03 · Siemens · UX Designer · 2018–2020
The $20M Spreadsheet
Vehicle safety testing at one of the world's largest engineering companies ran on spreadsheets nobody fully trusted. I redesigned the workflow as one governed data system for 50,000+ engineers. The screens are under NDA. The story isn't.
Role: UX design. Data workflows, accessibility, adoption
The problem
Safety test data is the kind of data that ends up in court. And it lived in spreadsheets: copied between teams, versioned by hand, reconciled by whoever noticed the numbers didn't match. Engineers spent their time confirming which value was current instead of testing vehicles. On the books, that habit had a price: a $20M process inefficiency.
The real problem sat underneath the operational one. Nobody distrusted the engineers. They distrusted the copies.
The decision
The interface was never the product. The record was. So the redesign treated data like the safety-critical asset it is: captured once at the source, validated by rules instead of by memory, published as a single governed record that every team reads and none quietly forks. When the current number is the easy number to find, re-checking stops being a job.
Accessibility wasn't a checkbox pass at the end. A system used by 50,000+ engineers includes engineers who don't see, hear, or move the way default UI assumes, so WCAG 2.1 AA was a design constraint from the first wireframe, and it made the system plainer and faster for everyone else too.
exhibit a · from the internal case study · specifics inked out
Test results for were maintained in and reconciled before every . The redesigned workflow moved capture into , with validation at entry and a single published record for downstream teams.
Result: the reconciliation step, and the attached to it, was retired.
18 months early
Working inside that data every day showed where it wanted to go next. At a 2019 Siemens hackathon, I prototyped an ML feature that predicted outcomes from the very records the new system was making trustworthy. It placed top 5, and made the case for machine learning 18 months before it reached the official roadmap. It was the first time I designed for a machine's judgment instead of a person's, and it hasn't stopped being the job since.
What happened
The $20M inefficiency came off the books. 50,000+ engineers got one current number instead of a folder of candidates, in a system built to WCAG 2.1 AA. And I left with the lesson that's run under everything since: big companies don't have design problems. They have trust problems wearing design clothes.