Designing a conversational interface that turns complex travel data into instant, trustworthy insights, without adding another dashboard.
✦ Try Iris, live ↗ITILITE helps businesses book, manage and optimise employee travel. Corporate travel teams own a sprawling, opinionated dataset, yet getting to a single answer required exporting spreadsheets, building dashboards manually, or asking an internal analyst to play archaeologist.
The problem wasn't missing data. It was the distance between the question and the answer.
The temptation was to ship another analytics page. We resisted.
Instead, I designed Iris as an AI-first analytics layer that lets admins self-serve insights in natural language, woven into the existing reporting and insights workflows rather than sitting alongside them as a new surface.
The work focused on three things: progressive guidance, clear guardrails, and trust-building patterns, so users could explore data confidently, and the system could decline confidently when it couldn't.
Four moves that shaped how Iris feels, less an AI feature, more a quiet member of the team.
Designing clear boundaries, error states, and fallback copy was as important as the happy path. Trust is earned in the failures.
The questions shown to users significantly influenced how they explored, and ultimately trusted, the system.
AI features gain adoption faster when they feel like part of an existing workflow, not a separate product on the side.
Decisions around navigation, limits, and data scope had more impact than visual polish alone. AI is a system problem.
Designing Iris was less about adding AI to the product and more about reframing how users access and reason about their data.
The work required balancing innovation with trust, flexibility with constraints, and discovery with simplicity, treating AI as a system design problem, not just a UI feature.