top of page
LLM-Powered Customer Shopping AI Service Product 🤖️
B2C | Service Design | Conversational Design | Ecommerce
“How can we alleviate anxiety during supply shortages and foster a positive mindset while awaiting customer service responses during COVID, to ensure the smooth operation of daily goods transactions?”
Problem
During COVID, 30% of orders were returned, and a lot of customers experienced unsatisfaction.
My Role
Interactive Designer
Solution & Impact
I designed the interaction and the interface of the AI Shopping Assistant and the Merchants-customer Chat Bot within Taobao to automate the service process, reducing the 27% return rate.
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
Affinity Diagram, interview documents, key persona, journey mapping, hi-fidelity prototyping, visual design
Team Members
1 Project Manager, 3 Algorithm Developers,8 Engineers, 2 Researcher
Create AI System One Click Away
A feature in the Taobao that assists customers with daily shopping queries.
Executive Summary Executive Summary Executive Summary
A Sense of Control & Visibility
Intelligent Prediction
Streamline the Inquiry Process
Personalize Conversation with LLM
Memory Mode to Make Customers Feel Cared
Business Goals
Long-term goal:
Ensure the smooth operation of daily goods transactions during the pandemic 😃.
-
Reduce refund rates
-
Increase AI agent task completion rate in conversations
-
Shorten real agent service time
-
Strengthen the emotional connection between customers and products
Short-term goal:
Impact 🎉 (User Test Results)
95%
AI Task Success
4.2/5
Star Rating
- 27%
Return Rate
85%
Engagement Rate
Research & Sprints Planning
Challenges
During the onset of the COVID-19 pandemic, with stores closed and people stuck at home, the demand for online shopping skyrocketed. Suddenly, lots of customers ordered on Taobao, and businesses experienced a surge in order rates like never before.
This should have been a period of unprecedented growth and success, but there was one significant problem: 30% of those orders were being returned 🥲.
Design Process
Customer Data Analysis
The target customer segments
Identify the right problem
Customer:
" I bought the wrong size because the seller didn't respond to me. "
Merchants:
“ There are too many inquiries and returns, my staffs are overwhelmed. ”
Taobao Agents:
" Too many repetitive inquiries... "
Design Solutions
New AI Shopping System Sitemap
AI Conversational Journey Mapping
After the team finalized the new sitemap, I began designing the AI-powered agent conversation journey to ensure customers get the info they need in the fewest steps possible.
Brainstorming
We decided to build a more conversational AI service. Within 24 hours, I generated a list of potential solutions, evaluated trade-offs, and subsequently facilitated stakeholders' workshops to gather feedback and validate assumptions.
Prototype & Test
Streamline the Process
We Use Machine Learning to sort a list of items
A Sense of Control & Visibility
We aim to provide a detailed delivery tracking feature to alleviate concerns and enhance a sense of control.
Predict target items
Provide an overview and checking detail option
Visualize the delivery
The detailed location breakdowns for each delivery point
Improve Engagement & CSAT
Considering customers' potential frustration after waiting, we partner with merchants to offer exclusive coupons, aimed at soothing emotions and encouraging future purchases.
Boosts Customer Service Efficiency
Use automation to handle frequent product usage queries, freeing up businesses to tackle complex customer support issues
The first version ❌
Start Live Chat With Merchants
Manually copy and paste the item name
However...
Customers: "Going back and forth to copy names is time-consuming... 🥲"
The second version ✅
Reduce the Steps ✅:
Predicted the Queries
AI Response:
Largely reduce the waiting time
Boost Product Exposure:
Automatically recommend products based on their data
Personalize Conversation with LLM
Reduced the frequency of direct inquiries to real agents, alleviating their heavy workload.
1. The Sorting Model & the LLM
2. Capture Customer's Preferences based on purchases, browsing history and comments
3. Generate Personalized Recommendation Lists
78.5%
of user problems were solved in 4 rounds 😃
Track Data & Iterate
1. Implemented a HERL loop for continuous customer feedback collection to optimize human-AI interaction.
2. Introduced 'Memory Mode' to automatically reference previous order inquiries, addressing user frustration with repeatedly searching for order details.
3. Designed cartoon avatars and emoji packs to build trust with new products, resulting in a 5% increase in customer satisfaction according to research.
HFRL LOOP
1. A/B Testing: Test new and old response strategies among different user groups.
2. Feedback Collection: Continuously refine the new version based on metrics like "No Help" feedback rate.
Detect Negative Words & Provide Positive Emojis
Design emojis to build trust and positive relationship.
+ 5%
improved in satisfaction with emoji use in conversation 😃
Memory Mode to
Make Customers Feel Cared
Proactively ask about their previous order inquiries, making customers feel cared.
Qualitative Feedback
Customer:
" No need for waiting for a real agent! "
Merchants:
“ “It saves me time. It’s hard to operate without the service. ”
Taobao Agents:
" It helps me to focus on more complex customer issues. "
What Did I Learn?
1. Thoroughly designing each touchpoint is crucial in service design.
2. Provide Emotional Design to Customers
3. Balance the Cost of AI with Project Profit
bottom of page