❊ Shipped in 2024
AI Sommelier Chatbot
I made 1,000+ wines into easy picks, and grew sales 10% in 2 months
Timeline
Sep - Dec 2023
Role
UX Designer
Researcher
Platform
Mobile Web App
Stakeholder
CTO
Data Team Lead
Dev Team Lead
Client
Carrefour
⦿ TL;DR
+10%
revenue growth within 2 months
95%
Users added wine to cart in ≤3 searches
1st
Google Gemini chatbot in Taiwan retail
"AI sommelier can really give me the right wine — in just 2 minutes."
— post-launch shopper feedback
❖ Context
Overwhelmed shoppers, stalled sales
✕ User Pain
"I don't know where to start.
The 100+ bottles all look the same."
— Customer feedback
74% of shoppers walk away when overwhelmed by choice.
$ Business Concern
Carrefour's wine sales were stalling.
They bet on AI to move bottles.
Growth collapsed from +8.5%/yr to almost 0% since 2020.
Once-steady wine sales never recovered post-COVID.
✧ My design strategy
An AI that thinks like a sommelier, talks like a friend.
Sommelier-style guided choices, number scales + profile snapshots, and food-pairing icons
Step-by-step guidance inspired by sommeliers for confident wine choice
I turned sommelier questioning into simple guided choices, to empower customers with expert-level selection.
Number scales and profile snapshots make wine easily imaginable
Replaced jargon like 'tannic structure' with 1-5 scales and plain-language profiles shoppers can compare at a glance."
Food-pairing icons turned wine shopping into dinner planning
Mirrored recommendations to real meals, so customers pick by what they're eating, not by grape varieties they don't recognize.
Full Story
⦿ Research
To ship AI under tight timeline
I dove into technology, domain, and users
"What great AI UX looks like"
"How experts guide decisions"
"What users actually ask"
Sommeliers have a proven system. Wine jargon locks people out.
Sommelier's 3-step guidance
Knowledge barriers blocked shoppers
Key Feature 1
Expert-informed AI conversation flow
◎ Context
People leave empty-handed when wines all look the same without guidance.
❖ Challenge
AI conversation was too slow, too costly, and too unpredictable for a 4-month MVP
↔︎ Key Decision 1
Step-by-step guidance inspired by sommeliers for confident wine choice
I turned sommelier questioning into simple guided choices, to empower customers with expert-level selection.
↔︎ Key decision 2
To ensure reliable, fast responses, I designed structured choices to lead conversation
Structured selections guaranteed reliable recommendations without model training. Reduced decision anxiety for beginners.
↔︎ Key decision 3
To prevent hallucination, I designed a workflow to pre-process wine profiles with AI and human verification
Key Feature 2
AI-simplified wine profiles and illustrations
◎ Context
Wine jargons confused customers and stop them from buying
❖ Challenge
How to make AI tone approachable without losing expert credibility?
↔︎ Key Decision 1
Number scales and profile snapshots make wine easily imaginable
Replaced jargon like 'tannic structure' with 1-5 scales and plain-language profiles shoppers can compare at a glance."
↔︎ Key Decision 2
Food-pairing icons turned wine shopping into dinner planning
Mirrored recommendations to real meals, so customers pick by what they're eating, not by grape varieties they don't recognize.
Key Learnings
