Safety stock formula and calculation for purchased parts

What is safety stock?

Few inventory decisions carry as much weight for as little attention. Safety stock (sometimes also called buffer stock or safety inventory) is the stock held above expected use during a replenishment lead time. Cycle stock represents the stock you are holding to cover expected use in production in between deliveries. When use and delivery both run to plan, the next order arrives as stock reaches the safety stock/buffer, and the safety stock is never touched.

Two things force dipping into safety stock: (1) unplanned additional use of a given material (e.g., due to additional scrap, or a changed production schedule), and (2) suppliers delivering later than expected. For purchased components with lead times longer than a month, late deliveries are typically the larger risk of the two by a wide margin.

Safety stock formulas make calculation simple

Measure lead time from your own receipts, not just the quote

The quality of a safety stock calculation depends less on the formula than on its inputs, and the input that matters most is how much supplier lead time varies. In the best case, a plant holds delivery history for each supplier and part over the past 6+ months: the order and received date of every PO. The standard deviation of those actual lead times can be easily calculated and used straight in the formula. Where that history is thin or missing, the spread has to be estimated from what the supplier provides, and several methods do this well.

Turn a quoted range into a lead time

Without delivery history, the supplier’s quote is the best place to start. A typical quote, such as “35 to 60 days, usually about 45,” can be used to estimate a useful lead time mean and spread.

Why a quoted range understates the tail

Quotes describe the range a supplier expects when nothing goes wrong. Material shortages, missed trucks and quality holds add days at the long end, while little makes a delivery arrive much earlier. Actual lead times therefore skew late. A quote is best treated as a starting point, to be replaced with receipt data once about ten orders are available.

Switch to your own receipts after about ten orders

Lead time is best measured from the PO date to the day the part is received into stock and available to issue; the supplier’s ship date understates it. The calculator accepts the order, promised and received dates for each PO line and returns the mean, standard deviation and 90th percentile of actual lead time, along with the average gap to the promised date. When σ_LT exceeds roughly a quarter of the average lead time, supply should be treated as unreliable. The open PO tracker keeps the requested, committed and actual date of every open PO line, and a supplier scorecard shows lead-time adherence by supplier from a PO export.

Measuring at the level of supplier and part pays for itself. Under the combined formula, a supplier that delivers within a tight range earns a smaller buffer at the same service target, with no special setting. The effect is material: in the worked example below, excluding lead-time variation cuts the buffer from 3,567 units to 915.

Promised and actual lead times should be kept apart. If receipts arrive late on average, the planned lead time in the ERP needs correcting first. A supplier that promises 42 days but delivers in 47 on average has a 47-day lead time, and that figure belongs in the ERP. Safety stock then only has to cover the swings around those 47 days, not the five days of lateness every order carries. For what a quoted lead time often leaves out, see our article on supplier lead time.

Cartograph does this automatically. It measures each supplier’s lead time and its spread from the order and received dates on your POs, and suggests a safety stock strategy for each part, with the formula and inputs behind every figure.

Measure daily use in full: scrap and every product the part goes into

Lead time is one half of the formula. The other half is daily use: d, the average daily consumption of a given part, and σ_d, how much that varies. Both are easy to understate on a component, because its use comes from more than one place: the parts that end up as scrap, and every product it goes into.

Scrap and yield

Scrapped parts draw on the same stock. Average daily use and its standard deviation should be grossed up by the scrap rate before any calculation. Omitting a 2 percent scrap rate in the example lowers lead-time demand by about 370 units and safety stock by 70, and the reorder point fires late in every cycle.

One component, several parents

When a component goes into several products, its daily use is the total across all of them. Multiply each product’s daily builds by how many of the component it takes, then add the results. The MRP calculator shows how a build plan turns into dated component requirements.

The variation adds up differently, depending on how the products behave:

• The products sell independently (the usual case, and the calculator’s default): square each product’s standard deviation, add the squares, and take the square root. Busy days for one product tend to be offset by quiet days for another.
• The products rise and fall together, for example because they share one end customer or one season: add the standard deviations as they are. Nothing offsets, so the buffer is larger. In the example, it rises from 3,567 to 3,774 units.

Calculate safety stock with the formula that fits your part

With sound inputs in place, the choice of method determines whether the buffer reflects the risk a part actually carries.

Calculate your safety stock

The calculator opens on an illustrative component, so a complete result is visible before any value is changed. Open a section to replace its inputs with your own, and the results update as you type.

Two simpler formulas for parts with one source of variation

The combined formula in Fig. 2 covers swings in both daily use and lead time. When only one of the two moves, a simpler statistical formula covers it and shows directly which risk the buffer is for. Hover over or tap any symbol to see what it stands for.

This formula fits when the supplier delivers on the same day every time and only consumption varies. Because high and low days partly offset over the lead time, it scales σ_d by the square root of LT.

This formula fits when consumption is steady and delivery dates move. Each day of delay costs a full day of use, so σ_LT is multiplied by d.

Z and the service level

Z comes from the one-sided normal table: it is the number of standard deviations of buffer required to cover a given share of replenishment cycles. In any spreadsheet, =NORM.S.INV(0.95) returns 1.645. The two-sided value (1.96 for 95 percent) covers both tails and overstates the buffer, since stockouts occur in only one.

The calculator’s service levels are cycle service levels: the probability that a replenishment cycle ends without a stockout. Fill rate, the share of demand served from stock, is a different measure and depends on order quantity as well as the buffer. On steady parts, a 95 percent cycle service level produces a fill rate above 95 percent, because a short cycle runs short only in its final days. On lumpy demand the relationship reverses, and a 95 percent cycle service level can coincide with a fill rate near 50 percent.

