Demand Variability: Why Your Forecast Moves

Two SKUs sell 50 a day on average. One sells exactly 50 every day; the other swings between 5 and 200. Same average, same monthly total, and they are absolutely not the same buy. The number that tells them apart is variability, and it's the most under-used number in inventory planning.

Demand variability is how much actual demand swings around its average from period to period. You measure it with the coefficient of variation (standard deviation ÷ mean). High variability means a forecast is harder to trust and forces you to hold more buffer stock.

Key takeaways

• The average is a liar when it stands alone: two SKUs with identical averages can need completely different stock strategies.
• The coefficient of variation puts the swing on one scale: standard deviation divided by mean, comparable across SKUs of any size.
• Variability sets the forecastability ceiling: some SKUs simply cannot be forecast tightly, and knowing which ones changes where you spend effort.
• Every safety-stock decision is a variability decision: the swing you measure here is the input the buffer formula prices in.

What is demand variability?

Demand variability is the size of the swings in demand around its average. Steady demand has low variability; spiky demand has high variability. It's a property of the product and its buyers, not a flaw in your planning, and measuring it is the first step in deciding how much protection each SKU needs.

Average alone hides the swings

Planning from the average is planning for a week that may rarely happen. The steady 50-a-day SKU meets its average almost daily; the 5-to-200 SKU almost never does. Everything a planner does downstream (forecasting, buffers, reorder timing) behaves differently across that gap, which is why two SKUs with identical totals can deserve opposite treatments: lean and automated for one, buffered and watched for the other.

How do you measure demand variability?

The standard measure is the coefficient of variation (CV): standard deviation divided by mean. Dividing by the mean is what makes it useful, because it puts big and small SKUs on one comparable scale. A raw standard deviation of 30 means chaos for a SKU averaging 20 a week and near-silence for one averaging 2,000.

Coefficient of variation as the measure

Take the two SKUs from the hook, measured weekly. SKU A sells 350 a week with a standard deviation of 25: CV = 25 ÷ 350 ≈ 0.07. SKU B also averages 350 but with a standard deviation of 280: CV = 280 ÷ 350 = 0.80. One number now says what a year of gut feel was hinting: B is over ten times swingier than A, and every planning rule they share is wrong for one of them.

What high vs low CV looks like in practice

• Low CV (steady): replenishment staples, subscription-like repeat purchases, core basics. Forecast tightly, run lean buffers, automate confidently.
• Moderate CV: most of a typical catalog. Standard forecasting works; buffers do real work.
• High CV (spiky): trend items, sizes at the edge of the curve, promo-driven products. Forecasts carry wide error bands, and buffers or responsiveness carry the load.

As a working shorthand, planners often treat a weekly CV under roughly 0.5 as comfortably forecastable and anything above 1.0 as demand that needs a different playbook, but calibrate those lines on your own catalog rather than importing them.

Why should planners care about variability?

Variability decides how forecastable an item is and how much buffer it needs. It's the input behind every safety-stock decision, and it's the honest explanation for why some SKUs miss forecast no matter how good the model gets.

Forecastability: some SKUs can't be forecast tightly

A high-CV SKU isn't a forecasting failure waiting to be fixed; it's a product whose demand genuinely arrives in swings. The mature response is to stop demanding precision where it can't exist and manage the risk instead: wider ranges, faster reorder loops, more buffer per unit of demand. Effort spent squeezing error out of a CV-0.07 staple is usually wasted; the same effort on classifying and protecting the high-CV tail pays immediately. (When variability tips into mostly-zero weeks with occasional bursts, it's a different animal entirely: intermittent demand.)

Variability is an input to safety stock

The swing you measure here is exactly what the safety-stock formula prices in: bigger swings, bigger buffer, for the same service level. This page deliberately stops at measuring; the buffer math that consumes the measurement lives in the safety stock formula. One warning worth carrying there: measure variability on real demand where you can, because stockout weeks read as calm (zero sold) when they were actually your spikiest demand going unserved. And when your order swings look wilder than your customer swings, that's usually not your customers at all: it's the bullwhip effect amplifying them.

Frequently asked questions

What's a good coefficient of variation for demand?

Lower is easier, but there's no universal "good." Working shorthand: weekly CV below about 0.5 forecasts comfortably, 0.5 to 1.0 needs real buffers, and above 1.0 the SKU deserves a different playbook (wider ranges, faster reordering, or intermittent-demand methods). Calibrate the bands on your own catalog; the ranking across your SKUs matters more than the absolute numbers.

How is demand variability different from the bullwhip effect?

Variability is how much real customer demand swings on its own; you measure it per SKU with the coefficient of variation. The bullwhip effect is amplification added by the supply chain, where ordering behavior turns small real swings into large artificial ones upstream. You plan buffers for the first and change ordering behavior to dampen the second.

Which products have the highest demand variability?

Trend and fashion items, new launches, promo-driven products, seasonal SKUs measured across their whole year, and edge-of-range variants (unusual sizes, niche colors). Steady repeat-purchase staples sit at the other extreme. Category is destiny only partly: two products in one category can differ wildly, which is why measuring per SKU beats assuming.

Can you reduce demand variability or only plan for it?

Some of it is genuinely yours to reduce: smoothing promotions instead of lurching between spikes, steadier pricing, and better launch pacing all calm self-inflicted swings. The remainder is real customer behavior, which you plan for rather than fight: buffers, responsiveness, and honest forecast ranges scaled to each SKU's measured CV.

Does high variability always mean more safety stock?

For the same service level, yes, more swing demands more buffer. But the smarter response isn't always maximum buffer: sometimes it's a faster reorder loop (shrinking the window the buffer must cover), a lower service-level choice on a low-margin item, or accepting stockout risk deliberately. Variability sets the price; you still choose what to pay.

How does variability affect forecast accuracy?

It sets the floor under your error. A model can only be as precise as demand is regular, so high-CV SKUs will post larger errors than staples no matter the method. That's diagnostic gold: compare each SKU's forecast error to its variability, and you'll see which misses reflect a fixable model and which reflect the product's nature.