Overview
Carrefour Taiwan's AI Sommelier is a pilot product aimed to quickly validate AI's potential in driving business value.
They picked the wine department, one of the iconic but struggling business unit, as the testbed.
I led the end-to-end process from research, design, and development to delivery, revamping EVALUE’s charging experience.
Duration
2023.09 - 2023.12
4 Months (Design)
My Role
Project Lead
UX Designer
UX Researcher
Responsibility
Product Strategy
User Research
UX Design
Direct Stakeholders
CTO
Data Team Lead
Dev Team Lead
Problem Context
In retail, product variety should drive business growth.
However, Carrefour Taiwan's massive wine collection make customers get lost and leave empty-handed.
Also, the knowledge barrier stops beginners from exploring.
1000+ Wine Offering
Carrefour offers the largest wine collection in Taiwan
Staggering Growth
As per our PM's briefing and Taiwanese wine market report
Core Challenge
To validate AI's business potential, how might we:
provide accurate wine suggestion
make wine world approachable
Solution Highlight
An AI-powered guide, offering:
smart pairing suggestions
simplified wine language for beginners
Impact
"AI sommelier can give you the right wine in just a few seconds."
Henry Ting, Carrefour Taiwan CIO
+10% wine sales
Achieved wine sales growth within 2 month of launch, a proven success in AI's business potential
1st Gemini in Taiwan Retail
The 1st retail product in Taiwan to successfully deploy Google's Gemini AI
Research Activity
I researched Taiwanese wine shopping behavior through store visits, customer interviews, and online forums.
Market Insight
Taiwan's wine market: Customer segmentation and sales distribution reveal mid-priced market potential
Key Challenges
Excessive selection and knowledge barrier are the reasons why beginners can't move upmarket
"They all look like the same!"
Excessive collection paralyzes selection and drives customers away
"Is this the right one?"
Beginners can't imagine flavors from descriptions or food pair
Expert Insight
Sommeliers' 3-step guide to provide personalized wine recommendation
Synthesis I
Client aimed to launch the product ASAP.
So I decided to apply AI chatbot solution for its technical maturity, user familiarity, and future scalability.
Synthesis II
Expert-informed design:
I translated sommelier's guidance into AI conversation flow
Challenge 1
Abundant selection, yet hard to choose:
Every bottle looks similar without expert guidance
Solution 1
AI-Powered sommelier making personalized discovery simple, approachable
Get personalized wine guidance from your wine expert
Challenge 2
Rich wine descriptions, yet hard to imagine:
Beginners can't visualize taste or think of food pairings
Solution 2
From wine jargon to plain language
Simple figures and everyday language
AI-Simplified Description and Pairing Insights
Concise, intuitive taste profiles and images aid imagination
Reflection
Key Learnings
Designing trustworthy AI requires transparency
Early internal testing revealed that users were skeptical of AI-generated wine recommendations when they didn’t understand the reasoning behind them. To build trust, I introduced clear explanations for each suggestion, showing how factors like flavor notes and user preferences influenced the choices. This experience reinforced that AI must not only be accurate but also explainable to create user confidence.
How I would approach this differently today: CustomGPT
If I were to redo this project now, I would leverage custom GPT models to fine-tune AI responses, ensuring the tone, complexity, and recommendations align with user expectations. This would accelerate iteration cycles and allow for tailor-made responses based on real-time feedback. Additionally, by collaborating more closely with engineers from the start, I could define clear AI behavior and explainability rules upfront, reducing the risk of AI outputs feeling too generic or opaque. This would create a more seamless and trustworthy user experience.
