Demand variability

Demand variability is the degree to which demand for a SKU bounces around its average from one period to the next. It's the difference between "we sell 100 a week, every week" and "we sell 100 a week on average, but it ranges from 30 to 200."

Both averages are 100. They look identical on a forecast. But the second one needs vastly more safety stock to deliver the same service level — because the variability, not the average, is what causes stockouts.

This is the single most under-appreciated concept in inventory planning. Brands optimise their forecasts (the mean) and ignore the variance. Then they wonder why they keep running out of items they "forecast correctly."

How demand variability is measured

The standard measure is the standard deviation of demand across recent periods (usually weekly):

σ = √(Σ(xᵢ − x̄)² / n)

Where x̄ is the average demand per period and xᵢ is each individual period's demand.

A simpler operational measure: coefficient of variation (CV), which is the standard deviation divided by the mean.

CV = σ / mean

CV is dimensionless, so you can compare a SKU that sells 5/week to one that sells 5,000/week. It's also the measure that XYZ analysis uses to classify SKUs as stable (X), moderately variable (Y), or erratic (Z).

• CV < 0.25 → low variability, very predictable
• CV 0.25–0.5 → moderate variability, normal for most consumer goods
• CV 0.5–1.0 → high variability, plan with care
• CV > 1.0 → erratic, traditional forecasting will struggle

Where demand variability comes from

Three structural causes for most consumer brands:

• Promotions and marketing pushes. A 30% sale lifts a week's demand 4–10×, then drops it below baseline for the next two weeks as customers pull-forward purchases. The mean is fine; the variance is huge.
• Channel concentration. Wholesale orders arrive in lumps — one buyer placing a quarterly PO can equal a month of D2C sales in a single day.
• Seasonality and weather. A heatwave week for sun care, a cold snap week for thermals. Predictable in direction, hard to time exactly.

Plus the ones that aren't really variability — they're forecast bias dressed up as variability:

• New product launches with no history
• Channel expansion (adding Amazon, adding wholesale)
• Price changes that shift the demand curve

Why it drives safety stock

The standard safety stock formula:

Safety stock = Z × σ × √(lead time)

The σ term is demand variability. The Z term is the service-level multiplier. The √(lead time) term scales it up by how long you're exposed.

So safety stock is proportional to variability. Cut variability in half and you cut required safety stock in half. The way most brands try to improve service level is by raising Z (chasing 99% instead of 95%); the cheaper way is to reduce σ.

Reducing demand variability — the levers

Real ones:

• Smooth promotional cadence. Fewer, smaller promotions cause less of the spike-and-trough pattern.
• Channel mix. D2C is smoother than wholesale for the same average revenue.
• Pre-orders and waitlists. Pull demand forward into a known window instead of waiting for the spike.
• Better pre-promo communication with ops. A flash sale that ops doesn't know about is a stockout waiting to happen — that isn't variability, that's an information failure.

Sometimes you can't reduce variability — you have to plan around it. That's what safety stock and dynamic reorder points are for.

Variability vs uncertainty

Subtly different concepts:

• Variability is what we can measure — historical demand bouncing around its mean.
• Uncertainty is what we can't — events outside the historical pattern (a competitor's failure, a viral moment, a recession).

Statistical safety stock handles variability. It doesn't handle uncertainty. For uncertainty you need scenario planning, supplier relationships that flex, and the ability to say no to ranges that look fragile.