Forecast Lab — Compare forecasting models

Forecast Lab is an exploratory laboratory for demand forecasting. Upload your historical demand data, choose the models to compare, and the tool runs a comparative backtest — showing you which model performs best on your specific data, with clear and interpretable KPIs.

Available models range from Naive (baseline benchmark) to Holt-Winters (exponential smoothing with trend and seasonality), including Moving Average, Simple Exponential Smoothing, Holt, and Croston/SBA for intermittent demand.

In demand planning, choosing the right forecasting model is one of the most important decisions — and one of the most often made by inertia. Many companies use the same method for every SKU, without ever testing whether it actually outperforms alternatives.

Forecast Lab lets you run that test in minutes: upload data, launch the backtest, see results side by side. You don't have to trust a vendor telling you "our algorithm is the best." You can verify it yourself, on your own data.

1. 1Data upload — upload a file with your historical demand data (required format explained on the page)
2. 2Model selection — choose which models to compare (all or a subset)
3. 3Backtest — the tool splits data into training and test periods, and runs each model on the same data
4. 4Results — dashboard with comparative chart, KPI table (MAE, RMSE, MAPE, Bias), result interpretation
5. 5Exploration — what-if slider to vary the backtest window, side-by-side SKU comparison, educational tooltips on every KPI and model

All computation happens in the browser via Web Workers. Your data never leaves your computer.

In the current version (Phase 1), Forecast Lab uses classical statistical models — not AI/ML models. AI enters in Phase 2 of the roadmap, with machine learning models for comparison.

This is intentional: starting from statistical fundamentals builds a solid foundation of understanding. If you don't know why a Holt-Winters beats a Moving Average on your data, adding an ML model won't help — it'll just give you one more number you don't trust.

• •This is an educational and exploratory tool, not a production forecasting system
• •Models are classical statistical (AI/ML coming in Phase 2)
• •Result quality depends on data quality — dirty data produces unreliable results
• •Does not handle exogenous variables (promotions, events, weather) — based on time series only
• •Not designed to replace a corporate demand planning system, but to understand how models work and which one fits your data best

• •Demand planners who want to quickly test which model works best on their data
• •Supply chain managers who want to know if their current forecasting method is really the best option
• •Students and professionals who want to learn forecasting hands-on, not just in theory
• •Anyone with historical demand data wondering: "which model should I use?"