Demand variability: why last week's average misleads you
Demand variability is how much a SKU's daily sales swing around their own average, day to day and week to week. It is the spread rather than the middle, and it is the reason two SKUs that both average four units a day need very different reorder points.
What demand variability means
Demand variability is the amount a SKU's sales move around their own average over a period, measured in units per day or per week. The average tells you how fast something moves. Variability tells you how wrong any single week can be.
That second number is the one that empties a shelf. An average is built from the normal week. Stock runs out in the abnormal one.
Where it comes from
Variability usually has causes you can name:
- A campaign, a discount or paid spend turned on and off
- A weekday or weekend pattern in your category
- Wholesale or bulk orders landing on top of steady retail sales
- A press mention, a creator post, a feature in someone else's newsletter
- A bundle or a kit that pulls one variant forward for a fortnight
Some of that is pattern you can plan around. Some of it is noise you can only leave room for. Telling the two apart is most of the work.
Two SKUs, one average, different risk
| Day | Steady SKU, units sold | Spiky SKU, units sold |
|---|---|---|
| Monday | 4 | 0 |
| Tuesday | 3 | 0 |
| Wednesday | 5 | 18 |
| Thursday | 4 | 2 |
| Friday | 4 | 0 |
| Saturday | 3 | 6 |
| Sunday | 5 | 2 |
| Week total | 28 | 28 |
| Average per day | 4.0 | 4.0 |
Why last week's average misleads you
Both SKUs above read 4.0 units a day. Put a ten day supplier lead time against that number and both land on a reorder point near 40 units.
For the steady SKU, 40 is honest. Its worst day was five. The order will land with stock still on the shelf.
For the spiky SKU, 40 is a coin toss. One afternoon took 18 units. Two of those inside a ten day wait and the shelf is empty with the pallet still in transit. Same average, same formula, a completely different outcome.
This is why a single figure is never enough on a SKU you care about. The average sets the middle of your runway. Variability decides how much of that runway you can afford to spend before you place the order.
How to judge variability without a statistics course
You do not need a standard deviation to act on this. Four habits get a small store most of the way:
- Look at the worst week, not the mean week. Pull the last eight weeks of units for the SKU and find the highest. That is the week your reorder point has to survive.
- Count the zero days. Lots of zeros with occasional spikes is a sign the daily average is a fiction for that SKU. Think in weeks instead.
- Name the cause before you react. A spike you caused with a campaign is a plan. A spike nobody can explain is noise, and noise needs room rather than a new forecast.
- Choose the window deliberately. A short window tracks a trend fast and reacts hard to one odd day. A long window smooths the odd day and is slow to notice a real shift. The trade-off is covered in the sibling pages on sales velocity and on picking 7, 14 or 30 days.
The honest answer for a high variability SKU is usually to hold more cover and order sooner, not to compute a cleverer average.
Related terms
Demand variability sits between several numbers you will meet on the same screen:
- Sales velocity is the average itself, in units per day over a chosen window.
- The velocity window is how many days of history that average is drawn from, and it is the main lever you have over how much variability shows up in the number.
- Safety stock is the buffer you hold specifically to absorb variability during the wait for an order.
- Lead time is the supplier wait that variability has to be survived across, and the one figure only you can supply.
- Days of inventory on hand is current stock divided by velocity, so it inherits whatever the average hides.
- A stockout is what variability does to you when the buffer was sized for the normal week.
Questions
How do I measure demand variability for a SKU?
Start simple. Export the last eight to twelve weeks of units sold for that SKU, note the highest week and the lowest, and compare both with the average. A high week that is double the average means the average alone cannot size your reorder point. If you want a single figure, a standard deviation of weekly units does the same job more formally, but the highest week is enough to make the decision.
Does a longer velocity window fix demand variability?
No. A longer window hides variability rather than removing it. A 30 day average smooths a one day spike into something small, which is useful when the spike was noise and misleading when it was the start of a trend. The swing in real demand is unchanged either way, so you still need cover for it.
Is demand variability the same as seasonality?
They overlap but they are not the same. Seasonality is a predictable pattern you can plan for by ordering ahead. Variability includes that pattern plus everything unpredictable around it. If you can name the cause and put a date on it, treat it as a plan. If you cannot, treat it as room you need to leave.
What should I do differently on a high variability SKU?
Three things, in order of effort. Hold more cover than the average suggests, because the buffer exists for exactly this. Order earlier, so the wait starts from a higher stock level. And check it by hand more often than your steady lines, especially after anything that moved demand on purpose.
Does Restoc report a demand variability figure?
No, and we would rather say so plainly. Restoc computes a reorder point from the velocity window you pick and the lead time you enter per SKU. It does not output a variance or standard deviation statistic. Choosing a shorter or longer window is the control you have over how much of a SKU's swing reaches the number.
My store is new and has very little sales history. What then?
With a few weeks of data, any variability figure is mostly guesswork, so lean on judgement and a larger buffer until the history builds. The sibling page on a newly connected store walks through what the numbers look like on day one and what to expect as sales accrue.