Transforming Customer Service with AI
Guided Conversational Chatbot
Reimagining Automated Support with Empathy
Transformed a 78% abandonment rate into 34% higher automation by redesigning from 100+ user journeys.
Project Summary
I led the end-to-end redesign of Zalando's AI chatbot, which had become the lowest-performing customer support channel with a 20% intent recognition failure rate and the highest repeat contact rate across all channels. Through mixed-methods research, I uncovered that the core issue wasn't usability—it was deep-seated customer distrust in chatbots. By architecting a guided, button-driven conversational flow and introducing counterintuitive design decisions like simulated "thinking" delays, I transformed the chatbot into an intelligent partner that proactively solves problems and rebuilds customer confidence.
Timeline
14 Months
Market
Europe: E-Commerce, Fashion, Customer Service
Role
Senior UX Designer
Team
- Otobong Okoko (Senior Product Designer)
- 1 Product Manager
- 2 Lead AI/ML Engineers
- 3 Software Engineers
- 1 Data Scientist
- 1 Content Writer
- External Partnership with UltimateAI
Strategic Objectives
Transform chatbot from cost center to strategic asset through improved automation and customer trust.
Increase Automation While Maintaining Quality
Transform the chatbot from a cost center to a strategic asset by dramatically increasing the automation rate for key use cases, particularly delivery issues, without sacrificing the customer experience.
Reduce Repeat Contacts
Solve customer problems effectively on the first interaction, eliminating the frustration of having to contact support multiple times for the same issue.
Rebuild Customer Trust
Address the fundamental psychological barrier preventing chatbot adoption—deep-seated distrust born from negative past experiences with conversational AI.
Designing for Trust, Not Just Efficiency
Our research revealed a critical insight: users feared "not typing the right keywords" in free-text chatbots. The solution wasn't to improve NLP accuracy alone—it was to eliminate the anxiety entirely through a guided, button-driven experience. I analyzed 100+ chat transcripts to identify failure patterns and mapped them against our technical capabilities, establishing three core design principles:
- Be a Partner, Not a Gatekeeper: Proactively offer solutions rather than forcing users to navigate complex menus
- Remember Everything: Maintain full context throughout the conversation to eliminate repetition
- Always Offer an Off-Ramp: Make escalating to human agents transparent and seamless with full conversation summaries
The "Patience" Paradox
Usability testing uncovered a counterintuitive finding that became pivotal to our success. Users described the bot's instant responses as "unnatural" and "irritating"—the bot was too fast, which paradoxically eroded trust. I introduced a 3-5 second simulated "thinking" delay with a typing indicator before responses. This small change made the experience feel more human and thoughtful, leading to a 60% improvement in user perception and acceptance in subsequent tests.
Proactive Intelligence Through API Integration
Rather than asking users to hunt for order numbers or parcel details, I designed a system that automatically fetches recent orders via API and displays them in a visual carousel. For clear-cut issues like lost parcels, the bot now initiates automated refunds directly in the conversation after performing risk checks—providing instant resolution that users never expected from a chatbot.
Graceful Handovers
When escalation is necessary, the chatbot summarizes the entire interaction and passes that context directly to human agents. This eliminated the user's biggest fear: having to repeat themselves and waste time after a failed bot interaction.
Automation Performance
Improved efficiency and customer satisfaction metrics
Increase in automation rate for delivery use cases (from 11.76% to 38.36%)
Decrease in Repeated Case Share, proving first-contact resolution improved significantly
CSAT scores rose significantly in post-launch monitoring
Estimated annualized operational savings from increased automation
Customer Feedback
Qualitative impact and user testimonials
Before this user test, I would have opted to call instead of using the chat. Just now, I realize how efficient and workable the chat is. It gives me all the information I need, and it's easy to use. If I had known this before, I would have used it instead of calling.
The Biggest Learning
Design for trust, not just speed. The primary barrier to chatbot adoption wasn't usability—it was psychological. The counterintuitive decision to intentionally slow down the bot's response time proved that in conversational AI, the perception of thoughtfulness can be more important than raw efficiency.
What I Would Do Differently
Involve the Content Writer from day one. We spent significant time in later stages refining the bot's tone of voice and empathy. Establishing this content strategy earlier would have streamlined the design process and strengthened the emotional connection with users from the start.
Protected Case Study
Case study is only available on request.