Healthdirect VOICE appS
Designing ANOTHER First in Australian Healthcare…
With Google Home launching in Australia in Mid 2017 and Amazon’s Alexa not far behind, Healthdirect Australia were keen to explore this new content channel and the opportunities it could provide to give Australian easier access to more healthcare information wherever and whenever they needed it.
Initial experiments with the platform highlighted that this project would be far from just taking existing content and making it available on the platform, it would require a whole new approach. Starting with research into how people use smart speakers, how to develop Natural Language Processing AI and how to help users find and engage with your content once it has been published.
Who was involved
The core team consisted of a product owner, solutions architect, a tester and myself. Outside of there core team there was an exec sponsor.
As this was a completely new technology stream there was interest all the way to the CEO level and frequent showcases to ensure all stakeholders were brought on the journey and able to give feedback as the product developed.
My ROLE
As Entrepreneur in Residence, I was responsible for developing the initial pitch to the business. This involved a demo of a health search app using voice and a proposed roadmap to explore the technology and build and launch a Minimum Viable Product. The pitch was well received by design, development and tech stakeholders as well as the executive sponsor.
As we developed from the initial pitch to launch my involvement included:
Product scope design and development
Facilitating a series of workshops and user interviews to establish the best use cases for voice
Technical design, evaluation of content systems including Dialogflow, Chat-Fuel, Lex and Google Speech
Development of Natural Language AI
Working with content creators to develop the content strategy for voice both how users phrase queries and how the responses were formulated
Running a series of knowledge sharing sessions across the business to build the conversational design and dev. capability within the company
What we did
Discovery
We first had to discover a baseline for who was using these new channels and what they were using them for. This involved a round of interviews with a range of users of different demographics and technical knowledge. One of the most interesting interviews was with a blind user who described these devices as “independent life tools”, capable of assisting around the house with a huge amount of things and delivering content in a much friendlier manner than interacting with desktop screen readers.
Design
To reduce scope for launch, we identified a series of subjects that people would be likely to ask their device for information on and settled on Summer health topics as they are something people always need to be aware of in Australia. The next step was to role-play what people would ask the device with both clinicians and users. This helped us develop a range of questions and fallbacks including being able to ask for the UV rating in your area and also helped us develop a tone of voice and persona for the assistant.
Finally we ran a series of “Wizard of Oz” user tests where users talk to a device but the responses are provided by an unseen human facilitator with a script of questions they can respond to. This really helped to solidify the personality of the assistant ("a trusted aunt" or "friendly nurse”) and put us in a position to build our first Minimum Viable Product.
Build
In the design stage we had built a technical prototype so a lot of coding and technical questions had been answered. There were still some challenges though, we used Dialogflow for our intents and responses in Google Home and Assistant but Alexa required a different technical stack. Fortunately we discovered the Jovo framework which simplified the code and allowed us to write one set of code for both.
The other challenge was setting up a process for non-technical content writers to write new content, edit existing content and train the Natural Language Processing to understand what the user’s intentions are. This required up-skilling within the company but gave the content writers a new set of skills and ability to write for conversational interfaces.
What we learned and delivered
We launched in Summer 2017 and immediately saw users talking to our assistant. As these responses came through it became obvious that we would have to monitor and train the language model for the assistant as the initial weeks and months went on. This is an aspect of launching an AI based app that many don’t appreciate, the AI model requires training for months after launch.
The result for the business and the team were many-fold, we launched a successful voice assistant, the “What’s the UV rating in…” intent proved especially popular and a good hook to get users asking the assistant more. Summer Health attracted attention in the media and it’s follow up for Cold and Flu Advice was featured on the Alex Skills store.
Try them today, on Google Home or Google Assistant say “Talk to Summer Health” or view the directory listing.
You can install their “Healthdirect Flu Advice” skill on Alexa from the Skills Store. It is also available as a Facebook Messenger bot.