Client Usage Dashboard
Office Support AI System
AI | B2B | Native Mobile | Desktop | Enterprise Product
Office Support AI Assistant helps managers streamline the work flow, automatically respond to questions from other associates, perform automatic approvals and more. It allows associates more time to focus on other important and creative tasks.
.png)
Problem
Managers at Alibaba were spending up to three hours daily addressing routine employee inquiries, which negatively impacted their productivity and personal time.
My Role
Interactive Designer & User Researcher
Solution & Impact
A mobile AI assistant including voice input features was developed to automate responses to these inquiries, significantly reducing the time managers spent on routine tasks and increasing workplace efficiency.
Scope
I worked full-time at Alibaba from 2020 to 2021, during which I led this main project, one of four total projects I managed.
Deliverables
Contextual Inquiry Documents, Journey Maps, Wireframes, Prototypes, Data Analysis Reports, User Testing Documents, Graphics
Team Members
2 Project Managers, 2 Algorithm Developers，12 Engineers, Research Lead
Why Do We Need It?
Problem
"This repetitive work takes up too much time; I even have to deal with it after I get home. 🥲"
We have 250,000 associates, including 27,000 managers. The large volume of employee questions requires managers to spend up to 3 hours addressing them per day, which even impacts their family time.
Solution
So we want to automate managers' tasks and streamline associates' information access, ultimately increasing work satisfaction.
Ask PM About Our Strategic Plan
For This Quarter
We’re aiming to reduce associates' wait time and managers' task management time by 30% over the next 60 days.
But What's the Challenge?
So far we have a desktop version where managers can input questions and answers to train the model, and associates can use the model.
Associates Side - Ask Questions
Managers Side - Train the Model
However, the engagement rate (11%) is still very low, and we found out through user testing that single-question training takes 5 minutes, which is a bit long.
One of our users said: “ It's time-consuming and complex 😔..”
Make Decision with the Whole Team
We discussed insights and user behavior data and ultimately decided to improve the training process.
However, before diving into any changes, I want to collaborate with the research team to conduct user research, identify pain points, and explore potential opportunities.
Problem Statement
How Might We Reduce the Training Time and Increase the Engagement Rate?
What Important Data Did We Gather?
Survey & Interview
83%
Associates Submit Requests Through Their Phones
>50%
Managers work away from the desk
3.2
User Satisfaction Rate
What are the Managers' Product Usage and Pain Points?
Managers' Pain Points Identified
Combine the Insights and Identify the Trends
We initially thought they used Macs for work since everyone has one at the office. However, our research showed that most interactions occur on iPhones. Employees typically ask questions using their iPhones, and managers usually respond via their phones when they have time.
Therefore, we've decided to develop a simplified mobile version to train models.
Prioritize the Features With Teams
Design Solution
Launch a Mobile App for Easy Training of Models
Iteration 1
But though user testing...
I found that managers spend 50% of their time only on typing😔
Iteration 2
Voice Input Feature
User Testing
We reduced model training time from 5 minutes to 2.5 minutes 😊!
High Fidelity Prototype
1. Voice Input to Easier the Process
Utilize NLP, Automatic Speech Recognition, and Signal Processing for Accurate Speech-to-Text Conversion
.png)
Offer Support
Throughout the Process
Positive Feedback Increases 5% Satisfaction
2. Reduce Cognitive Load
Guide Users to Set Up Their First Q&A in 3 Easy Steps
Provide Video Tutorial
😄
Second QA Training Rate
We got some improvement, however, it still didn’t meet the success metrics 😔.
24%
Engagement Rate
3.5 (out of 5)
Satisfaction Rate
Our launch event increased engagement to some extent, but it didn’t meet our expectations. So I decided to collaborate with the research team conducting user interviews to identify the reasons behind this.
We overlooked the users' emotional responses and thoughts after training the model
Finishing the training is not the final step!
The design solutions I’m going to propose, which I’ve already tested, will help us to achieve the goal of our strategic plan.
I brought everyone together in one room to gain feedback and share my results from research and data.
Finally, we decided to Collect and organize Q&A data into different packages to meet various needs!
Iteration 3
Preset Q&A Packages to Save Time
Collect Q&A data and train the model to produce Q&A template to save users time
Become a Main Paid Feature After Commercialization
User Testing
We improved the Model Training Time from 2.5 minutes to 1 minute! 😊
User Testing is in a controlled environment, but I want to see how users interact with it in real-world settings.
40%
I conducted a contextual inquiry over 3 days and found that 40% of managers were unconsciously saving employees' questions in Apple Notes!
Iteration 4
One-Click Save Question
Eliminate the Need for Note-Taking in Apple Notes
😄
Popular Feature
From the Business Side..
Started the Commercialization in April 2020!
But We Have New Challenges
For a new product in the market, the two main risks we face are potential misalignment with actual user needs and low market acceptance.
Solution 1
A Closed-Loop Design Process
Quickly iterate the product to meet new market users' needs by continuously gathering feedback.
From the Product Side..
Human Feedback Reinforcement Learning
From the User Base..
Support Groups for Partner Companies
Solution 2
Mascot Design
Enhance Loyalty & Build Trust
Outcome
Business
Design
Learning
1. Team Collaboration:
When disagreements arise, effective communication becomes crucial. By providing data-supported analysis, one can rationally explain the importance and necessity of a decision
2. Real User Contexts:
By observing in real user contexts, researchers often identify issues that even the users themselves have not realized.