DeepScribe

UX Design for Automated Intelligence

Reducing physican workload through AI-powered note taking.

Project Highlight

The TL;DR. Read the full case study, here.

Final Design:

Original

My Design:

My Design

Wireframes:

iOS, Apple Watch

Wireframes:

Chrome Extension

Research:

Persona

Wireframes:

User Journey Map

Case STudy

Reducing physican workload through AI-powered note taking.

My Role:

Product Designer

Key Colaborators:

Kairui Zeng, DeepScribe CTO

My Responsibilities:

UX Researcher and Synthesizer
I consolidated existing research and produced deliverables clearly illustrating insights from it. I also conducted specific design reasearch in the project scope.
UX Designer
Brainstorming, user stories and flows, feature prioritization, wireframing.
UI Designer and Prototyper
UI patterns, interface design, prototype development, and user testing.

Design Brief:

DeepScribe has created an innovative tool help fight burnout in medical professions through technology. The increased complexity of insurance systems has caused doctors to spend large amounts of time outside of work documenting their exams. This can result in 50-60 hour work weeks and skyrocking rates of physician burnout. The team at DeepScribe has developed an AI engine that can pull relevant concepts from a transcript of a doctor/paitent exam and file them correctly into a correct electronic health record (EHR), requring only a quick check and approval by the doctor.

While the product works well, DeepScribe’s user experience needed some help. Specifically, the team needed an interface for navigating a physican’s daily schedule of DeepScribe exams that created trust in the AI system working in the background.

Research and Scope Definition

My first task was to review the existing research DeepScribe had done into their users and familiarize myself with the problem they solve.

I had a long coversation with DeepScribe’s CTO about the genesis of their automated medical notetaking idea, their implementation, successes and failures. He provided me with common user profiles, reviews, and an overview of a physician’s day-to-day interaction with DeepScribe.

Using this raw material, I created a User Journey Map and Persona to illustrate key characteristics of DeepScribe’s users and process. These assets would help me and the rest of their team keep the user at the forefront of new development work.

User Journey Map

User Persona

After familiarizing myself with DeepScribe’s users and problem space the CTO and I defined the scope of the project and the most pressing problem that needed addressing. Users of the DeepScribe platform sometimes experienced anxiety during complex patient exams and would sometimes refrain from using the product and take notes manually for fear that they wouldn’t be recorded properly.

To gain more insight into this experience and the motivations for it, I designed a survey to distribute to existing DeepScribe physicians that would investigate their feelings about the efficacy of the product.

The primary takeaways from the survey were as follows: 

  • DeepScribe sometimes doesn’t catch data for complex paitents and can file it into the wrong places, requiring more effort to review and edit notes at the end of the day.
  • Physicans often don’t know if a recording session has started or ended, creating confusion and anxiety.
  • Physicans would like to know ahead of time if a particular exam will require a detailed note review so they can do so sooner rather than later.
With this new insight, we defined the project scope using “How Might We..” questions. 

  • How might we increase physician confidence in DeepScribe’s recording of their notes during an exam?
  • How might we let physicians know where possible problems with the recording are?

Feature Prioritization and MVP

A crucial piece of inspriation during this phase of the project was Googles People+AI Guidebook, particularly the section on Explainability and Trust. From this resource, I learned the value of strategic transparency for UX design in AI-powered products. Giving users a peek into the statistical nature of the AI system and not treating it like magic helps them feel more comfortable with it and overcome problems more quickly.
A crucial piece of inspriation during this phase of the project was Google’s People+AI Guidebook, particularly the section on Explainability and Trust. From this resource, I learned the value of strategic transparency for UX design in AI-powered products. Giving users a peek into the statistical nature of the AI system and not treating it like magic helps them feel more comfortable with it and overcome problems more quickly.

Feature Concepts

Prioritization

Not all of these features could be included in a coherent Minimum Viable Product. I used the RICE method to narrow down the list and choose what to work on first.

Using the RICE method, I organized the various features by priority. Communicating with the development team to determine how difficult each feature would be to implement.

The RICE method prioritizes feature by assigning them as score based on this formula:

  • RICE Score= (Reach x Impact x Confidence) / Effort

Where,

  • Reach is the number of customers affected by the feature in a given time period. We’ll rate this on a 1-3 scale with 3 being the entire customer base and 1 being a small number of customers.
  • Impact is the resulting increase in trust and confidence in DeepScribe caused by a physician using the feature. Measured on a 1-3 scale.
  • Confidence is the level of surety associated with the impact rating. Measured as a percentage with 100% being high confidence and 50% or lower being a total moonshot.
  • Effort is the amount of time and resources associated with shipping the feature. Measured in person-months.
For example, when I evaluated the feature “Text bubbles when each person is talking.”, I assigned it the following values:

Reach: 3

The feature would impact all users.

Impact: 2

The feature would be moderately useful for increasing trust in the system since it would be responding directly to users.

Confidence: 80%

I’m fairly sure it would have this effect based on the results of my initial survey.

Effort: 2 person-months

Developers indicated that this feature would be difficult to build and require a lot of engineering resources.

These values resulted in a RICE score of 2.4, putting it low on the list of priorities. 

The method helps cut through subjectivity through quantification and identifies features with high value and low cost that should be worked on first.

 

MVP FEATURES

  • Clear start/stop messaging for exam sessions.
  • A survey to capture system failings and negetive expereinces.
  • A summary of basic paitent information to be delivered soon after an exam’s conclusion.
  • Live audio visualization of a conversation when recording.
  • A marker to indicate when an exam was complext and the note is likely in need of review.

Initial Designs

With a clear list of high-value and realistic features established, I set to sketching out interface ideas. DeepScribe wants to develop a Chrome Extension,  an iOS App, and an Apple Watch App for their product, so I designed with all three in mind.

Wireframes

Prototyping and User Testing

After gathering feedback on the wireframes, I built a mockup set of the iOS app. We decided to start with this platform since most of DeepScribe’s users were already using their existing app.

I used their existing design as a basis and made improvements from there.

Before

After

Before

After

I built a prototype from these mockups that was distributed to several doctors for feedback.

Reception was positive but there were several suggestions for improvement.

Final Designs

Card-based platform to personalize paitents and a physician’s schedule.
Faster, more intuitive room selection.
Brand-conscious audio visualizaiton.
Simplified exam flow.
Seperate note review interface with confidence rating.
Get In Touch

Let’s Work Together!

Email

brattonra@gmail.com

Homebase

Charlotte, NC

Phone

(704) 661-5565

Email

brattonra@gmail.com

Phone

(704) 661-5565

Homebase

Charlotte, NC

9 + 8 =