Sales & Inventory Forecasting with sktime: A Robust Time Series Toolkit

In the fast-moving world of retail and quick-commerce, staying ahead of demand is tough - especially when you're dealing with daily sales data across hundreds of items and stores. In this talk, we’ll show how the open-source sktime library can help you build forecasting pipelines that are not just powerful, but also modular and easy to reason about.

We’ll walk through real examples where hierarchical forecasting (like item → store → region) and probabilistic methods (to capture uncertainty) that are essential. You'll see how different approaches - from Poisson-based models to non-parametric methods like ENBPI compare when predicting item-level demand.

Along the way, we’ll cover common traps (like confusing actual orders with sales data), and how to bring in extra signals like promotions or weather to improve your forecasts. Whether you're new to time series or looking to apply it in a real-world setting, you’ll leave with tools you can use.

* Understand how to build and evaluate hierarchical forecasting pipelines in sktime.

* Learn the difference between parametric and non-parametric probabilistic forecasters.

* See how to include external factors like promotions or weather in forecasting models.

* Recognise common pitfalls in retail forecasting and how to avoid them.

* Get hands-on ideas for applying sktime to real-world datasets