Expedia Comms Manager
A fully agentic communications system. Designed end-to-end — from pipeline architecture to high-fidelity prototype.
- annual touchpoints
- 11.9Bannual touchpoints
- reduction in time-to-market
- 90%+reduction in time-to-market
- projected revenue & cost impact
- $100M+projected revenue & cost impact
- operational efficiency gains
- 40%+operational efficiency gains
- Who
- Expedia, Vrbo, Hotels.com
- What
- Invented an agentic, multi-channel communications system
- Role
- Staff Experience Designer & Strategist (Lead)
- Industry
- Travel
01
The problem
Sending a single customer communication across Expedia, Hotels.com, and Vrbo took 4–6 weeks. Most of that time wasn't spent improving the content — it was eaten by design review, legal sign-off, and compliance routing.
The deeper issue: channels were managed as standalone artifacts instead of one coherent customer journey.
02
The invention & architecture
Universal Messaging Platform (UMP) Comms Manager is a fully agentic pipeline. It takes a natural-language prompt or a JIRA ticket URL and generates all four channel formats — email, SMS, push, and in-app — simultaneously.
Each output is scored for quality and compliance by Helm, an internal evaluation model originally built by the content design team, then auto-routed to the right reviewers (Legal, Design, Compliance) based on its risk tier.
“The goal wasn't just to make content generation faster. It was to create a system that could look at an entire customer communication journey and ask whether we were actually doing right by the traveler.”
03
The build
This is a production-architecture multi-agent system, not a clickable mockup. All four channel agents generate in parallel rather than sequentially. I built it with Claude Code and Cursor.
The hardest part wasn't the agent architecture — it was integrating with existing Expedia systems that were never designed for this kind of automation.
- LangGraph 1.0.2 for agent orchestration
- Next.js 16 + React 19 frontend
- OpenAI GPT-4o as the underlying model
- CopilotKit's AG-UI protocol for the human-AI interaction layer
- Zustand for state management
- Turborepo + pnpm workspaces for the monorepo
04
The vision
The project was only weeks away from its internal debut when a re-org ended the work before it found its audience. It's included here for what it demonstrates, not as a final shipped product.
Short term, it cuts a 4–6 week process to roughly one to two days while improving consistency and compliance. Longer term, it points to genuine one-to-one personalization at scale — across every channel, for every traveler.