Course Objective & Scope
Generic time-series methods struggle when confronted with thousands of SKUs, stores, promotion calendars and sales censored by stockouts. This programme addresses uncovering true demand from censored sales, price and promotion effects, new-product introductions, Croston/TSB methods for slow-moving items and global cross-series models with LightGBM.
Hierarchical reconciliation across product and geography trees (hierarchicalforecast) and quantile forecasts connected to service-level safety stock are practised hands-on. Participants build a demand forecasting pipeline evaluated with WAPE, bias and Forecast Value Added (FVA).
Who Should Attend?
- Data scientists in retail, FMCG, manufacturing and logistics
- Demand planning, S&OP and supply chain analytics teams
- Inventory and category management analysts
- Prerequisite: Python, pandas and basic forecasting knowledge
Key Takeaways
- Detect stockout periods and adjust sales history closer to true demand
- Build global and local models across thousands of SKUs with LightGBM and statsforecast
- Model intermittent and slow-moving demand with Croston, ADIDA and TSB
- Reconcile forecasts coherently across category and location hierarchies
- Connect quantile forecasts to target service levels and safety-stock rules
Curriculum
From Sales to True Demand: Data Preparation
- Stockout censoring, returns and outlier-day adjustments
- SKU segmentation: smooth, erratic, intermittent, lumpy (SBC)
Global ML Models and Promotion Effects
- Cross-SKU LightGBM models with price and promo features
- Cannibalisation, halo effects and new-item cold start
Intermittent Demand and Spare Parts
- Croston, SBA, TSB and ADIDA approaches
- Error metrics suited to zero-heavy series
Hierarchical Reconciliation and S&OP Alignment
- Bottom-up, top-down and MinT reconciliation
- Aligning operational item plans with financial budgets
Probabilistic Forecasts and Inventory Link
- Quantile regression, WAPE, bias and Forecast Value Added (FVA)
- Lead-time demand distribution and safety stock
Other Courses in This Category
-
Bridging predictive models and mathematical optimisation: LP/MILP formulation, capacity, scheduling, routing and budget/price allocation with OR-Tools and Pyomo, plus stochastic and contextual decision systems under uncertainty.
- Live Course
- 5 modules
-
Mastering tabular gradient boosting with XGBoost, LightGBM and CatBoost: tree-growth strategies, native categorical and missing-value handling, ranking/quantile/custom objectives, monotonic constraints, Optuna and fast inference.
- Live Course
- 5 modules
-
Automated model selection, hyperparameter search and multi-layer stacking with AutoGluon, FLAML and Optuna, covering compute budget management, leakage checks and moving from rapid baselines to production.
- Live Course
- 5 modules
-
Glass-box models (EBM, GAM), TreeSHAP/KernelSHAP, ALE, counterfactual explanations (DiCE), uncertainty quantification and regulatory compliance for producing auditable, actionable explanations in high-stakes decisions.
- Live Course
- 5 modules
-
Moving from correlation to causation with DAGs, propensity scores, Double Machine Learning (EconML/DoWhy), heterogeneous treatment effects (CATE) and Difference-in-Differences to measure price, campaign and policy impact.
- Live Course
- 5 modules
-
Collaborative model training across institutions or edge devices without centralising raw data: Flower framework, FedAvg/FedProx, non-IID data, differential privacy (Opacus) and secure aggregation.
- Live Course
- 5 modules
Get in touch
