TL;DR: my team designed, measured, and improved an AI Assistant that helps 60k+ monthly users solve real problems.
Background
Eventbrite is a global ticketing platform that brings the world together through live experiences. As a senior content designer on the Service Design team, I designed experiences for event organizers and attendees who need help. My team’s mission was to empower customers to find a solution to their problem and connect to human support.
The challenge
In 2023, the business invested in AI. We designed a new channel for our audiences to get answers to their questions.
Audiences:
- Organizers: people who create events and sell tickets on Eventbrite
- Attendees: people who explore events and buy tickets
My role involved shaping strategy, leading research, and designing content for the AI Assistant.
To shape our AI strategy, I started by clarifying business requirements and user needs. I surfaced existing insights from my past research on our audiences and interviewed eight internal stakeholders about AI. This early exploration surfaced several challenges for the project.
User problems and constraints
- Our customers ranked “chatbots” as their least-preferred method for troubleshooting issues.
- Various teams tried to launch “help” chatbots numerous times before, with little success or adoption.
- Eventbrite attendees and organizers of free events are not allowed to contact a support agent for help, although they expect to. I was concerned our target audiences would expect to reach a real human if the chatbot did not resolve their issue.
Without a compelling use case and user-first design, I knew this AI project would fail.
The opportunity
My team and I followed a design thinking approach to envision, launch, measure, and iterate our AI solution.
In the early phases of the project, I aligned the team and helped shape our AI strategy–focusing on sound use cases. I collaborated closely with my team and other folks across functions. My team included engineers, UX designers, UX writers, operations folks, and business stakeholders
AI Help Assistant goals:
- Provide better ways for customers to quickly solve their own issues
- Expand help options for customers who currently lack access to human support (event attendees and organizers of free events)
As we devised an AI strategy we considered questions like:
- What user habits and behaviors can we leverage to make our AI solution more adoptable?
- How many users are affected by the use cases we’re trying to solve?
- Who will interact with our AI product? What will they be doing, thinking, and feeling?
- How will we know if the AI is working for our customers and achieving business goals?
The design process
1. Discover
I facilitated a workshop with folks across Eventbrite to explore our users’ needs, business goals, and AI capabilities. We brainstormed ideas to solve high-priority use cases and drafted our vision for an LLM-powered conversational interface.
2. Design
I created a product requirements document. With the team aligned around a vision and problem to solve, we designed experience principles, voice and tone guidelines, wireframes, and other artifacts to guide our AI Assistant prototypes.
3. Test
We launched two versions of AI Assistants to pilot different vendors. I led user testing to evaluate how our users would respond to the chatbot, assess clarity and usability, and explore other use cases. After testing, we made improvements to the chatbot based on user feedback.
4. Measure
Given the task success rates from the user testing and user analytics met our goals, we launched the AI Help Assistant to an expanded audience. In the months following, I conducted a content audit to measure how often the AI assistant resolved users’ issues.
User testing results
Observing our customers interacting with the AI Assistant gave the team insights to improve the user experience. We were pleased to see high task success rates and eager to fix pain points. Participation in user testing research evals also boosted customers’ attitudes toward Eventbrite (measured via NPS score).
Most importantly, the user testing guided our decisions about future use cases. Since launch, the team has iterated and optimized the AI Help Assistant.
Impact
Key metrics from the Eventbrite AI Assistant
Resolved 40-89% of issues
I led a content audit that showed the AI Assistant resolved the help-seeker’s issue in 40-89 percent of conversations
27k users found lost tickets
An automated flow I built helped ~27,000 users find their lost tickets in its first year after launch—addressing a critical customer pain point
Helping 60k users per month
In its first year, customer interactions increased 70%, resulting in 60k AI Assistant interactions per month
Expanded globally
Based on its initial success, leadership expanded the AI Assistant to 26 countries in local languages
I’m most proud that this project was truly collaborative and user-centered. The team of designers, engineers, business stakeholders, operations leaders, and customer support representatives sought insights about our customers’ needs every step of the way. We stayed focused on the problem we were solving, and remained diligent on moving the needle.
Help Assistant provides one-click access to lost tickets
I also led projects to build automations: powering the chatbot with customer data from disparate systems to solve more help issues. For example, when users ask about lost tickets, the AI Assistant can “scan” their account and provide one-click access to their tickets.
Select image to enlarge
