Building a new way to chat with a doctor


GYANT is a San Francisco-based startup combining artificial intelligence with telemedicine to partially replace in-person doctor visits for acute illnesses. You can think of GYANT as a friend to text when you’re feeling unwell.

🤔 How does it work?

Chat with an automated bot to answer questions about your symptoms. If you’re in California, you can then text a licensed medical provider directly in the app. If you’re outside of California, GYANT will compare your symptoms to millions of records in order to give you the best short list of possible diagnoses and recommended next steps. Finally, GYANT will send friendly messages to check-in on you in the days following your original chat to make sure you’re feeling better.

📝 Project details

Between 2015 and 2017, the company launched a Facebook Messenger chatbot that had helped over a million people check for Zika virus in Brazil, run pre-diabetes screens in Europe, and check acute illnesses in North America. But to comply with HIPAA, the company had to pivot from a Messenger bot to building their own native app. The company launched an MVP iOS app in November 2017.

While their first iOS app was a step in the right direction, it left much to be desired. From December 2017 to March 2018, I worked with one iOS Engineer, two back-end engineers, a Data Scientist, and a small team of medical practitioners to lead the end-to-end redesign and development of GYANT’s second iOS app. You can download the new GYANT for iOS in the App Store.

The founders felt strongly about a few things so I viewed them as constraints. (1) the platform would be an iOS app, (2) the solution would be texting; not video, (3) our target demographic would be college-educated and U.S.-based people aged 22–35, and (4) we should take product and design risks in order to break existing messaging app paradigms and differentiate our app.

What we’re shipping — the new GYANT for iOS

Messages from GYANT appear front and center with quick reply buttons to help make your replies faster. The overwhelming majority of your interactions are in the bottom-half of the screen so reachability is a breeze.

In California? Text with a human doctor that can write you a prescription if needed.

View your doctor’s profile to see more about his or her location, background, hobbies, and even fun-facts.

Not in California? No problem. GYANT will compare information about your symptoms to millions of records to get you the most accurate medical advice an AI Assistant can provide.

How we got there — our process of guessing less

From December 2017 to March 2018, I worked with one iOS Engineer, two back-end engineers, a Data Scientist, and a small team of medical practitioners to lead the end-to-end redesign and development of GYANT’s second iOS app. The app shipped in March 2018 (download the new GYANT for iOS in the App Store).

In pursuit of gaining empathy for our users, we learned what they value most

In the early stages of the project, we spoke with over a dozen people in our intended audience to learn about their experience with acute illnesses. What do they do when they first feel sick? What goes into their decision-making when going to the doctor? What would make them have a good experience texting with a doctor? We learned from our users and reflected on a few key takeaways:

1. When researching symptoms, our users value accuracy and approachability. Unsurprisingly, we found out our users don’t search online for their symptoms. The overwhelming majority of people we spoke with didn’t search online for their symptoms because “it always tells me I have a terminal illness” and “I don’t understand a lot of the stuff on those sites.” This showed us the importance of being accurate in our diagnoses and emphasized the importance of explaining medical information in a friendly and approachable way.

2. Pre-doctor visit, our users value convenience and efficiency. We heard story after story of users going to Urgent Care to avoid making appointments; even when that meant paying 3–5x the copay of a Primary Care Physician. This showed us the importance of getting users through efficiently though the app experience. Additionally, we found the steady rise of Urgent Care clinics in the US supports our qualitative feedback.

The rise of Urgent Care clinics agreed with our qualitative feedback. Data: CB Insights.

3. During doctor visits, our users value empathy and compassion. The importance of empathy, compassion, and trusting providers became a large cornerstone of our interviews. We questioned “how do patients and doctors establish trust?” and explored the offices of doctors and surgeons to get insights. We learned they build trust, in-part, by showing professional credentials (education, certificates, etc) and their life outside of work (family, pets, hobbies, etc).

