Overview
A dashboard drowning in its own data
A semiconductor tools team had telemetry for everything and insight into nothing. Engineers scrolled through endless tables to answer one question: is this chip batch healthy? Alerts fired constantly, so they were ignored constantly.
Over six weeks I researched their workflow, prototyped a status-first dashboard, and delivered a component system their front-end team built from without a single redesign round.
Challenge
Signal buried under noise
Shadowing sessions showed engineers kept three tabs open and a spreadsheet on the side just to triangulate one verdict. The old UI treated every metric as equally urgent, which meant none of them were.
- Alert fatigue — hundreds of daily warnings, almost all benign.
- No shared vocabulary — each shift described batch health differently.
Approach
One verdict, then the proof
Every screen now opens with a single batch verdict — healthy, watch, or act — followed by the three metrics behind it. Drill-downs kept the full depth power users needed, but the default view answers the morning question in seconds.
Severity rules were redesigned with the on-call team, cutting daily alerts by an order of magnitude and restoring trust in the ones that remained.
Gallery
Screens and sessions
Deliverables
What the team kept
- Research report — twelve shadowed shifts distilled into five workflow findings.
- Clickable prototype — the full dashboard flow, tested with nine engineers.
- Component library — charts, tables and status patterns documented for reuse.
Related work
More case studies
Have data users can't read?
Let's turn your densest screens into your clearest ones.