Manufacturing AI training
Demand forecasting
A practical course on forecasting future demand from sales and production history with AI time-series models, then optimizing production planning, inventory, and purchasing.
What is demand forecasting?
Demand forecasting applies AI to multivariate time-series data, including sales history, seasonality, and market factors, to predict future demand and optimize production planning, inventory, and purchasing.
Multivariate time-series analysis
Beyond sales history, the model takes in seasonality, promotions, and external economic indicators, which is what makes the forecast usable in practice.
Hierarchical forecasting across many SKUs
Thousands of SKUs are sorted by ABC classification, and the pipeline picks the best model, deep learning or statistical, for each item automatically.
Wired straight into production planning
Forecasts flow into APS and ERP automatically, so replanning takes less time when demand moves.
What you'll learn
The core material this course covers.
- Collecting and preparing forecast data (seasonal decomposition, outliers, external variables)
- Classical models (ARIMA, exponential smoothing) vs. deep learning (LSTM, Transformer, N-BEATS)
- Time-series deep learning forecasting workflows
- Multi-item and hierarchical forecasting strategy
- Forecast accuracy metrics (MAPE, RMSE, bias)
- Integration with APS (advanced planning and scheduling)
Curriculum
What we cover
- 1
The business case for forecasting, and the shape of the data
Look at what forecast misses actually cost, then break time-series data into trend, seasonality, and residual components.
- 2
Hands-on statistical forecasting models
Work through ARIMA and Holt-Winters in statsmodels and visualize the forecast on a simple dataset.
- 3
Deep learning demand forecasting
Walk through LSTM and Transformer architectures, then forecast on real data using Prophet and N-BEATS.
- 4
Multi-item forecasting and pipeline design
Assemble a Prophet and LightGBM ensemble with an automatic retraining pipeline, and evaluate it on MAPE and RMSE.
- 5
Hands-on integration with production planning
Build the interface that pushes forecasts into production planning and APS, and define how to run it automatically.
How teams put this to use
Deep learning demand forecasting from production and shipment data
Production and shipment history trained a deep learning model into a single tool that predicts near-term demand and equipment condition together.
Demand forecasting for battery materials
During a period of volatile EV demand, monthly forecasts for cathode and anode materials allowed purchasing lead times to be managed ahead of the swing.
ABC-classified forecasting for high-mix, low-volume manufacturing
Deep learning for A items and statistical models for C items secured forecast accuracy and operational efficiency at the same time.
Expected outcomes
What changes once your team puts this to work on site.
- Getting ahead of demand swings reduces both excess inventory and stockouts.
- Production planning takes less time to put together.
- Supply chain visibility makes delivery commitments more reliable.
- Building in external variables makes the forecast usable in real decisions.
Demand forecasting — FAQ
You can, but the amount of data changes which model to pick. With a short history, statistical models such as ARIMA and Prophet are more stable, and deep learning can fill the gap through transfer learning that trains on similar items alongside yours.
Major ERP systems such as SAP and Oracle accept planning data through an API or a file interface. The course covers forecast output formatting and APS integration during the hands-on sessions.
We'll design the training around your operation
Schedule, group size, and curriculum detail all get set with you.