Building Real-Time Production Dashboards That People Actually Use

I have built a lot of dashboards. The ones that fail have one thing in common: they were designed around the data instead of around the decisions. Early in my career I would build dashboards that showed every metric I could extract from the database. Yield by wafer, yield by lot, yield by test insertion, bin pareto, parametric distributions, tester utilization, test time breakdowns. Comprehensive, technically impressive, and mostly ignored. The dashboards that actually get used are the ones designed around a specific question someone needs answered daily. Design for Decisions, Not Data Before building anything, I ask: "What decision will you make differently based on this dashboard?" If the answer is vague, the dashboard will be vague. If the answer is "I need to know which test house to send the next lot to based on current yield performance," then I know exactly what the dashboard needs to show. For our global test house dashboard, the primary decision was sourcing allocation. Operations needed to compare real-time yield, test time, and cost across five outsourced test facilities worldwide. The dashboard shows a ranked comparison with trend lines, not a data dump. Architecture That Scales Our data flows from UltraFlex test systems through KQL ingestion pipelines into SQL databases. PowerBI connects via DirectQuery for real-time metrics and Import mode for historical aggregations. DAX measures handle the complex yield calculations that account for multi-insertion test flows and retest logic. The trick is getting the data model right. A well-designed star schema with proper fact and dimension tables makes every downstream report fast and maintainable. A poorly modeled dataset means every new dashboard page requires custom SQL workarounds. I maintain the data architecture underneath: the relational database schema, the ETL pipelines that clean and transform raw test data, and the refresh schedules that keep everything current. The Metrics That Matter For semiconductor production, the metrics that drive action are surprisingly few: First-pass yield by product and test insertion. This is the health indicator. Yield trend over time with control limits. This catches excursions. Bin pareto showing which failure bins are growing. This directs debug effort. Test time distribution. This catches tester or handler issues. Side-by-side facility comparison. This drives sourcing decisions. Everything else is supporting detail that should be one click away, not on the front page. The result: dashboards that operations managers check before their morning coffee, because the information directly drives what they do that day.