Conversational Technologies
5 min read

How We Adapted Our Virtual Assistants At Warp Speed

Rotem Fisch Head of Design, Hyro
How We Adapted Our Virtual Assistants At Warp Speed

It’s been a little over a month since the World Health Organization (WHO) — concerned both by the alarming levels of spread and the severity of infection — declared COVID-19 a worldwide pandemic. On January 21st, 2020, the first confirmed case in the United States was detected in Washington State.

 

At the time of writing, Americans comprise a third of the world’s coronavirus patients.

Source: Worldometers

To accommodate this enormous pressure, the country’s healthcare ecosystem had to transform itself in a matter of days, playing a national game of catch up with a disease that was circulating at an unprecedented rate.

At Hyro, we had to adapt to this new situation just as quickly. Through our existing virtual assistants, we soon witnessed the incredible need patients had for care, information, and clarity regarding COVID-19:

Frequent Keywords Via Hyro’s Virtual Assistants — March 2020

As our healthcare clients worked on all fronts to serve their hundreds of thousands of patients, we reexamined, rethought, and reshaped our virtual assistants to help relieve the sudden, and overwhelming stress on their support centers.

A changing use case

Up until recently (and in many ways to this day), most patients used our existing virtual assistants to find physicians and schedule appointments using natural language. Through iterative design, we improved and expanded our platform’s scope over time to the point where patients could filter search results by varying attributes such as specialties, location, languages, and gender.

 

Gradually, as the virus made landfall in the U.S., we noticed a remarkable shift in the manner in which patients communicate with our virtual assistants. Increasingly, more and more user conversations revolved around COVID-19. Users began asking a wide array of coronavirus-related questions. One of our very first lessons was that patients’ concerns were hardly limited to the virus itself but to the many ways in which it has affected or disrupted their care.

A Sample Of User Queries Revolving Around Covid-19 Directly And/Or Indirectly

By the time 20% of all user-queries could be tracked back to the virus, we added a “Coronavirus updates” button, which directed users to our clients’ newly erected COVID-19 information web pages. Drawing upon these critical usage-insights, we were also able to identify that many patients were confused as to how the crisis affected their healthcare provider’s online scheduling policy, which prompted us to add a “How do I schedule an appointment” option to the conversation-flow.

“Coronavirus Updates” And “How Do I Schedule An Appointment” Options Added

Taking a closer look

The “Coronavirus updates” button proved highly effective and received the lion’s share of clicks and requests during initial COVID-19 related conversations, but it left us with a blind spot. Though we knew we were now offering our clients’ patients with a service they very much needed, the intent behind these clicks remained ambiguous. What exactly did they want to know?

Based On A Sample Size Of 100 Conversations, March 19–21

To address this issue we turned yet again to the data we’ve collected from user conversations and extrapolated the main patient pain points to create a new set of options. One of the key features we added was a coronavirus self-assessment tool which asks users a series of questions to help them better assess whether they may have contracted COVID-19, based on recommendations we scraped from the CDC’s website.

These additions were also extremely well-received and immediately provided us and our clients with an understanding of the services patients needed the most. But there was still one more challenge we needed to solve. The patients using our virtual assistants have become accustomed to being able to type or speak freely in their natural language. This indirectly encouraged fragmented sentence structures, disambiguates, slang, and even spelling mistakes. To tackle this issue, we used the wealth of knowledge we’ve amassed to allocate specific trigger words that when uttered by a user would direct them to the right page or service.

The way forward

What started as a sprint is now becoming a marathon. The COVID-19 crisis is very much ongoing and every day brings with it new questions and problems to be solved. We continue to use iterative design to address dynamic, real-time issues and find ourselves (like the rest of the world) on a continuous learning curve. Each step we take to better serve our clients and their users throughout this pandemic is another capability we might have never conceived otherwise. This experience has encouraged us to release a complimentary COVID-19 virtual assistant geared directly towards queries surrounding coronavirus, which has been deployed by some of the nation’s leading health organizations such as Montefiore, Mercy, and Fisher Titus since launching on March 18th.

As more healthcare organizations continue to deploy our assistants on their websites, we’re already hard at work on “phase two” of our COVID-19 relief effort.

 

As of March 23rd, the average waiting time in crisis hotlines has more than doubled, leaving patients unanswered and operators exhausted.

 

We’re currently in the final stages of development on a COVID-19 AI call-assistant that will be able to automate and navigate these patient calls, triage symptoms, and answer coronavirus related questions.

We’re looking forward to what will surely be another source of fascinating insights.

 

To learn more about our COVID-19 Virtual Assistant, visit our website and follow along on Linkedin and Twitter for updates. Want to talk more about iterative design? Shoot me an email at [email protected].

Stay safe and healthy, everyone!

About the author
Rotem Fisch Head of Design, Hyro

Rotem is Hyro’s Head of Design & UX and an industry maverick with more than ten years of experience leading, devising, and crafting graphic and product design for some of the world’s top tech unicorns (think IronSource and HiBob). Although she teaches interaction design at the Bezalel Academy of Art & Design, Rotem probably struggles more than most when interacting with microwaves and remote controls.