AI-Driven Health Search · Roon

IZEL MARAS
Information

Resume
Linkedin
Read.cv

Lead Product Designer
Chime
2024 - Current

Head of Design
Roon
2021 - 2024

Lead Designer
Verizon
2016 - 2020

Visual Designer
frog Design
2014 - 2016

AIGA Mentor

Tech Crit Advisor at
Cornell Tech

Brooklyn

AI-Driven Health Search

What

Help users found medically vetted answers to their questions

Personas

Armchair Arman

Reads and learns at own pace
Extreme attention to detail
Aims to know as much as their doctor
Creates documents
Tries to understand research papers

social sandy

Navigates with others
Values community over information
Joins FB groups
Follows health influencers
Active on social media and Reddit

pay-it forward pen

Advocate
Experienced a challenging time
Has hard-won insights
Wants to help others

remote rajiv

Lacks access to quality care
Worried about health
Seeks information

Diagnosed Denise

Recently diagnosed
Overwhelmed by learning
Doesn't know where to start
Feels alone, sad, and anxious
Struggles with emotions

chronic carl

Long-term condition
Knows basics of management
Seeks improvement in quality of life
Doesn't want condition to define him

navigation nancy

Caregiver for ALS patient
Reads extensively about condition
Manages appointments
Stressed by caregiver role
Financially distressed
Puts own needs second

User Feedback & Metrics

On a weekly basis, 11% of active users perform a search.
The search exit rate shows how users use search to find videos. Search exit rate has been on decline by 20% each month, indicating users can’t find videos with search.Number of active users dropped 44% in the past 4 weeks, indicating, people can’t find the content they need, and exit the app.Users are confused around where to ask and where to search. “I thought that was ask” “I’d go to here to search [showing ask bar]”

Problems to Solve

Product Issues

On a weekly basis, 11% of active users perform a search.

The search exit rate shows how users use search to find videos. Search exit rate has been on decline by 20% each month, indicating users can’t find videos with search.
Number of active users dropped 44% in the past 4 weeks, indicating, people can’t find the content they need, and exit the app.
When we asked users what they needed the most, they explained they need in 30 character or less.
Users are confused around where to ask and where to search. “I thought that was ask” “I’d go to here to search [showing ask bar]”

Usability Issues

Medically vetted content on Roon is hard to find through the current navigation.
Users can’t locate the search bar.
Users confuse the ask and search functions.
Users can’t find the Search Landing Page (SLP).
Many users are unaware of the different types of content available.
There's no way to search for experts, topics, or subtopics.
Many users can’t find the 'Ask an Expert' feature.

Goals

Reduce search drop off rates by 20%
As measured by conversion of search page view to search result press

Increase search engagement by 20%
As measured by increase in unique weekly searches by active users

Reduce ‘Doctor Ask’,

As measured by reduced questions submitted to Roon.

Integrate AI to aid users find content
As measured by video watches and use of AI-powered feature

Ideation & Validation Process

Phase 1: Problem Identification

Understand the Problem
- Define context and constraints.
- Study the audience and gather qualitative and quantitative data.
- Review similar issues, companies, and features.
- Gather team input and align on the problem and audience.
- Draft and refine a user story map.
- Discuss engineering constraints, prioritize, and define the MVP and long-term goals.
- Determined the hypothesis, scope, and schedule for the project.

Phase 2: Idea Generation

Brainstorm & Concept Development

Run a Crazy 8s session to gather ideas from all teams.
Create and refine concepts.
Develop early responses through concept sketching and wire-framing.
Align with stakeholders on what to test

Phase 3: Prototyping & Testing

Create & Test Prototypes

Build workflows and create high-fidelity prototypes.
Prototype interactions to test usability.
Develop a test script and conduct user tests.
Synthesize results, update the user story map, and iterate based on feedback.
Conduct additional tests and make iterations based on feedback.
Refine the design through repeated testing and feedback until the audience responds well to the proposed experience.

Phase 4: Delivery & Improvement

Finalize & Review
- Review the prototype with engineering to finalize requirements.
- Conduct additional tests and make necessary iterations based on feedback.
- Create and finalize the high-fidelity prototype with the design system.
- Q&A
Post-Delivery Analysis
- Conduct a full analysis of initial results post-shipment.
- Interpret results, discuss with stakeholders, and structure additional tests.
- Continuously improve the design based on feedback and analytics.
- Begin the investigation phase of the product experience again, continuously improving the design.

Solution A

Improve Search answers by using AI to serve reliable and accurate content to user queries.

Fine tuned Chat GPT with our question and answer library transcripts, and created a prompt that ensures all the answers are generated only from Roon’s medically vetted database.
Created an interface that allows users to ask a question to Roon and get answers with sources to our proprietary medically vetted video content
Added an entry point from home
Added an entry point from search

Before

User Problem
Roon has medically vetted content, but they are hard to find from the current navigation

User Problem
Users can ask doctors questions, but wait times for answers are too long.

User Problem
Some users don’t even know where to start, what to ask, need to be guided.

User Problem
Users want trustworthy answers to their medical questions.

Solution B

Make Search more accessible from high-traffic pages and improve navigation and query suggestions.

Merged topics tab and search tab and called it explore with search icon.
Made search function accessible from tab bar
Redesigned a tabbed search landing page, where users can get topic, subtopic and expert match

Before

User Problem
The search function is confusing, and difficult
to find.

User Problem
There's no way to search for experts, topics, or subtopics. Many users are unaware of the different types of content available.

Impacts & Results

AI-powered Search is helping users ask meaningful questions and find answers with deep medical expertise.

We observed a 24% increase in users clicking through a question instead of closing the Search page, well above our original goal to reduce Search drop off by 20%

We observed a 9% increase in search page visits by active users.

We observed an 85% decrease in users asking doctors questions, as AI now answers most questions accurately.

80% of AI answers are voted up, indicating users are finding answers they need with the AI-powered ask.

AI-powered ask e has increased by 6.29% since it’s launch on March 2024.

AI-powered ask has been stealing traffic from global search.

Users ask 2-3 questions a day to AI-powered ask.

Lessons Learned

AI can be used to generate questions that users are likely to ask, that can be displayed as “Related Questions” suggestions to make search more efficient.
Design with search and AI computation times in mind, using motion to mask delays.
Clearly communicate AI limitations to users, preparing them for potential hallucinations and misleading answers.
Educate users on how to effectively use AI to achieve the best results, focusing on instructional guidance.