Shaping the Future of Data Analytics at Porsche
I’ve done a 3-months internship at Porsche, advancing the analytics landscape for the e-powered Porsche Taycan. Here are six lessons learned.
Turns out, building a car is even more complex in reality than you would already think.
Every week, there are people
- testing different engine configuration behaviors on the test track,
- supervising the automatic engine inspection before it leaves final production,
- examining vehicle maintenance protocols to keep quality at highest standards,
- developing connected-car services for mobile-app car control,
to give just some examples of everyday activities.
Data is everywhere
As diverse as these tasks are, they‘ve one thing in common: They all require and produce data. Lots of it. Not only processes but also products become more and more digital. On the one hand, the vehicles themselves get „more digital“, impelled by the connectivity trend. On the other hand, digital business models are expedited to offer value-added services, for example by offering integrated mobility services such as parking via mobile app.
Learning 1: There is more data in manufacturing business than one might assume
Interning as a Data Rockstar
Therefore, a main focus of Porsche’s IT strategy is to enable data-driven business. For three months, I’ve interned in a team in Ludwigsburg, South Germany, that goes by the name of „the Data Rockstars“ — yeah, just unofficially, but still…
Formulating a Data Analytics Strategy
While a data scientist can always use matplotlib or ggplot2 to deep-dive into their data, most people outside a Data Rockstars team cannot quickly cobble together some lines of Python. The way to go instead is to rely on a self-service data visualization tool which can be operated by anyone without much training. In the same way, anyone is able to have a look at their numbers in Excel, anyone should be able to have a look at their data in a visual and easy-to-use fashion (in the best case thus abandoning Excel altogether for looking at data).
Even so, just making some charts isn’t enough to really understand the complex phenomena that can be hidden in data. Therefore, tooling is needed that also allows Non-Techies to explore the data for yet unknown „treasures“, for example by use of interaction, what-if analyses, predictions or clustering.
Learning 2: Anyone should be able to make use of their data in an advanced, yet easy-to-use and visual way
But still, just analyzing data is not enough. In order to solve inter-disciplinary problems, you need inter-disciplinary teams which talk through ideas based on data, discuss odd patterns, and communicate and debate findings. So data-driven collaboration becomes a vital capability.
However, once you’ve settled for a tool, this is where the hard work begins in corporations. You’ve got to convince those that decide for or against a tool and of course also those that are supposed to use it. You’ve got to understand which people would prefer a different tool and why. You’ve got to lay out a tactics on how to bring the data analytics culture into the business units. You’ve got to architect the tech landscape. You have to estimate a more or less reliable cost plan. You need to do good old project planning to bring the actions on a timeline, you must — yeah, you get the point…
We have designed a strategy on how to do this and presented it to our CIO who greenlighted the concept. It was a pretty cool experience to be „just an intern“ and yet prepare and participate in a workshop that gives directions for the future of the company.
Learning 3: If there is any company, where data analytics can be successful, it’s a company that puts focus on what is said and not by whom
Also beyond the workshop, I was really astonished at the culture towards interns at Porsche. They are a highly valued part of the organization and its innovation processes. My ideas were always appreciated and my colleagues considered it their personal duty to empower me as much as possible so I could give my best.
Developing Explorative Data Visuals
In the second part of my internship, I got to start implementing the suggested analytics strategy with select business units. Very cool about Porsche is that it is a comparatively small and centralistically managed company. For example, there is exactly one person in charge of managing the world-wide warranty system. There is exactly one person in charge of managing the world-wide vehicle database. Always having one go-to person who pulls all the strings globally for the whole company for a specific topic was beautiful.
Learning 4: Quality-oriented data governance in times of data-drivenness and AI is more important than ever
Therefore, when I started to work my way in the Remote-controlled Vehicle Maintenance Prediction project, there weren’t umpteen people to confer with. Instead, it was just
- one team from Research and Development (R&D),
- one person in charge of the Smart Mobility Backend, and
- our team providing the Analytics Solution.
Vehicle log files including sensor data of a variety of components from the Porsche Taycan are sent over-the-air into a pipeline. The pipeline transforms the data and inserts it into a Hadoop-based storage system. From there, engineers from R&D will analyze that data. Previously, those analyses were performed on data gathered during maintenance appointments in the customer service centers.
And this is where we came in. In order to predict failure or just wear of a certain component, it is not sufficient to consider just a single sensor measurement. Instead, we also had to take into account the frequencies and intensities the component is used, how interrelated components behave, what outside temperature prevails, etc. The use cases are endless, so we settled for a couple of components and designed different Explorative Data Visuals in Tableau.
Developing these Explorative Data Visuals requires a mixture of skills such as
- design (thinking in terms of the user, not the product),
- data savvy (knowing how to analyze numbers and visualize patterns),
- domain knowledge (understanding the field of application).
We started out with some basic analyses and then iteratively developed more complex visuals that would not only answer basic questions but allow to interactively discover relationships you wouldn’t even know are there. Each iteration lived from exchanging ideas between the R&D team on the one hand and us as the data people on the other hand.
Learning 5: If you lock a Data Scientist and a Subject Matter Expert into a room, great things will happen
We handed over a prototype, kind of a Version 1.0, that will be extended over time by the R&D people themselves. Of course, when questions get harder and pushing boundaries is needed again, the Data Rockstars won’t be far…
Once, the Taycan is gliding down the roads, a lot of the assumptions we made about how to derive decisions and predictions from the data, will be put to the test. Also, a lot more of ad-hoc questions will be asked — and that’s a good thing! With the complexity of real-world data, it would be naive to believe that analytics is a linear approach. Learning and understanding is always an iterative process based on a circle of question and answer. All the more, it is important that everyone in a company is enabled to intuitively consult the data that is relevant to them — in a visual way.
Most important learning: Working at Porsche means having a fun and creative job with great people every day!
Thanks to Dr. Andrada Tatu, Martin Mayer, Sandra Schmidl, Jochen Mohr, Martin Ebeling, Fabian Schick, Ben Matheja, Dirk Hemminger, Helge Silberhorn and everybody else from the IT Service, Warranty and Quality team as well as Matthias Schinko, Peter Nerz, Max Weber, Moritz Rabe, Tetjana Aymelek, Tobias Schug, Claudia Feiner and Sebastian Kox for making my time at Porsche a great and memorable experience!
