Ask how a business sets its safety stock and the usual answer is a rule of thumb: "we keep two weeks of everything." It is easy to remember and easy to apply. It is also wrong for most items.
A flat rule ignores the two things that create the need for a buffer: how much demand swings, and how reliable the supplier is. An item with steady sales and a dependable supplier needs very little. An item with erratic sales and a supplier who is sometimes two weeks late needs a lot. "Two weeks" gives both the same cover.
In Post 1 we used ABC analysis to find which items matter most at Northshore Supply, a fictional distributor. This post sets a safety stock and reorder point for three of those items, one from each class, and shows what the flat rule gets wrong.
What safety stock is
Safety stock is the extra inventory you hold to cover the gap between what you expected and what actually happened during a supplier's lead time. It is not the stock you plan to sell; it is the cushion underneath it.
It feeds directly into the reorder point, the stock level at which you place the next order:
Reorder point = (average demand × lead time) + safety stock
Three things decide how big the cushion needs to be:
| Driver | The question it answers | Where to find it |
|---|---|---|
| Demand variability | How far does weekly demand stray from its average? | Standard deviation of weekly sales over the last 12 months |
| Lead time and its variability | How long does the supplier take, and how often are they late? | Purchase order date to receipt date, for the last 10 to 20 orders |
| Target service level | How often are you willing to run out? | A business decision, set by ABC class |
A 95% service level means that in 95 out of 100 replenishment cycles, stock lasts until the next delivery arrives. The other five times, you run short. Service level is a choice, and a higher one always costs more stock.
Three methods, from simple to rigorous
Method 1: Average-max
This needs no statistics. Take your worst realistic case and subtract the normal case.
SS = (max weekly demand × max lead time) − (average weekly demand × average lead time)
It is quick and works with very little data. Its weakness is that it assumes the busiest week and the latest delivery happen together, so it usually overstates the buffer. One freak week in your history can inflate it badly.
Method 2: Statistical, demand variability only
This is the standard textbook formula. It sizes the buffer from how much demand actually varies and the service level you choose.
SS = Z × σd × √L
Here Z is the service factor, σd is the standard deviation of weekly demand, and L is the lead time in weeks. Use it when your supplier delivers on time, every time.
Method 3: Statistical, demand and lead time variability
Most suppliers are not perfectly reliable. This version adds the lead time's own variability.
SS = Z × √( L × σd² + d² × σL² )
Here d is average weekly demand and σL is the standard deviation of lead time in weeks. This is the method to use for A and B items whenever you have the lead time history.
The service factor Z
Z converts a service level into a number of standard deviations. In Excel it is =NORM.S.INV(service level).
| Service level | 90% | 95% | 98% | 99% | 99.9% |
|---|---|---|---|---|---|
| Z | 1.28 | 1.64 | 2.05 | 2.33 | 3.09 |
Keep the units consistent. If demand is measured per week, lead time must be in weeks too.
A worked example: three items from Northshore Supply
We take one item from each ABC class in Post 1. Average weekly demand is the annual usage divided by 52. The other figures come from Northshore's sales and purchase order history.
| Input | NS-101 Nitrile gloves (A) | NS-108 Hi-vis vest (B) | NS-111 Shop towels (C) |
|---|---|---|---|
| Average weekly demand (units) | 100 | 17.3 | 10 |
| Standard deviation of weekly demand | 20 | 6 | 5 |
| Maximum weekly demand | 140 | 30 | 22 |
| Average lead time (weeks) | 4 | 3 | 2 |
| Standard deviation of lead time (weeks) | 1.0 | 0.5 | 0.5 |
| Maximum lead time (weeks) | 6 | 4 | 3 |
| Target service level | 98% | 95% | 90% |
| Unit cost | $62 | $68 | $44 |
Step by step for the gloves, using Method 3
- Demand part: L × σd² = 4 × 20² = 1,600
- Lead time part: d² × σL² = 100² × 1² = 10,000
- Add and take the square root: √11,600 = 107.7 units
- Multiply by Z for 98%: 2.054 × 107.7 = 221.2, rounded up to 222 units
- Reorder point: (100 × 4) + 222 = 622 units
When stock on hand plus stock on order falls to 622 cases, Northshore places the next order.
