The problem
Most customer requests are routine — hours, scheduling, order status, simple account changes — yet they tie up agents and, after hours, go unanswered. The opportunity was an AI employee trained on a business that could handle that volume autonomously, while escalating anything sensitive to a person without losing context.
What I designed
- Onboarding & setup. A guided flow to teach XBert a business — its services, policies, hours, and FAQs — with preview-and-test before going live, so teams could launch with confidence.
- Conversation & routing. Patterns for answering across voice, chat, and text, reading caller intent, and routing to the right team or task — resolving when possible, escalating when not.
- Human-in-the-loop handoff. A seamless transfer that hands the agent full conversation context, so customers never repeat themselves and people stay in control of edge cases.
- Trust & oversight. Guardrails, transcripts, and summaries that make every AI interaction visible, auditable, and easy to take over.
Approach
I worked end-to-end — from discovery research through pixel-level UI and dev-ready prototypes — partnering closely with product, engineering, and data. Components were built on the team's token-based design system, and I used a code-based prototyping loop to align stakeholders quickly on how the agent should behave in real conversations.
Outcomes
- ↓20%Lower call handle times
- ↓50%Fewer missed calls
- 24/7Always-on coverage
Figures reflect impact ranges on the platforms I contributed to. Full case study and process walkthrough available on request.