Glossier
Glossier
Glossier
EmpathAI is a Zendesk plugin that helps Glossier's support team scale without losing people. It reads emotion and urgency in real time, checks product issues through photo analysis, and suggests personalized fixes based on order history. A human agent still steps in to handle the relationship.
Category
Category
Beauty
Beauty
My role
My role
Product Concept
Product Concept
Timeline
8 weeks
Process
AI chatbots often create emotional disconnect.
When a chatbot misses emotional cues, users feel unheard, especially when they're already stressed. That gap breeds frustration, drop-offs, lost trust, and avoidable escalations.
Process
AI chatbots often create emotional disconnect.
When a chatbot misses emotional cues, users feel unheard, especially when they're already stressed. That gap breeds frustration, drop-offs, lost trust, and avoidable escalations.
Solution
An emotionally intelligent chatbot that detects real-time user emotions.
It adjusts its tone, communicates transparently, and hands off to a human when needed. This builds customer trust, wastes less time, minimizes unnecessary escalations, and feels human, not robotic.
Process
Journey Map
Users don't actually trust AI to understand their problem. They brace for slow, canned responses. When they get a robotic tone, repetition, and rigid policies, it confirms their expectations.
Solution
An emotionally intelligent chatbot that detects real-time user emotions.
It adjusts its tone, communicates transparently, and hands off to a human when needed. This builds customer trust, wastes less time, minimizes unnecessary escalations, and feels human, not robotic.
Process
Journey Map
Users don't actually trust AI to understand their problem. They brace for slow, canned responses. When they get a robotic tone, repetition, and rigid policies, it confirms their expectations.
Main Flow
Empath’s key features
Real-time signals and context: emotion, escalation risk, and urgency, plus instant access to order history, returns, and loyalty status.
Smart recommendations: on-brand response suggestions and personalized resolutions (gift cards, escalation, etc.) based on behavioral patterns.
Faster photo verification: automatic detection of product issues and image quality checks, cutting back-and-forth and speeding up refund/replacement approvals.
Main Flow
Empath’s key features
Real-time signals and context: emotion, escalation risk, and urgency, plus instant access to order history, returns, and loyalty status.
Smart recommendations: on-brand response suggestions and personalized resolutions (gift cards, escalation, etc.) based on behavioral patterns.
Faster photo verification: automatic detection of product issues and image quality checks, cutting back-and-forth and speeding up refund/replacement approvals.
Major Improvements
Swapped chat and customer info panels to match users’ established chat mental models.
Moved AI-suggested responses into the chat itself instead of a side panel, reducing cognitive load.
Added resolution time tracking, reinforced by a completion pop-up, to keep agents focused on efficiency goals.
Major Improvements
Swapped chat and customer info panels to match users’ established chat mental models.
Moved AI-suggested responses into the chat itself instead of a side panel, reducing cognitive load.
Added resolution time tracking, reinforced by a completion pop-up, to keep agents focused on efficiency goals.
Lofi - UX Pilot
HiFi - Magic Pattern > Figma
Reflections
The key product decision was knowing when to loop in a human. Since the goal was to scale customer service without sacrificing loyalty, responses couldn't feel AI-generated.
The main process takeaway: AI prototyping tools like Magic Pattern collapse the traditional lo-fi-to-hi-fi journey. That used to force problem definition and scoping decisions through slow iteration. Without that built-in checkpoint, the rigor has to happen upfront, since no slow middle exists to catch what was missed.
Reflections
The key product decision was knowing when to loop in a human. Since the goal was to scale customer service without sacrificing loyalty, responses couldn't feel AI-generated.
The main process takeaway: AI prototyping tools like Magic Pattern collapse the traditional lo-fi-to-hi-fi journey. That used to force problem definition and scoping decisions through slow iteration. Without that built-in checkpoint, the rigor has to happen upfront, since no slow middle exists to catch what was missed.
