Tracking AI token usage.
One clear view of AI usage across a Fortune 1 enterprise, for associates using the tools, leaders running teams, and the executives investing in the future of technology.
Everyone was using AI. No one could see it.
The company was investing in AI everywhere: Copilot, ChatGPT, Gemini, and a growing family of homegrown tools. But the picture was fragmented across ten dashboards. Leaders couldn’t tell which tools earned their cost, which teams built the habit, or where licenses sat dormant.
I led design end to end: the research, the information architecture, and every dashboard, working daily with PM and data engineering to turn ten inconsistent sources into one trusted view.
The hard part wasn’t the charts.
Three audiences needed the same data at very different altitudes: associates checking their own habits, leaders comparing teams, executives steering spend. And a dashboard about people’s work is easy to get wrong. Usage gets read as performance; incomplete data gets read as truth. Most of my design decisions were really policy decisions: what each role sees by default, where drill-down stops, and how honestly every chart admits what it doesn’t know.
Start with the questions.
I interviewed leaders, associates, and analysts, then audited what each tool could actually report.
Three altitudes, one system.
Personal, team, and enterprise views of one system, every metric backed by a written spec.
Cards, tested with leaders.
Consistency heatmaps, adoption funnels, dormant-license alerts, and token meters.
Pilot first, then expand.
V1 launched with three tools and pilot orgs, then widened coverage with every release.
A dashboard about people’s work is one careless default away from becoming surveillance. The privacy guardrails weren’t a constraint on the design; they were the design.
Answers that took a data pull now take a filter.
V1 is live. The first tools are unified, pilot orgs are on real data, and coverage is expanding release by release. We committed to three measures up front: how much of the enterprise’s AI activity the platform captures, and whether leaders and associates keep coming back on their own.
The most valuable artifact I made wasn’t a screen. It was the metric spec. Agreeing on what a number means, who can see it, and what it should never imply did more for the product than any visual decision. I’d start there again.