We learned the people problems with our first app centered around a lack of clarity, efficiency, trust, and delight

Our intuition (as a designer and as entrepreneurs) gave us a starting point: our next version of GYANT should be more efficient and more delightful. Without efficiency, the benefits of texting a doctor would never outweigh seeing a doctor in-person or video chatting with a doctor. Without delight, we would never stand out in an increasingly crowded market and get people to change their routines when they begin feeling sick.

We asked users about their experience using GYANT’s first iOS app. When asked “How would you improve GYANT?” our hypothesis on efficiency was supported and we realized the importance of clarity.

Talking with both users and non-users in our target audience highlighted the importance of trust and delight in telemedicine while reinforcing the importance of clarity and efficiency.

But how would we know once we had solved those problems?

We first sought to define our goals in simple terms. A framework Julie Zhou has written about at length in her blog (a couple of my favorite posts here and here) is to ask “what will people will do radically different once your product is successful?” In GYANT’s case, the behavior change we hoped to see is people visiting the doctor in-person less often.

Our overarching goal is to send people to the doctor less often.

Adding this onto our findings from surveys, in-app feedback loops, and interviews, we arrived at our high-level goals.

  • Improve clarity. Feedback from users made it clear our first experience lacked clarity in what next steps were, how medical information was displayed, and how GYANT arrived at the recommendation.
  • Increase efficiency. One of the inherent values of a conversational interface is being able type as if you’re speaking with a human. Just as you can efficiently community with Alexa and Google Assistant, users want to chat with GYANT like they’re talking to a human. But this was lacking in the first iOS version. Additionally, we knew our users prioritize efficiency when making decisions about in-person doctor visits and believed this would carry over into digital doctor visits as well.
  • Build trust. Our conversations with users gave us reason to believe there is a positive relationship between the level of information you know about your provider and the level to which you trust him or her. This hypothesis makes sense intuitively — trust between both our users and GYANT and our users and the provider is important because it increases the likelihood users provide us with accurate health history, share personal information, and stick with us while we ask several questions about their symptoms.
  • Bring delight. The CEO was adamant about representing the GYANT brand and creating an experience that would be surprising, delightful, and memorable. Without delight, he believed we would never be top-of-mind for users when it came to discussing their health. Our goal was to do this by bringing character and empathy to how we animated GYANT as an artificial intelligence.

Establishing metrics for success

Qualitatively, we knew we would notice a positive change in tone and feedback from users. The pain points they previously mentioned would be softer or not mentioned at all and their overall experience would be more delightful — something you can see while you watch as they experience your product or prototype.

Measuring and getting accurate quantitative data at early-stage startups is often tough. This redesign was no exception. I partnered with our Data Scientist to think through and track our success metrics.

  • % of messages to GYANT asking an unsupported and typed question regarding process or design should decrease
  • % users completing signup, symptom check, and diagnosis should increase
  • % users returning to provider chat should increase
  • % of new users visiting the app again within 1 week should increase
  • % users giving 4 or 5 star rating within app should increase
  • Net Promoter Score should increase

Forming Design Principles: the foundation of the experience we want to build for iterations to come

We formed our design principles alongside our goals to guide the the GYANT user experience for iterations to come. Interestingly, the many rounds of user feedback and analysis made our design principles come easily.

GYANT’s first Design Principles

What we learned from rapid prototyping and iterating

We tested prototypes with users and collected feedback from team members in order to learn quickly. With users, we erred on the side of testing higher-fidelity screens. With team members, we didn’t shy away from showing wireframes and half-baked concepts.

Testing prototypes, team off-sites, and hosting a “Design Lunch” feedback session

Main chat

We focused heavily on designing the core of the GYANT experience — the main chat. We started with a belief (okay, my belief) that the majority of taps should be reachable with one hand. That is, the majority of user interactions should happen in the bottom half of the screen.

