Nicolas Vandeput - From forecasting to inventory: machine learning and simulation in action

In this episode, ⁠⁠⁠⁠⁠⁠⁠⁠Ben⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Wim⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ talk to ⁠⁠Nicolas Vandeput⁠⁠, forecasting expert, founder of ⁠⁠Supchains⁠⁠, and author of several books on demand forecasting and inventory optimization.

Nicolas shares how machine learning can be used for demand planning, the typical myths around it, and the (im)possible use of external signals.

We dive into simulating inventory policies with real data, and challenging long-held assumptions in safety stock formulas and min-max strategies.

Key topics covered include:

• Getting started with forecasting and Python
• Why machine learning can be easier than statistical models
• Corporate resistance to open-source tools
• The myth of AI cleaning your data
• Why outlier smoothing may hurt your models
• Forecasting vs. explainability: which matters more?
• The evolving role of planners in ML-driven supply chains
• Inventory optimization through simulation, not theory
• Nicolas’ take on safety stock formulas and biased forecasts
• Outside-in planning: the reality behind external data drivers
• Preview of the first global inventory competition

Key timestamps:

• (01:00) – How Nicolas fell in love with forecasting and supply chain
• (02:00) – Learning Python from scratch and applying it in demand planning
• (04:40) – Why machine learning is easier than many think
• (06:00) – Working with IT: who supports Python in corporate environments?
• (08:00) – Why AI doesn’t clean your master data
• (10:00) – Detecting errors like zero prices or missing sales manually
• (12:00) – Sales input ≠ demand plan: advice on statistical vs. ML models
• (13:00) – The trade-off between explainability and forecast quality
• (16:00) – Decomposable ML: is it real or just marketing?
• (19:00) – LLMs and forecasting: potential vs. hype
• (22:00) – How ML changes the role of planners
• (24:00) – Is ML about cost savings or better planning?
• (26:00) – Forecast-driven inventory policies vs. theoretical methods
• (29:00) – Safety stock: why theory isn’t always applied correctly
• (31:00) – A surprising policy that beat all others in simulation
• (33:00) – Setting inventory targets vs. minimizing cost
• (36:00) – Forecast accuracy benchmarks: what actually makes sense?
• (38:00) – Top priorities for demand planners (and why external data is rarely one)
• (41:00) – Building the first global inventory competition
• (44:00) – Udemy course and future book plans
• (47:00) – Final thoughts and resources

‍