Can machines run businesses? – A practical perspective.
Today’s businesses collect an ever-growing amount of data every day: about their customers’ behaviour, usage and sales of their product, their internal operations, … . For some, like mobile game publishers, their business is entirely digital from the first moment. Does this enable us to hand over their day-to-day running to machines? Or could machines at least recommend what we should be doing every day?
While I will not be able to answer either question adequately during this presentation, I would like to focus on the humble goal of demonstrating what, in the context of machine learning, makes running a business uniquely different from driving cars, playing computer games, or folding proteins.
For this purpose, I will examine the more concrete question: Why did the performance of my business change over time? Using a hands-on example, I will demonstrate the fundamental reasons and practical considerations limiting the usefulness of regression techniques for this problem. In particular, I will look at where predictive models fall short and why you should not ignore questions of causality. Finally, I suggest an alternative approach using unsupervised learning techniques.
Michael Klaput is the CTO and co-founder of Kausa, a decision intelligence platform helping companies to understand why their metrics are changing. Educated as theoretical physicist with a PhD from the University of Oxford, Michael worked as a Quant for large investment banks in London for five years, before starting a career in applying machine learning to solve real-world problems.
Michael’s methods are strongly influenced by his background in differential geometry and topology (string theory) and he keeps trying to bring more of them into the world of machine learning. With Kausa, Michael is focused on developing new approaches that help organisations make decisions based on their data.