So we cut the screen “in half” (it isn’t actually 50/50) and focused messages from GYANT in the middle and the quick reply buttons and keyboard in the bottom half of the screen. This meant content would be in the upper half and interactions in the lower half. We found this resonated very well with users — no more struggling to reach the top portion of the screen (I’m looking at you, Apple).

From there, we explored layout variations and background ideas from wild to mild. Pop-ups that animate from the GYANT logo, form-like quick reply buttons, and quick reply buttons that scroll horizontally. We found bubbles positioned to the right side of the screen worked better because they indicated the user was sending something and not filling out a form.

We saw the background screen as a great opportunity to add delight to the experience and enhance the content from GYANT. We explored different colors and even toyed with the idea of adding a subtle background video; but we found this to be distracting and opted for a safer solid color that was on-brand.

We viewed the menu as both a problem and an opportunity. On one hand, the menu represented an interesting opportunity to bring delight to users by animating GYANT as an AI in a cool and surprising way. But on the other, it was a non-essential part of the experience and, at launch, would have very few options (one, in-fact… Settings).

In the spirit of pushing boundaries and breaking paradigms, we explored menu interactions that had no buttons (😱) and relied on users to swipe or tap-and-hold a sidebar to access the menu and other screens. As expected, we found this to be unintuitive and struggled to give enough affordance to show a menu was accessible without a button of some kind.

We then explored a side menu screen that could have a video looping in the background to animate GYANT as an AI. The menu would be accessible through a small icon in the bottom left corner of the screen. We found this interaction to be too distracting from the main Chat experience and that many users didn’t immediately understand what the icon was.

Our latest iteration is a bottom tray menu accessible by tapping a settings icon. One tap brings up the most common interactions (for now, “Reset chat”) and one more tap brings up Settings. So far, we see users finding accessing the menu and settings intuitively and easily this way. We will continue to revisit the menu interaction as our feature list grows.

User insights informed our big bets; our big bets influenced our design decisions

Bet 1: Design a content-first experience to improve efficiency and clarity

Content from GYANT and doctors are shown front-and-center, 95% of user interactions occur in the bottom-half of the screen, and the menu doesn’t distract from the main Chat experience.

Bet 2: Onboard new users through conversation to improve clarity and establish trust.

Greeting new users with an introduction from GYANT was our way of explaining what value the user would gain (feeling better), how to extract that value (chatting), and building trust with the user before asking for signup information.

We continued with this line of thinking when asking for push notifications and payment information.

Bet 3: Give a personal touch doctor profiles to increase trust

While doctor profiles hasn’t yet shipped, we have reason to believe that showing the doctor’s life outside of work will help build trust between the user and the provider — something that is not easy to do when texting a stranger for medical advice for only a few minutes.

Bet 4: Bring life to GYANT as a friendly AI to create delight

Illustrating GYANT as a friendly AI has proven to be one of the most difficult challenges in the redesign. We’re working with a talented brand and motion designer to get away from the traditional artificial intelligence brand and create a more friendly and delightful way of illustrating GYANT. More to come on this front 🙂

“The devil is in the details”

For such a simple app, this saying really held true. I communicated constantly with the iOS developer to work on details like the way quick replies animate up on the screen, the active and inactive states for buttons, and adding haptic feedback to give users another level of responsiveness.

Results and reflection

The app was released in the App Store in March 2018. As we await the results of our work, I’ve had time to reflect on our wins and losses aside from the metrics we’re tracking.

What we did well

We moved quickly — taking the holidays into account, we went from 0 to launch in just under 12 weeks. We didn’t let process be our product — while the case study frames the redesign in terms of process, our goal was always to build the best user experience and ship quickly. We got technical feedback early — I looped in the iOS developer early-on to get his technical feedback. We prioritized ruthlessly and left out features that would cause delays.

What we can improve

We didn’t test prototypes as early and as often as we should have — while the app nearing the end of development, we got user feedback that prompted us to change a small feature. This could have been avoided by testing with more people earlier.