What Is Demand Variability?
Demand variability is the standard deviation of daily or weekly demand for a SKU. It quantifies how much your actual sales bounce around the average. A SKU that sells exactly 30 units every day has zero demand variability. A SKU that swings between 5 and 60 units day-to-day has high demand variability, even if the average is also 30.
This number is the σ in the safety stock formula. Get it right and your buffer is sized correctly. Get it wrong and you either stock out or carry excess for months. Most FBA sellers either pull a number from a stale spreadsheet or skip the calculation entirely and use a flat percentage of lead time demand. Both approaches misprice risk.
Demand variability matters most when paired with long lead times. A 5-day lead time forgives sloppy variability estimates because the buffer required is small. A 60 to 90 day ocean freight lead time amplifies any error in σ by the square root of lead time, which means a 20% error in your daily demand variability becomes a much larger absolute error in lead time demand variability.
How to Calculate Demand Variability
The formula for sample standard deviation:
σ = √( Σ(xᵢ - x̄)² / (n - 1) )Where x_i is each day's sales, x̄ is the mean daily demand, and n is the number of days observed. In practice, you do not calculate this by hand. In Excel: =STDEV.S(range). In Google Sheets: =STDEV(range). In Python: numpy.std(data, ddof=1).
For inventory planning, what you need is the standard deviation of demand during lead time, not just daily. Scale it:
σ_dLT = σ_daily × √(lead time in days)This square-root scaling assumes daily demand is independent across days, which is approximately true for most FBA SKUs once you remove obvious anomalies. It is the most important step that planners forget.
Worked Example: Demand Variability for a Real FBA SKU
You sell a $26 ASP supplement at average velocity. Your last 12 weeks of organic daily sales (Lightning Deal days removed) look like this sample:
Week 1 daily units: 18, 22, 19, 24, 16, 21, 23
Week 2 daily units: 20, 17, 25, 22, 19, 18, 24
Week 3 daily units: 21, 23, 20, 17, 26, 22, 19
... (continuing for 12 weeks total = 84 daily observations)
Step 1: Calculate the mean. Average daily demand x̄ = 21.0 units/day.
Step 2: Run STDEV.S on the 84 daily values. Result: σ_daily = 2.7 units/day.
Step 3: Calculate coefficient of variation. CV = σ / mean = 2.7 / 21.0 = 0.13. This is steady demand.
Step 4: Scale to lead time. Lead time = 75 days (50 days production + 25 days ocean freight + Amazon receiving).
σ_dLT = 2.7 × √75 = 2.7 × 8.66 = 23.4 units.
Step 5: Use it in the safety stock formula. At a 95% service level, Z = 1.65.
Safety Stock = 1.65 × 23.4 = 39 units.
If you had instead used the rule-of-thumb "20% of lead time demand," your buffer would be 0.20 × 21 × 75 = 315 units. That is 8x the math-driven answer. At $9 landed cost, that's an extra $2,484 in cash tied up per SKU. Now multiply that across 200 SKUs.
FBA-Specific Context for Demand Variability
Three FBA realities make demand variability harder to measure than a textbook example suggests.
1. Suppressed listings. If your listing was suppressed for compliance reasons for 4 days, those days show zero sales but are not real demand-side variability. Treat them as missing data and exclude them, rather than letting them inflate your σ.
2. Stockouts hide true demand. Days where you ran out of inventory show artificially low demand. Use a censored demand approach: cap your data at the days you had stock, and infer the missing demand from comparable periods.
3. Promotional bleed. A Lightning Deal often boosts the next 3 to 7 days through algorithmic ranking. The promo itself ends but the elevated demand persists. Strip not just the promo days but the trailing tail when measuring true demand variability.
Once you've measured σ cleanly, you can calculate safety stock for the SKU directly in our free tool.
Common Mistakes
1. Using STDEV.P instead of STDEV.S. STDEV.P treats your data as the entire population. STDEV.S treats it as a sample, which is correct for inventory planning. The difference is small but real, and the population formula understates variability slightly.
2. Mixing time periods. Calculating σ on daily data and then plugging it into a formula expecting weekly σ produces nonsense. Pick a unit (daily or weekly) and stick with it through the entire calculation chain.
3. Confusing variability with forecast error. Demand variability is the natural noise in actual sales. Forecast error is how far your prediction missed. They are related but not identical, and using one in place of the other gives wrong safety stock numbers.
Once σ is clean, you can calculate your reorder point in our free tool to see exactly when to fire the next PO.
Related Terms
Frequently Asked Questions
How do I calculate demand variability for an FBA SKU?
Pull 8 to 12 weeks of daily sales data for the SKU, strip out promotional spikes and outage days, then run STDEV.S in Excel on the cleaned series. The result is your demand variability in units per day. Multiply by the square root of lead time to scale it to lead time demand variability.
How much demand variability is normal for FBA?
A coefficient of variation (standard deviation divided by the mean) of 0.15 to 0.30 is typical for steady-velocity FBA SKUs. Below 0.15 means very steady demand, often Subscribe & Save heavy. Above 0.40 means erratic demand that may not be normally distributed and needs a different forecasting approach.
Should I use daily or weekly demand variability?
Use daily demand variability if your lead time is measured in days and your sales are smooth across the week. Use weekly demand variability if you have noisy day-of-week effects (heavy weekend skew). Match the time unit of your variability calculation to the time unit of your lead time.
Why does my demand variability change over time?
Real demand variability shifts as your SKU matures, your ad spend changes, competitors come and go, and seasonality moves through the year. Recalculate demand variability monthly using a rolling 8 to 12 week window. Static variability calculations from a year-old data pull are usually wrong by 30 to 50%.
Does promotional activity belong in demand variability?
No. Lightning Deals, coupons, and Prime Day spikes are deterministic events you control, not random variation. Strip them from the data before calculating demand variability, then add a separate promotional buffer if needed. Including promotions in your variability inflates safety stock by 30 to 60%.