Disease Control Butler - Taiwan CDC line CHATBOT
–
YEAR: 2017-2020
pRODUCT pLANNING: JC YEH
UX & ui DESIGN: JC YEH, FIFI HUANG
visual design: CONNIE KANG
character design: CONNIE KANG
user research: jane wang, Yin Peng, JC YEH, FIFI HUANG
This AI chatbot answer questions about more than 90 infectious diseases 24 hours through MNLP (Medical Natural Language Processing) technology and evolved through machine learning. Early in 2020, Disease Control Butler offers the latest information about COVID-19 like disease explanation, prevention measures, guidelines for arriving in Taiwan, and worldwide pandemic situation reports. It also integrates with the mask map function to help the public get access to disease prevention resources.
Goal
Create a chatbot that can share the Taiwan CDC toll-free hotline service loading in pandemic and influenza season.
Design challenge
Gaining trust from the user. Getting user familiar with conversational UI and interaction.
Solution
We create a virtual character that feels like an actual person who works in CDC. User can text to this chatbot just like texting to a friend and ask questions about infectious diseases.
STRATEGY
To create a virtual character that knows how to interact with the user, we carefully design every reply message from Disease Control Butler. Very much like writing a screenplay - we conduct the interaction and props design of every scene.
Interaction Design
How the chatbot behave when answering a question should be similar to human texting.
Props Design
When displaying complex information, chatbot use metaphor of actual documents like cards, sheets, forms and diagrams.
Character Design
Through co-creation workshop, we defined the appearance and characteristic of Disease Control Butler.
INSIGHT
Research Method1. Desk research
2. Statistical data of CDC hotline
3. Interview doctors and professionals
4. Interview related NGOs
After CDC decided on the topics they want to raise awareness in the next version, they will send the statistical data of related hotline calls' record from the last five years. Therefore we can analyse it to find out what the public usually asked about these diseases. From the hotline record, we've also learned about the context and connection of each question. Meanwhile, we conduct several interviews with the doctors from CDC. These insights inspired the features which helps the user to get disease prevention resources much more effortless.
INFORMATION ARCHITECTURE
Learning from the research that a caller usually asks 2-3 related question in the same call. We tried to map out the connection between every question and create the IA of infectious disease Q&A. IA is crucial when designing a chatbot. Every intent defined in our NLP training data and the conversation flow design were all based on IA.
INTERACTION DESIGN
Menu
The menu of Disease Control Butler is the key visual to communicate the scope of his service. It helps users understand what sort of question they can ask and the latest disease information they need to know.
1. The upper buttons are about the latest pandemic and the features they might use rapidly at the moment.
2. Others are the entry point of the general topics about infectious diseases prevention. These items were defined based on our IA.
Conversation Flow
With the NLP technology, user can text their question the same with their everyday messaging. They can even have some small talk with the chatbot and learning more about his character.
After Disease Control Butler answered the question, he offers 3-4 related questions to carry out the following conversation. Continuous dialogue helps the user learn the details without explaining everything in long messages.
Caring
If a user asked questions about personal illness, Disease Control Butler will remember it and asked about it moments later to show that he cares about him. When the user asks about the pandemic in other countries, he will remind him about the preventive measures and the vaccine he needs to inject before travelling.
Micro Survey
When we deliver a survey, the format doesn't have to be a questionnaire. With the micro survey, we can ask user 2-3 questions in a conversation. Since the dialogue looks very much like small talks initiated by the chatbot, users usually answered them willingly and instantly.
Using Props
When we have a conversation with a person, we explain complex information with written words or drawings on paper as a prop. Therefore we designed our chatbot to have the same behaviour. The user can find the answer he needs from virtual props and keep them in the conversation for later use.
PROPS DESIGN
The props Disease Control Butler sent out were constructed from LINE flex message elements. We designed a prop's layout based on its actual document and created three different formats - cards, sheets, and forms. To have a consistent design across designers, I have created the LINE chatbot design system and made a sketch library for my team.
CHARACTER DESIGN
Disease Control ButlerGender: Male
Age: 30-35
Profession: Staff from CDC
Character Traits:
-Trustworthy
-Fashionable
-Communicative
-Sense of humour
As a virtual character, Disease Control Butler is one of the public faces of CDC. Therefore we defined its appearance and characteristic with their PR team in a co-creation workshop.
Appearance
We offered photos of faces to help CDC find Disease Control Butler's image and extract the possible facial and figure features for the appearance design. In order to create a memorable character, the designer put in some distinctive features like birthmarks and messy hairs.
Characteristic
Disease Control Butler express his character traits through the pet phrases he uses in the conversation. We defined these phrases in the workshop and inserted them in different dialogues.
2.2M
Users
20%
Hotline callS reduced
25%
W1 retention RATE
46
NPS SCORE
FEEDBACK & DESIGN UPDATE
Learning from: Product Data
Issue: 40% of users drop-off in the first dialogue
The early design of Disease Control Butler's onboarding message tried to introduce himself and nudge the user to ask questions in the first dialogue. There was too much text to read and makes the character overly hospitable.
In the re-design version, Disease Control Butler only introduces himself and makes acquaintance. The user can easily hi back and learn more about what he can do in the later dialogues.
Learning from: Usability Testing
Issue: Lack of the following action when Disease Control Butler doesn't have the answer.
User Feedback: "When the chatbot doesn't have the answer, I assume he will ask someone for me."
With NLP technology, we can identify the intent of the user's question. We've established the workflow to hand over these question to CDC staff and get them answered. Once Disease Control Butler got the updated answer, he will text the user who asked about it.
Video from Taiwan CDC