Results for all three items
| Result (units) | NS-101 Gloves (A) | NS-108 Vest (B) | NS-111 Towels (C) |
|---|---|---|---|
| Method 1: Average-max | 440 | 69 | 46 |
| Method 2: Demand variability only | 83 | 18 | 10 |
| Method 3: Demand and lead time | 222 | 23 | 12 |
| Reorder point (Method 3) | 622 | 75 | 32 |
| Safety stock value (Method 3) | $13,764 | $1,564 | $528 |
| "Two weeks of everything" rule | 200 | 35 | 20 |
Four lessons come out of this table.
- Lead time variability can matter more than demand variability. For the gloves, the lead time part (10,000) is six times the demand part (1,600). Method 2 ignores it and gives 83 units, well under half of what is needed.
- Average-max overshoots. It asks for two to four times the Method 3 figure, because it plans for the worst week and the latest delivery arriving together.
- The flat rule puts stock in the wrong place. Two weeks is 10% too little for the gloves, about 50% too much for the vests and 67% too much for the towels.
- The right buffers do not cost more. Method 3 holds $15,856 across the three items; the flat rule holds $15,660. The money is nearly the same, but it now sits where the risk is.
Choosing a service level for each class
Service level is where ABC analysis and safety stock meet. A common starting policy looks like this.
| Class | Typical target | Reasoning |
|---|---|---|
| A | 97 to 99% | A stockout loses the most revenue, so aim high, but calculate carefully because the buffer is expensive |
| B | 95% | A solid default for items that matter but are not critical |
| C | 90 to 95% | A stockout costs little, and a simple generous buffer is often enough |
The catch is that service gets more expensive the higher you go. The chart shows the safety stock value for the gloves at eight service levels.
Moving from 95% to 99.9% nearly doubles the buffer, from $11,036 to $20,646. The last step alone, from 99% to 99.9%, costs $5,084 for less than one point of service. No level reaches 100%, because the formula has no upper limit.
The same step is cheap on a C item. Raising the shop towels from 90% to 99% adds 9 cases, or $396. That is why many businesses do the careful calculation only for A and B items, and give C items a generous flat buffer. The effort goes where a wrong answer costs real money.
Practical cautions
Lumpy demand breaks the formula. Methods 2 and 3 assume demand is spread roughly evenly around its average. An item that sells nothing for five weeks and then 40 in one week does not behave that way. As a rough test, if the standard deviation is larger than the average, do not trust the statistical result. Use average-max or a simple min/max instead.
Measure real lead time. Use the date the goods were received, not the date the supplier promised. Count from the day you placed the order, and include your own receiving and put-away time.
Remove one-off events. A single project order or a one-time promotion inflates the standard deviation for a year. Take it out before you calculate, and note that you did.
Safety stock does not fix a biased forecast. If demand is consistently above forecast, the problem is the forecast. Safety stock covers random swings, not a steady error in one direction.
Compare the reorder point to the right number. Use stock on hand plus stock already on order, less any customer backorders. Checking on-hand alone leads to double ordering.
Review on a schedule. Recalculate A items every quarter, B items twice a year and C items once a year. Recalculate immediately when you change supplier or a lead time shifts.
Your checklist
- Pull 12 months of weekly demand for your A and B items
- Pull order and receipt dates for the last 10 to 20 purchase orders per item
- Calculate the average and standard deviation of demand and of lead time
- Set a target service level for each ABC class
- Calculate safety stock with Method 3, and the reorder point
- Use average-max or min/max for C items and for lumpy demand
- Load the reorder points into your system and set a review date
Next in the series: Post 3 applies ABC analysis to a manufacturer, where raw materials, work in progress and spare parts each need their own treatment.
Chmltech Ltd. helps small and mid-sized businesses improve operations and adopt digital tools. If you would like help applying this to your own inventory, get in touch.
