Thanks for checking out my work. Below, you’ll find video prototypes of speculative designs exploring the possibilities for conversational speech in the medical world.
I’m always looking to improve and to explore new, more conversational approaches to designing voice and chat interactions. The examples below are hypothetical use cases I’ve used to explore design options taking a personalized and pro-active approach to solving users’ problems.
Inhabiting an in-home voice assistant or mobile device, the Evergreen Appointment Assistant can assist patients with setting and managing their medical appointments. The assistant knows the user and their providers, and tailors the conversation to meet the individual’s needs.
Stephanie has an upcoming doctor’s appointment and needs to re-schedule.
Here, I explore a conversation where the assistant knows the user has an appointment coming up and proactively asks if that’s why they’re engaging the system. I chose to allow either a straightforward “yes/no” response or, as shown in the video, a more complex response that includes the action the user wants to perform. I like this task- or goal-based approach because you can solve for a large percentage of interactions and avoid making the user choose from a list of options. And, I feel this is more like what a human agent might ask.
There’s also a small-but-important prompt writing decision in here, where I chose to leave off the second instance of “AM” when listing out three times of day (“I have 10 am, 11:30, and 1:30 pm”). It doesn’t add much value to say “11:30 am” in that context.
In this example, Stephanie wants to change the day of the appointment she just set with the assistant, so the assistant must manage a complex dialog based on the doctor’s availability and offering different days for Stephanie to choose from.
An interesting design challenge here is tracking what the user means by “next available.”
In this case, our user has multiple upcoming appointment, so the assistant needs to ask if they want to manage one of those existing appointments. I experimented with starting by asking the user if they want to make a new appointment, but decided that it’s more likely a user with multiple appointments will want to change or cancel one rather than make a new one. I’d want to do some quick guerilla research to confirm that assumption, plus look at usage logs later on.