An ALIS LABS Research project
Forecast & Budget
Machine-learning demand forecasting connected to a full corporate budget.
Context
Budgets are usually a spreadsheet disconnected from demand forecasting. This research tests a single flow: from transactional data to the P&L, through evaluated ML models.
Challenge
- Hierarchical forecasting by product, channel and month.
- Internal assumptions and external indicators in one model.
- Executive output that finance can audit.
Solution
A versioned package that starts from simulated ERP data, builds a feature store, trains forecasting models and feeds a budget engine with P&L, cash flow, working capital and taxes.
Architecture
- 01100% synthetic ERP data
- 02Public external indicators (inflation, rates, retail)
- 03Models from seasonal baseline to gradient boosting, hierarchical
- 04Evaluation with WMAPE, MAPE, bias, RMSE and FVA
- 05Outputs in Excel, dashboard and a BI base
Governance & security
- Versioned evaluation metrics per model
- No real company data
Current state
- Research project