Demand variability: classifying how demand actually behaves

Separate frequency from size

Variability has two dimensions that are usually conflated: how often demand occurs and how much arrives when it does. An item ordered every week in similar quantities is smooth. One ordered every week in wildly different quantities is erratic. One ordered rarely but predictably in similar amounts is intermittent. One that is both rare and irregular in size is lumpy, and it is the hardest to plan. Measuring the two dimensions separately produces a classification that actually guides method selection.

Method follows pattern

Smooth demand suits conventional time-series forecasting and tight replenishment triggers. Seasonal demand needs a method that models the shape rather than the level, and parameters that move through the year. Intermittent demand breaks ordinary averaging methods, which spread a rare large order into a small constant expectation that is wrong in every period; techniques designed for sporadic demand, or simply ordering against confirmed requirements, work better. Lumpy demand is often best handled by agreement with the customer rather than by statistics.

How much of the variability is self-inflicted

A substantial share of observed variation is created inside the commercial relationship: promotions concentrated into short windows, quarter-end order pushes, minimum order quantities imposed on customers, and pricing that rewards bulk buying. This kind of variability is a policy outcome, not a market fact, and it can be reduced by changing the policy. Distinguishing it from genuine end-market variation is what stops planners from buffering against their own company's incentives.

Aggregation reduces variability, disaggregation restores it

Combining demand across items, locations or periods smooths it, because peaks and troughs partially offset. This is the underlying reason central stock holdings need proportionally less buffer than the same demand served from many separate points. It also means variability measured at one level says nothing definite about another, so the level at which a pattern is classified must match the level at which the planning decision is taken.

Frequently asked questions

How should intermittent items be forecast?

Usually not with conventional averaging, which converts occasional orders into a misleading small constant. Methods built for sporadic demand estimate the size and the interval separately, and where a handful of customers drive the item, asking them for their plans beats any statistical technique.

Does high variability always mean high safety stock?

It pushes in that direction, but the consequence of a stockout and the cost of holding matter too. For an inexpensive item where a short wait is acceptable, accepting the occasional shortage is often the better economic answer than buffering the full spread.

How often should items be reclassified?

On a regular cycle and at life-cycle transitions. New items, items entering a decline phase and items whose customer base has changed shape frequently move between categories, and a classification refreshed once and never revisited quietly misdirects the planning method for years.