows
In some specific scenarios, it was necessary to make the intelligence and logic inside our code, since it depended on information and logic that was not supported by Google Assistant. In order to check the flow and provide documentation to the development team, I designed the flows to make it easy to be understood and checked.
Training the intelligence
After making internal tests with volunteers, it was already possible to train the Assistant and check if it is understanding the user's intention (what means it is finding the correct Intent for the user's input). The training feature is especially important for UX Designers since it allows us to see the kind of inputs real users are providing, what give us fast feedback about how people construct the phrases, what kind of information they are expecting from our Assistant, what is not clear for them, etc.
After the 3 months working part-time with development team and performing internal tests, a first version were available to be released public, as can be seen in the video below.
This first version was able to: welcome and farewell, explain what kind of information the Assistant could provide, handle errors and unexpected messages, sign in user inside our platform, deliver general and specific (filtered by sports or soccer team) news from Globo Esporte, save user soccer team, deliver soccer team position on the championship and when a football team would play in the future.