Forecast baseline

As explained in Demand Forecasting Best Practices

A baseline forecast is a forecast produced before teams apply additional human judgments or modifications.

A forecast baseline is the demand forecast generated by forecasting software or a model before people modify it. It provides the starting point for planning decisions such as ordering, production, and inventory movements, while keeping routine forecasting work automated. section 1.2.4 section 12.3

Build the baseline around the decision it must support: choose the right forecast horizon and aggregation level, use appropriate demand and driver data, and monitor accuracy and bias. A highly accurate number alone is not useful if it is produced at the wrong level or cannot support the intended decision. section 1.2 section 1.3

A useful baseline model should:

• Automate most forecasts so planners can concentrate on exceptions rather than produce every forecast manually. section 1.2.4
• Use relevant demand drivers, such as promotions, prices, shortages, or weather, when the available data and method support them. section 1.2.4 section 1.2.5
• Use time-series models for limited data, and consider machine learning when larger datasets or demand drivers can capture more complex relationships. section 17.0
• Serve as a stable reference for measuring whether subsequent human changes improve or worsen the result. section 1.2.5

Improving the starting point without replacing judgment

Suppose an existing baseline has 45% MAE and planners improve it to 41%. Replacing it with a machine-learning baseline at 41% can free planners to focus on new products or client information, allowing the final forecast to reach 39%. The goal is not to eliminate review, but to reserve it for insights unavailable to the model. section 15.3.2 section 16.1

Review the baseline only when people possess information the model lacks, such as a client’s expected sales change, a new product, an unmodeled price increase, or a major market disruption. Do not repeatedly override trends or seasonality without new evidence; fix the model when it misses obvious recurring patterns. section 16.1

Track forecast value added (FVA) across each handoff from baseline to final forecast. Comparing accuracy and bias after each team’s adjustments creates accountability, reveals harmful changes, and helps prioritize review effort where people are most likely to add value. section 1.2.5 section 12.3

Figure 1.2 5-step framework for demand planning excellence

Use the five-step framework to place the baseline between data and metrics work and the human review process. section 1.2

How Demand Forecasting Best Practices covers it

How deep it goes

Defines the baseline operationally and compares time-series, predictive, and machine-learning approaches without detailing their construction.

Examples

Illustrates a 45% MAE baseline improved by planners, then a 41% machine-learning baseline improved to 39%.

What it assumes

Starts from the planning role of forecasts while requiring decisions about horizon, aggregation, data, metrics, and model choice.

What you can do after

Apply it via forecasting models and FVA to automate routine forecasts and target human review where it adds value.

How it connects

Connects baseline forecasts to demand planning, forecasting models, demand drivers, forecast value added, accuracy, and bias.

liveBookmanning.com

Read Demand Forecasting Best Practices on liveBook Get the book at manning.com

Related concepts

• judgmental forecastused by · A judgmental forecast is a prediction of future demand that incorporates human assessment rather than relying solely on an automated forecasting model.
• time-series modelproduced by · A time-series model represents or analyzes observations ordered over time to characterize temporal patterns or dependencies and support inference or prediction.
• language modelproduced by · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• forecast value added frameworkmeasured by · The forecast value added framework is a method for evaluating each step or contributor in a forecasting process by comparing its forecast performance with that of the preceding step.
• machine learningproduced by · Machine learning is a field of computational methods in which systems learn patterns or relationships from data to make predictions, decisions, or other outputs.

This explainer was generated by Manning's knowledge graph from the text of Demand Forecasting Best Practices. The section links point to the parts of the book it draws on.