Decades of quality know-how, locked in people's heads
For years, catching the cars worth a closer look came down to the judgement of people who knew them inside out. They'd built that know-how over decades. Meanwhile the production line was generating an enormous amount of data that can be used to make that call. Repairs, paint-shop anomalies, line events, the full history of each car sat in the systems and never reached the person deciding what to inspect.
Give them data and a predictive model to put the data to even better use.
How it came together
We went through Škoda Auto's data and systems to see what was there and where the potential was, then agreed the direction together: how to describe a car's whole life in data and use it to catch faults earlier. This is where the Digital Twin took shape.
We replaced reporting that people had been pulling together by hand from PDFs and spreadsheets with automated, centralised reports, giving teams back around four days a month they used to spend preparing data by hand.
On that foundation we built and piloted the model on the line, scoring real cars and putting the results straight into the auditors' hands. That's where the step-change in catch rate came from, and it keeps improving the more it runs. The results spoke for themselves: other quality teams now want predictive models of their own, and the interest has since spread more widely.
A model that scores every car on its risk of a defect
We built the Road Test Predictor, a model that reads each car's data from production, from the quality checks, and from other systems, and works out how likely that car is to carry a defect. That lets the team focus its closer inspections on the cars that most warrant one.
The model learns from a combination of production and quality data, the signals across the build that tend to go hand in hand with a defect. We trained machine learning on that data, and that's what lets the model surface more defects than a person picking by hand ever could at this stage of the quality control process.
The Road Test Predictor lifted this stage's defect catch rate substantially.
This was never about defects slipping through to customers. Škoda Auto has always run several quality checks along the way, and still does, catching issues the model doesn't cover. What the Road Test Predictor changes is timing: it finds defects earlier and closer to where they arise, which optimises how, and at what cost, they get fixed.
Sitting on production data you're not using yet?
Tell us where quality slips through. We'll tell you honestly whether your data can catch it, and how fast.