Coinbase · 2025
AI powered customer support
Role
Lead Designer. Defined strategy & vision, presented to leadership.
Team
Customer Support + ML Engineering. Collaborated across 4 orgs with different XFN partners.
Duration
6-month project to rethink A.I in customer support.
Impact
24pp decrease in ticket escalation.
The problem
46%
of support cases escalate to a live agent
~$1.5M
monthly cost of agent-led support
Experience evaluation
Two problems stood out:
- Accessing support was hard.
- Once users found it, the experience was overwhelming.
Solution
I ran a series of cross-functional workshops with Customer Support, ML Engineering, Content, and Product to align on two clear pillars the experience would be built around.
Pillar 1: Easy-to-follow answers
Pillar 2: Be available till the end
Pillar 3: Personalised support
Visual design for all things A.I
A.I responses can take a few seconds to generate, and that wait starts to feel long when you're staring at a static screen. Instead of a plain spinner, we designed a branded loading animation that signals progress as the model thinks, and gives people something a little more interesting to look at than a boring circle.
Since then: Concierge for high-net-worth customers
Chat as a platform held up. The same foundation we built for mass-market support was extended to power Concierge, a premium, human-first chat experience for Coinbase One customers. End-to-end encrypted, personal, always-on.
Key learnings
A team is only as sharp as the goal it's pointed at. The hardest part of this project wasn't the craft. It was getting 4 orgs, 4 roadmaps, and dozens of people aligned on what "good" looked like. I also learned how much it matters to build trust with your peers' managers, not just your own. That's where the real leverage lives.