CHATBOT
REDESIGN
2018 Internship In Thinking Chat
Improve the user interface and the user flow of the AI software to increase overall user experience
OVERVIEW
For customers, artificial intelligent(AI) chatbot can help to find the information they are looking for on their own terms in a more dynamic and efficient way.
For companies, the AI chatbot can use artificial intelligence to provide their business with a steady stream of qualified leads, which is more effective and low labor costs.
As a design intern in Thinking Chat, I was given the MISSION to redesign the user interface and chatbot auto-response, creating a user-friendly interface and conversation to enhance the overall user experience.
MY ROLE
UX Researcher
UX Designer
UR METHODS
Heuristic Evaluation
Competitive Analysis
Contextual Inquiry
Usability Testing
DESIGN METHODS & TOOLS
User Flow (OmniGraffle)
Prototype & Wireframe (Sketch)
Interactive Animation (Adobe AE)
DURATION
Three Months
Jun - Sep, 2018
ANALYSIS
Brainstorm & Find a Break Point & Break Down the Mission
AI chatbot as an agent is automated but appear like a real human to the response. BUT HOW? What can we do to make it appear like a real human?
Before conducting the research on this project, I broke down the mission and brainstormed about the task points, as well as what possible changes could be made.
Heuristic Evaluation & Competitive Analysis & Contextual Inquiry
I spent four weeks to identify what potential problems exist, identify how competitors deal with a series of questions, understand the opinions of users in terms of some specific issues.
Finally, I found FIVE major potential problems.
RESEARCH
Problem I. Users Prefer Not to Provide Privacy Before Trust is Established
Privacy Problem
- Are users willing to fill in the name and email before they chat?
- Whether asking for the name and email in the form of a chat is more easily accepted by users?
User Quotes:
I just want to ask a very small, simple question, and I don’t want to fill in my email…
Before I know we need further connection, I will not leave my personal contact information…
I probably will shut down the chat window, when I see I must register to start a chat…
In the semi-structured interview with 10 participants, 8 of them preferred not to provide their personal information and voted for answering during a chat. Starting a chat without filling name and email is more accepted by users. So I proposed the following solution with a SIMPLE start and a NATURAL request for contact information during the chat.
Problem II. One Long Message is Hard to Read
Our current chat bot responds with one long message, which is completed and clear.
- Is this like a real human?
- When we send messages to our friends in real life, we usually send several short messages continuously.
From the feedback of the participants, all of them consider using short sentences is more readable than a long paragraph. To make the AI agent to behave more like a real human, I suggested to break down long sentences, use beyond just text with emoji, buttons, images, links, and cards, as well as add typing behaviors.
Problem III. The Chat History is Not Visible to Users
From the chat window on Thinking Chat, the chat history is saved in the company’s database, which is not visible to customers.
- Customers cannot review the contents they have ever asked when they log in again.
- Describing problems repeatedly can reduce chat efficiency and lower user satisfaction.
User Quotes:
It’s the second time I log in the chat window, I’d like to talk to the human agent who helped me last time since he knows my problems better…
After I described my questions to the chatbot, if I am transferred to a human agent, I still need to repeat my questions again. Why not talk to a human directly?
Also, as it can be seen from the chatbot I reviewed in the competitive analysis, there are two ways to follow up a conversation, one is a single page with all chat history, the other one is classifying chat history by different agent.
• One single page with all chat history
Pros:
- The old chat history can be visible to all human agent.
- There is no need to switch between human agents.
Cons:
- It could be very long and messy.
• Classify chat history by different agents
Pros:
- Users can choose a certain agent to continue.
- It looks tidy.
Cons:
- Need to switch between different human agents
- Hard to locate a certain content.
Considering the users' feedback, as well as the pros and cons of competitors, I proposed to display the chat history to users, classifying by different agents and adding a search bar on the top of the different conversations which helps on locating certain contents.
Problem IV. It Cannot Deal with Non-Programmed Questions
When facing a non-programmed question, the AI smart agent of Thinking Chat tends to transfer to a human agent directly.
- Is there a better way to deal with non-programmed questions?
- How would a real human agent answer these kinds of questions?
User Quotes:
Sometimes I just want to test whether it is a bot on the other side, I would ask “fooling” questions, if being transferred to a human agent at this time, I may feel embarrassed…
I would not prefer to wait for a human agent for a long time. If the bot can provide a relatively good answer for me, that would be amazing…
I also test some non-programmed questions with competitors. When I test “what is the color of the moon”, two of the bots transferred to a human agent. And what humans answered are “The moon is grey by the way”, “I'm guessing you are trying to check if I'm a bot”, etc.
I think this is one of the differences between human and bot. A human can answer non-programmed questions based on their understanding, while bot cannot. If we want our chatbot to be “real”, we can try to answer these “fooling” questions, like by linking it to Google result or asking for a different way of describing.
Suggentions:
- Try to match an answer from the database, based on the keywords of the question.
- If the chatbot cannot match an answer, search an answer from Google.
- Always ask for feedback and provide access to a human agent.
Problem V. It is Hard to Control the Best Time to Transfer to Human Agent
Some complicated problems that automated live chatbot cannot solve will be transferred to a human agent, following up by email, phone or live chat.
- When is the best time to transfer to a human agent?
- For further contact, which way is more preferred by users, phone, email, or live chat?
- For further contact, which way is more preferred by users, phone, email, or live chat?
User Quotes:
My issues are very complicated. I believe the bot cannot solve them. And sometimes I hope to save my time and talk to a human agent directly…
There are 6 of 10 participants considered the best time to transfer to human agent depends on the issues, whether the bot could answer it, whether I am in a hurry, etc. And they have different preferences on the contact ways for further connection, either phone or email. All of them hope to be able to choose the preferred contact way by themselves. So I suggested providing a button for calling a human agent at any time and request for contact information in the form of a tab.
ITERATIVE DESIGN
User Flow & Wireframe & Prototype
Overall, based on the data I collected and considered the technical limitation, I made the following changes to improve the user flow and user experience.
• User Flow •
• Wireframe •
• Check more on this video •
REFLECTION
This is my first internship on UX design/research area and first time to work on Artificial Intelligent area. Also, it is almost an independent project which means a very good opportunity for me to exercise. I have learnt a lot of things from it, including how to rapidly gain knowledge from zero to one in an unfamiliar field, how to work remotely and efficiently, how to think about problems from a business perspective, how to support the ideas on the basis of research, as well as how to present my design concisely and professionally. I felt so excited when my recommendation was adopted by others. Also, I was very happy that my work was satisfied by my boss. Hopefully, my design could be shown in the next version of Thinking Chat.