Machine Learning Forecasting in PLAIO

This article describes PLAIO’s machine learning forecasting methodology and how the model produces forecasts across the portfolio — what information it uses, how it behaves, and what distinguishes it from statistical forecasting.

PLAIO’s ML method uses a model trained on historical data from all products in the portfolio. By learning shared patterns across SKUs, it generates consistent forecasts even when individual items have limited history or irregular demand.

How it works

The ML Forecast model learns from the entire dataset at once rather than one SKU at a time. That lets it recognise portfolio-level patterns and apply them to individual items.

Historical demand is the foundation, but the model also incorporates:

• Product attributes such as category, grouping, or other metadata
• Seasonality
• Trend behaviour

From this combined dataset the model identifies relationships between products, repeated seasonal cycles, changes in demand level, and other structural patterns.

Because training spans the whole dataset, newly launched or low-history items benefit from patterns learned from established SKUs. The model produces stable initial forecasts and avoids the extreme behaviour that comes from fitting to a handful of data points. When underlying demand patterns shift, the model adapts during retraining without manual model adjustment.

The result is a forecast line that behaves consistently across SKUs, reflects shared seasonal and trend patterns, and stays stable even for items without long or regular histories.

ML versus statistical forecasting

The difference is mainly in how information is used.

Breadth of information. Traditional methods look only at each SKU’s own history, which limits pattern detection when data is sparse or irregular. ML forecasting learns from the entire portfolio at once.

Model consistency. A single global model serves all products, so forecast behaviour is uniform across the portfolio. Seasonal patterns, trend interpretation, and sensitivity to recent change follow the same logic for every SKU. Per-SKU methods often behave differently depending on data length and parameter choices.

Maintainability. The global model structure lets PLAIO retrain more efficiently — updates happen in one model rather than many.

Adaptability. If market dynamics shift or structural change appears in the data, the model incorporates it during retraining without SKU-level intervention.

In the Demand Planner

The machine learning forecast is displayed as the DemandML series type alongside the other series:

This lets planners compare forecast perspectives, observe how the ML model responds to changes in historical demand, and see where manual market insight adds context.

Understanding forecast performance

PLAIO evaluates forecast behaviour with two complementary metrics. Error is the magnitude of deviation between forecast and actual. Bias indicates whether forecasts sit consistently above or below actual demand. Together they show how the ML forecast behaves across time and across products.

Performance is typically reviewed by comparing the ML forecast, the market forecast, and the simple benchmark. Those comparisons reveal whether a product is inherently hard to forecast, whether the model is capturing the relevant patterns, or whether market input contains insight not present in the history.

Full detail in Forecasting Approach & Performance Evaluation.