Forecasting and causal inference are distinct but fundamental tasks in data science. While forecasting predicts future outcomes based on history, causal inference explores the "why" behind those outcomes and helps simulate "what if" scenarios. Confusing the two can lead to misleading results.

At Blue Yonder, we encountered a case where a customer's forecasting model predicted demand accurately based on price. However, when they used the model for simulations to explore "what if" scenarios, the results were counterintuitive: lower prices led to lower demand. I will share how we resolved this issue and emphasize the importance of incorporating causal thinking when addressing questions like, "Why did this happen?" or "What if I do X?"

In this talk, I’ll show how to identify common pitfalls, like confounders, when integrating causal inference into forecasting workflows. We’ll also explore Bayesian models, powered by Markov Chain Monte Carlo (MCMC) methods, to bridge the gap between forecasting and causality using the PyMC library on practical examples.

By the end of this talk, you’ll learn how to:

• Consider causal reasoning when building models that forecast well but can also be used for interventions and "what if" scenarios.
• Visualize causal hypotheses with Directed Acyclic Graphs (DAGs) to understand relationships.
• Leverage PyMC to build Bayesian models for testing causal hypotheses and answering "what if" questions.