Free Consulting
One of the pleasant truths of the modern world is that knowledge has become widely accessible. Concepts that once lived inside academic journals or consulting reports can now be discussed openly. The mathematics behind demand classification, for instance, requires nothing beyond school-level arithmetic. Yet many planning organisations continue to treat forecasting as a mysterious art practiced only by specialists.
A useful starting point is understanding intermittent demand.
When Demand Becomes Intermittent
Demand does not behave the same way across all products. Some products sell frequently and in stable quantities. Others sell irregularly with long periods of zero demand between transactions. Planning methods that work well for the first group often fail completely for the second.
Two simple statistical measures help classify demand patterns.
ADI — Average Demand Interval
ADI measures the average number of time periods between non-zero demand observations.
CV — Coefficient of Variation
CV measures the variability of demand relative to its mean.
Using these two values, demand patterns can be categorised.
Smooth demand
ADI < 1.32
CV² < 0.49
These series behave relatively predictably. Traditional forecasting models such as exponential smoothing or ARIMA can often perform reasonably well.
Erratic demand
ADI < 1.32
CV² > 0.49
Demand occurs frequently but varies widely in quantity. Forecasting is still possible, though uncertainty remains high.
Intermittent demand
ADI ≥ 1.32
CV² < 0.49
Demand occurs infrequently but quantities are relatively stable when orders appear. Specialised forecasting approaches such as Croston-type models may perform better here.
Lumpy demand
ADI ≥ 1.32
CV² > 0.49
Demand is both infrequent and highly variable. Forecasting becomes extremely unreliable. In many cases prediction becomes impractical and inventory policies must compensate for uncertainty.
These classifications provide a quick way to understand which products are statistically forecastable and which are not.
Why This Matters
Many planning systems treat every SKU as though it belongs to the same statistical universe. Forecasts are generated for thousands of products using identical models and parameter settings. The result is predictable. Some forecasts perform well while others fail completely.
Demand classification prevents that mistake. Once products are grouped by demand behaviour, forecasting approaches and inventory strategies can be tailored accordingly.
The goal is not to produce perfect forecasts. The goal is to understand the limits of predictability.
Moving Beyond the SKU Mindset
Another advantage of demand classification is that it encourages planners to move beyond a rigid SKU-level perspective. Many insights appear only when demand is analysed across clusters of products rather than individual items.
Modern data warehouses make such analysis straightforward. Products can be grouped according to multiple properties such as volume contribution, revenue contribution, correlation with other items, seasonal behaviour, customer segments, or product associations.
These properties can be stored as dynamic attributes within the planning environment and used as filters for analysis. Instead of examining thousands of SKUs individually, planners can analyse demand patterns across meaningful clusters.
Traditional ABC and XYZ classifications represent only the beginning of this process.
Using the Tools Properly
Most enterprise planning platforms already contain the computational capability required to calculate ADI and CV². In systems such as SAP, simple macros or data warehouse queries can generate these values automatically. Once calculated, the results can be used to create statistical forecasting groups or planning segments.
The technical effort required is modest. The conceptual clarity it provides can be substantial.
Unfortunately, many organisations spend millions on supply chain software while using only a fraction of its analytical capability. Systems become expensive typewriters that produce reports rather than insights.
Designing Analytical Flexibility
A well-designed planning architecture should therefore allow flexible analytical exploration. Product attributes, cluster definitions, correlation indicators, and demand classifications should all exist as configurable properties rather than fixed design elements.
This flexibility matters because analytical questions evolve. As markets change, planners may wish to segment products differently or analyse demand behaviour from new perspectives.
Rigid system design restricts that exploration.
The Role of Expertise
Technology alone does not produce analytical insight. The individuals implementing and operating these systems must understand the domain they are working in. Configuration expertise without statistical understanding often leads to mechanically correct systems that answer the wrong questions.
Businesses should therefore seek consultants and internal analysts who understand both the mathematics and the operational context of planning.
Configuring software is not the same as designing decision systems.
Simplicity Behind the Mathematics
Despite the sophistication of modern analytics, the underlying ideas often remain straightforward. Demand classification using ADI and CV² requires only basic arithmetic. The complexity lies not in the formula but in applying the results correctly within planning processes.
Statistical models can support decision-making. They cannot replace judgement about market behaviour, operational constraints, and business priorities.
The real transition from data to decisions occurs when organisations combine simple analytical tools with disciplined thinking about how demand actually behaves.
And for everything else, increasingly, there is artificial intelligence.
