NYP CHATBOT-Find a doctor

NewYork-Presbyterian Hospital

User Experience Designer

Spring 2018 - Present

Background & problem

NewYork-Presbyterian Hospital (NYP Hospital) receives over 2000 phone calls from patients and caretakers asking various questions every day. Syllable partners with NYP hospital trying to help reduce their burden by building a chatbot on their website.

design statement

Building a chatbot to assist NYP website visitors to solve their problem faster and easier.

Product: AI + design. Design should help user to use bot smoothly. Whether the bot can be successful is not decided by design. But design plays a huge impact on guiding users and providing valuable information.


Work with NYP staff - Understanding problem
Online intercept survey - Learn user's need
Onsite user testing - getting feedback from user

working process


NYP wants us to help them reduce the burden of call center. But how can this chatbot really meet the user's need?

Our goal: How can the chatbot assist users? What do the users need from the chatbot?



early research
Intercept online survey

The aim of the survey was to understand people’s needs, and expectation on NYP website. Moreover, we wanted to see how familiar people are with the idea of an assistant on a website and with the products powered by artificial intelligence.

Screen Shot 2018-08-05 at 6.29.53 PM.png

Majority of the visitors searching NYP website for:

-  Lab results
-  Finding doctor's number
-  Find a new doctor
-  Schedule an appointment with a doctor
-  Learn about a disease/illness



decisions for building mvp


After discussing with NYP and our product team, we decided to set "find a new doctor" as a priority before launching. So my responsibility is to think and design an experience for patients/caretakers to easily find a new doctor to meet their needs.


Before jumping into design, it was important to define success and ask the right question. Here are several questions I was trying to understand.

before design


different phases of design

phase 1

Focused on how to put doctor result card into the chatbot


Question to ask


What doctor information is needed in the profile page? How to display them?
How the current doctor website help people "find a doctor"?

nyp web.jpg

The doctor information includes:
Doctor name, Profile picture, Phone number, Expertise, Specialty, Biography, Insurance, Location



design challenge


design exploration

Early exploration:

Open a new page or Collapse & Expand?

How to put all the doctor information in a small chat window?
How to provide better experience for users to view the doctor information and choose the doctor they want?


From list view to full profile page: Open a new page or Collapse & Expand?

0 newpage-vs.-expand.jpg

After discussing with the product team, we decided to go further on "Expand & Collapse" version, since "opening a new page" action would cost more efforts for users to choose between different doctors.


Further exploration:

2 versions of Expand & Collapse
Version 1: Too many 'view all' clicks for different doctor information from users, which was not efficient
Version 2: One time click of "view all" button to see full profile, which solved the problem of version 1


Participants: ID 101-104
Online platform: TrymyUI

Result highlight:
1. The ways user asking for 'find a doctor' are various. Some people typed key words, some people typed long sentences.
2. The needs of people 'find a doctor' are different. Some people were more casual, some others can be very specific (they had different preferences and demands).

User testing _ online


phase 2

thinking about the whole user flow and ‘find a doctor‘ feature


- What different questions may individuals ask the bot for 'find a new doctor'?Should bot ask some questions to narrow down the results?
- What information do users expect to know about doctors?

Questions to understand

Process and outcome

What different questions may individuals ask the bot for 'find a new doctor'?
To understand this question, I asked for the query sheet from the product team, which helped me understand different ways people may ask for 'find a new doctor'.

Screen Shot 2018-08-26 at 10.41.24 PM.png

The way users ask the bot for ‘find a new doctor’ can be diverse. It can be very broad, very specific, or something in between. The bot should have different reactions according to the different queries users input.

The questions users ask the bot can be:


What information do users expect to know about doctors?

As shown above, there are too much information to display for each doctor. How to arrange them appropriately in a small chat window has become the challenge.
Which information is more important to the user? What the hierarchy should be?

Intercept online survey


How to insert the concept of 'filter' into 'find a doctor' design?

Design challenge


Plan 1: Asking options before showing results
   P 1-1: Bot sending Quick Replies to ask user's preferences before showing results


 P 1-2: Adding a filter page before showing results


plan 2: Showing results right away with filter options


User testing &
group discussion


Result highlights:

Plan 1-1
The design of bot providing Quick Replies may result in too many clicks from the user, which is not efficient.

Plan 1-2

Too many options need user to choose before they see the doctor results, which confuse users a lot. And it feels not intelligent.

Plan 2:
This plan earns most likes. Many participants had ‘Wow‘ moment when they see the doctor result right away. And filters are also clear for them to click.

Based on the result, we decided to use Plan 2 for further development and iteration.



NYP onsite testing

Before launching, I went to NewYork-Presbyterian hospital with our UX researcher to run the onsite UX testing with patients and caretakers to gather feedbacks from the real users, which was critical for our product development.

testing tasks

Screen Shot 2018-08-19 at 10.27.30 PM.png
Screen Shot 2018-08-19 at 10.40.54 PM.png


qualitative data analysis


- 36% (8/22) of patients expected to book an appointment or see doctor’s office hours
- 23% (5/22) of patients wanted to see doctor’s reviews and number of years in practice
- 23% (5/22) of patients wanted to search a specific doctor by name


- After seeing doctor cards, patients were satisfied with the information on cards
- The 'view all' button was not obvious to users
- Using filters (e.g.: Insurance, gender) was not obvious, but patients got excited once they learned about filters from researchers

Highlights of the results


updated design


Design component in doctor result page:
- Result description     - Filter      - Doctor cards    - 'View all' & 'View less' button



-'Specialty' search bar was removed from the top of the doctor result
-Show users their own query on the top of the filter. It should stick to the top when users scroll the results
-Filters always stick to the top
-Once user selects an option in any filter, display user their choice in the filter itself (If the name is too long, use ellipsis)


-Change ’VIEW ALL’, ‘VIEW LESS‘ button color from light grey to a stronger blue color (selected from NYP branding color)
-Only keep profile image, doctor name and the specialty information in small card (first priority)
-The order of different doctor information is changed
-Only keep 1 call-to-action button. Delete “direction” button

NYP Assistant


How can users find a doctor using NYP assistant?

flow chart.jpg

what i learned

1. Design with real data