We’ve teamed up with Amazon Alexa to advance convenient access to health care. Through new HIPAA-compliant skills, users will be able to set up health care visits and manage medical appointments using their voice on any Alexa-enabled device. This is a big step towards accessibility in health care. Read more about the launch in The Wall Street Journal, and TechCrunch.
We sat down with two Providence St. Joseph Health caregivers and talked about their experience. Dean Guo is a Senior TPM focused on Amazon Alexa and chat bots. Timofei Bolshakov is a Senior Software Engineer II on the engineering efficiency team, building on DIG’s AI & NLP initiatives.
Congrats on the launch! What was a memorable moment from the project?
TB: My favorite part was realizing I could reuse the framework that was used for Grace, my chat bot. This significantly shortened development timeframe and also proved that the framework for Grace is flexible enough to use in many different projects.
DG: There were so many phases for this to come to life, including design, development, and testing and certification phases. One example of collaboration and learning was around the use of the Amazon time slot type, which can help convert many statements into time such as “nine in the morning” and “four in the afternoon”. However, there was a constraint against our business cases in assuming A.M. when no specific time of day was given by the user. So if a user requested a 7:00 appointment they would be routed to a 7 AM appointment. However, our Express Care clinics don’t open until 8 AM! Instead of asking a user to clarify morning or evening, we defaulted to “7 PM” with a non-standard implementation.
What was a challenge in making this skill?
DG: Designing for a voice user interface is a new way of thinking about development. Sometimes clinics will have a lot of available appointments that can traditionally be listed, but we couldn’t have Alexa reading off a list of 20 timeslots. So, we analyzed existing appointment data and looked at patterns how appointments were made. We found most of the appointments were made within 2 hours (48%) with the second most popular at the end of the day. Using this information, we made the decision to present the soonest available timeslots with the options of later today and later tomorrow. A typical opening prompt is:
“Your closest clinic is Downtown Seattle, on 222 Pike St. There are appointments at: 10 AM, 10:20 AM, 11 AM, later today and later tomorrow. Which would you like?”
TB: Traditionally when someone logs into Health Connect, they have a live token from Auth0 that lasts for 12 hours. This means they’ll log in almost every time they use the app, so their data is secure. We didn’t want to do this for users on Alexa, but we also didn’t want to store their password. We worked around that constraint by asking users to set up a voice code. Ultimately, we were unable to use the same orchestration level as our other services, but prioritized keeping the skill user friendly.
Do you need a health connect account to use this?
DG: Initially yes, the skill only allowed existing Health Connect users to access and use the skill. It didn’t account for new users.
TB: We actually finished early on the main project. Once we realized we had more time, we used it to add in a register process as a part of the account linking process. Now anyone can use the skill!
How did you use data in your appointment design?
TB: When presenting appointment times, I always accounted for travel between the user address and their nearest clinic. This way, they couldn’t book an appointment if they wouldn’t get there in time.
DG: With most apps or services, behavior is the same every time you use it. With this Alexa skill, the behavior is little different every time. Depending on the time of day and how busy a clinic is, the user will hear different options in the morning, in the evening, and at night after the clinics are closed.
What was the process for launching this project on time?
DG: We used the agile approach with a weekly plan combining with daily demos to the product owner, and worked closely with Amazon. Our product owners provided clear scope, timely feedback and support. We also had contributors from different teams to provide necessary reviews, technical guidance, and infrastructure support for end to end skill development and release. Timofei and I worked closely together to design and develop the skill iteratively. I focused on the voice interface flow, script, and testing the skill, Timofei focused on the end to end implementation. Once the skill passed our own internal certifications, we worked with Amazon to make sure it met the standards of the Alexa skill certification.
What’s next for PSJH & Alexa?
Swedish Express Care clinics in the greater Seattle area went live on April 4, 2019. Providence Express Care Clinics in Washington and Oregon are projected to be available in the coming weeks.