Probabilistic Business Analytics

The boom of Machine Learning and AI over the past decade has led to amazing achievements in many fields such as Computer Vision, Natural Language Processing, and Predictive models and fueled the advent of Data Science as a key method in Business consulting. Data Science projects typically focus on conventional Machine Learning methods that solve problems of prediction by finding patterns of association in large datasets, i.e., they answer questions such as: how can I predict the sales of my product based on data observed in the past?

Typical business problems, however, ask questions of interventions, i.e., they seek actions that lead to desired outcomes, e.g., in pricing (How should we set the price of my product to increase sales?), process optimization (How can we optimize our supply chain? ) or customer service (How can we prevent customer attrition?) to name just a few examples. To answer such questions, we need to employ Causal Inference, a method that goes beyond conventional Machine Learning by estimating causal effects between actions (interventions) and outcomes.

Moreover, the influence of uncertainty is often neglected or underestimated in Data Science applications. We think that considering uncertainty in data and business processes is essential, not only because it improves the reliability of analyses, but also because it enables nuanced decisions that calibrate the risk involved in a decision.

We here describe a principled workflow that ties together Causal Inference (causal effects), Probabilistic Programming (statistical modeling), and Bayesian Decision Making (optimal decision making) to tackle business problems with a Data Science approach (Figure 1). These technologies are increasingly gaining traction in the industry.

Follow us on KPMG Ignition Tokyo LinkedIn for the latest news.