UX
Turning a Data Dashboard Into an AI Conversation
A data-heavy analytics application, rebuilt as a conversation instead of a maze of charts and filters.
System overview
The old dashboard grid, retired. A single input replaces the maze of filters and chart pickers users had to learn before they could ask a question.
Conversation states
The core interaction patterns: how the system asks a clarifying question, handles ambiguity, and decides when an answer needs a chart versus a sentence.
Verification surface
Charts didn't disappear — they moved. Now they appear only when a visual actually helps confirm an answer, nested inside the conversation instead of leading it.
Response formatting
A documented system for turning a raw data answer into text, table, or chart, so every response feels considered rather than generated on the fly.
Challenge
The existing dashboard had grown into dozens of charts and filters. Any business question that didn't map to an existing chart became a backlog ticket, and new users needed real onboarding just to find the right view.
Process
Design shifted from screens to conversation flows — intents, clarifying questions, fallbacks, formatting rules — tested against real support tickets and sales questions before anyone wrote production code.
Research
The pattern across support tickets was consistent: people weren't looking for a chart, they were looking for an answer. The chart was always a means, not the goal — and the interface had confused the two.
Design
The chat interface became the front door. Charts, tables, and plain text became interchangeable response formats, and the system picked whichever the question actually needed.
Outcome
New users reached their first answer without training, and time-to-insight dropped from minutes of clicking around to the time it takes to type a sentence. That pattern is now the template for other legacy dashboards.