Pricing systems are one of those areas that many data scientists rarely work on and often barely notice, even though prices quietly shape demand, revenue, and customer behavior every day. They may not be as fashionable as recommender systems or large language models, but pricing problems are both intellectually challenging and critical for real business decisions.

This talk aims to demystify pricing analytics and show that it is a rich and accessible playground for data science, without requiring an advanced background in economics. Starting from real world pricing systems, we will explore the kinds of decisions they support and the data science tasks that naturally emerge in this context.

The focus then shifts to one of the most fundamental questions in pricing: price elasticity of demand, namely how demand reacts to changes in price. While the question appears simple, it lies at the heart of almost every pricing decision and is precisely where many standard modeling approaches fail.

In this setting, the classic workflow of taking historical data, training a model, evaluating it, and deploying it can be deeply misleading. Models trained only on past observations risk capturing correlations that do not reflect the true impact of price changes. To reason correctly about pricing decisions, we need a causal perspective that makes assumptions explicit and allows us to think in terms of interventions rather than predictions.

The talk is structured around the following steps:

• Pricing systems and data science tasks: How pricing systems work in practice, which decisions they support, and where data science typically intervenes.
• Why price elasticity is harder than it looks: Confounding variables and bias effects that break standard regression-based approaches.
• A practical introduction to causal machine learning: Core ideas such as treatments, outcomes, confounders, and causal graphs, explained with intuition rather than equations.
• Estimating price elasticity with DoWhy: How to formalize assumptions, express causal questions, and estimate elasticity using the dowhy Python library.

Rather than offering a new “silver bullet” model, this talk emphasizes a shift in mindset: from predicting demand to reasoning about interventions. Attendees will leave with a clearer understanding of when causal methods are needed, what assumptions they require, and how Python libraries can help make those assumptions explicit and testable.