Climate analysts were exporting CSVs from Orbital every morning and rebuilding the same five charts in Excel. The dashboard was technically complete but practically unused. Activation sat at 12% after fourteen days.
- Ship in 14 weeks — before a new enterprise contract went live.
- Two frontend engineers, no design-system in place.
- Must support 50M+ rows of satellite telemetry without spinner hell.
I reframed the product from 'analytics tool' to 'morning briefing'. Instead of asking users to construct views, we pre-assembled the five views they were already rebuilding — and made every cell a doorway into the raw data.
- 01
Research
Shadowed nine analysts across three orgs. Watched them copy-paste for 90 minutes a day. That was the insight — the product was a commute, not a destination.
- 02
Frame
Wrote a one-page brief: 'Orbital should answer a question in under ten seconds'. We killed three quarters of the existing surface area on day four.
- 03
Sketch & prototype
High-fidelity Figma prototypes with real data piped in via a mock API. Tested twice a week with the same five users. Iterated 11 times.
- 04
Build
I paired with engineering in the repo — Tailwind, Radix, TanStack Table, Recharts. Shipped the design system alongside the dashboard. 47 components, all documented in Storybook.
- 05
Ship & measure
Rolled out behind a flag to 20% of users on day 70. Graduated to 100% after three weeks of clean dashboards.
“The feature nobody ships is the hardest: restraint. Killing three quarters of the surface area cost more political capital than building the remaining quarter. Worth every meeting.”
