Inside the Release: Amazon Alexa Skill

Small and Fast

Kingfishers are astonishing to watch. Small and fast — they get their food faster than a heron, defter than osprey. They will speed down a river and stop on a dime, hovering until they dart into the water to pick out a fish to take back to their nest. Our team is like that. Small, fast and built for purpose. We were going to implement our most commonly used feature — scheduling a same day visit in a retail clinic — using Alexa. This was part of the new Alexa Healthcare Skills program and it was a whole new world.

Our team started with three people; a TPM, software engineer and a product manager. The first thing we did was agree to keep our processes as simple and streamlined as possible. We were going down this new path with voice technology, so we took out as much friction as we could without sacrificing quality. We’d have daily stand-up meetings test the skill every day and worked closely with Amazon to deliver the best experience for customers.

Ecosystem

The ecosystem that a kingfisher lives in is a big part of how it’s able to be so effective. The same thing is true of our team. All of us were able to leverage enormous amounts of work. Need an API to the EMR? Got one! Need a bot framework? Got one! Need a custom time library? Got one! To top it all off we knew this specific Alexa skill was technically feasible because of the work we did at an internal hackathon to test the concept several years before. We were just waiting for a HIPAA Business Associate Agreement (BAA) to make our skill public. Our little team started down the river with a ton of confidence and talent.

Learning Voice

We started out hyper focused on how the voice interface impacts the user, and how we could deliver the best experience for customers in different situations.

Voice is a different user workflow. Imagine displaying a list of appointments that are available for scheduling. Showing 20 appointments and their availability visually is common. The human eye can scan a screen and be drawn to the right information quickly. With voice that information must be spoken to the user. Imagine calling a receptionist to make an appointment. You are usually able to get to two or three options quickly because both people can understand the others intent.

In a visual experience, it’s usually easy to show the user what step they are on in the process. Breadcrumbs, titles and even the different fields being presented give the user common visual cues to what they are supposed to do.

But voice is different. Imagine every turn a conversation can take. How would the skill respond if someone was confused? The response would be different for each part of the conversation. We wrote and tested dozens of such responses. It’s important that the most relevant information was the first to be surfaced through the skill.

We adopted best practices like writing for the ear, not the eye, presenting information in consumable pieces. The team spent a week looking at internal testing data to effectively offer appointment options. We decided only to explicitly offer a few options at a time. For times that had broader availability — like “tomorrow’s schedule” — Alexa would ask for the user’s time preference.

Going Faster

The team had a deadline to meet so we tested and iterated from day one by building simulations of the interactions before the actual programming was complete. We tested different utterances while developers were coding the skill to the APIs, building out the account linking and configuring infrastructure that would make that experience work.

Every morning in our stand-up meetings, we did a demo. In the early days, we would demo early iterations of our simulations. That helped our small team stay connected so that when the infrastructure and voice UX were ready nobody was surprised and bringing it all together was smoother.

Eventually, we were able to test the skill with a group of users. This step was critical to tell us if what we’d hatched would work.

Certification

Just as kingfishers work in pairs, we did so with Amazon. They provided much needed support in checking our voice interaction and ensuring what we delivered based on their hard-won experience of learning what does and doesn’t work. We stayed in close communication throughout the process of building the skill.

This is Just the Beginning

As a product manager there are few things more satisfying than seeing the work of a great team out in the wild. Now it will fly from the nest and the real work begins as we monitor how it performs, learn and adjust. There will be more to come as voice continues to grow. Stay tuned.