Order size matters as well. A large minimum order quantity (MOQ), the smallest order a supplier will accept, already protects most cycles, so a cycle service target overstocks the part; for such parts, the buffer is better sized on fill rate. Any stated service level should specify which of the two measures it uses. To size the order itself, the EOQ calculator fits it to the MOQ, pack size and price breaks.

Less common methods exist for parts these formulas handle badly: a Poisson or negative binomial quantile for lumpy demand, a buffer based on the P95 lead time for suppliers whose deliveries skew late, and fill-rate sizing for parts with large order quantities. Fill rate comes up with the service level above, and Boundary conditions covers the other two. The quickest method of all, safety days, holds a set number of days of average use, such as 5 × 408 = 2,040 units for the example. It needs no statistics and carries no service level, so keep it for cheap, steady parts where a rough buffer costs little.

Set the service level from what a shortage costs

Every service target implies a price for a missed unit, whether or not that price was chosen deliberately. The critical ratio, a standard rule of inventory planning, makes the link explicit: cycle service level = p ÷ (p + h), where h is the annual cost of holding one unit and p is the cost of one unit short. Rearranged, p = h × CSL ÷ (1 − CSL).

A part that costs $2 a year to hold, targeted at 95 percent, implies about $38 per unit short. In the calculator’s example, a $3.20 part at a 25 percent carrying cost costs $0.80 a year to hold, so a 95 percent target values a missed unit at $15.20. The figure works as a sanity check: a target that values a $3 fastener’s stockout at $400 is set too high. Fill-rate targets and continuous review shift the exact number, so it is best read as an approximation.

The same arithmetic explains why the right targets differ by part. The more robust approach is to set one objective for the whole inventory, such as the least stock for a given service floor, and to derive each part’s level from it. Most inventories sit inside the best achievable trade-off between stock and service, which means the same service can often be delivered with less capital.

Worked example: a purchased component

Illustrative example. A machined component goes into two pump housings and a service kit, which use 180, 140 and 80 units a day respectively (400 in total), with standard deviations of 58, 46 and 30. Scrap runs at 2 percent. The supplier quotes 35 to 60 days, usually 45. The part costs $3.20, and the target is a 95 percent cycle service level.

1. Daily use: 180 + 140 + 80 = 400. Each parent varies independently, so σ_d = √(58² + 46² + 30²) = 79.9.
2. Scrap: 400 × 1.02 = 408 a day, and σ_d = 79.9 × 1.02 = 81.5.
3. Lead time from the triangle: LT = 46.7 days and σ_LT = 5.14 days.
4. Demand term: 46.7 × 81.5² = 309,762. Lead-time term: 408² × 5.14² = 4,392,800.
5. Safety stock: 1.645 × √(309,762 + 4,392,800) = 3,567 units, 8.7 days of use, $11,414 at cost.
6. Reorder point: 408 × 46.7 + 3,567 = 22,607 units.

The lead-time term accounts for 93 percent of the variance. Excluding lead-time variation yields about 915 units, roughly a quarter of what the part requires. Max/average, with a peak of 560 a day before scrap, calls for 15,232 units, about $37,000 more stock than the combined result. Reading the same quote as uniform raises σ_LT to 7.2 days and the buffer to 4,931 units.

Use it in a spreadsheet

The same formulas are available as a free Excel and Google Sheets workbook: download the safety stock template (.xlsx). It turns a supplier quote into a lead time and spread, reads your own receipts from a second sheet, and returns the four methods, the reorder point and the cash in the buffer. It holds the worked example, so you can check it against Fig. 11 before replacing the inputs.

What each method delivers in simulation

To see how the methods behave, we simulated one million replenishment cycles for the worked example: daily use normal with a mean of 408 and a standard deviation of 81.5, a service target of 95 percent, and lead times drawn two ways. In the first, each delivery follows the supplier’s quote, a triangle from 35 to 60 days, usually 45. In the second, 85 percent of deliveries follow a tighter 35 to 55 day triangle and 15 percent are delayed to between 60 and 80 days, a stand-in for quality holds and missed trucks. In the second case every buffer is sized from the measured mean (48.8 days) and spread (9.9 days) of those deliveries, as it would be from your receipts. Fig. 12 shows the share of cycles that ended without a stockout.

Three results stand out. When deliveries follow the quote, the combined formula comes close to its target (94 percent against 95), while the demand-only formula, which ignores the supplier, protects only two cycles in three. When some deliveries run weeks late, the combined formula falls to 89 percent, because it assumes a bell curve and the delays form a long tail; sizing the reorder point from the 95th-percentile lead time restores the target at 95 percent. Max/average never ran short in either case, but it held 4.3 times the combined buffer in the first and 3.8 times in the second. The P95 row sets the reorder point at average daily use × the 95th-percentile lead time, and its buffer is that point minus lead-time demand at the mean. These are model results for one part under these assumptions, not data from live orders; change the assumptions and the numbers move. Download the simulation script (Python) to rerun it.

Safety stock formulas are no silver bullet: check their boundary conditions

A calculated buffer is a hypothesis about how a part behaves. Before it goes into the ERP, it should be tested against the conditions in which the formula breaks down, and then maintained as those conditions change.

Boundary conditions

The formula assumes that daily use and lead time each follow a bell curve and vary independently of one another. For many components, at least one of these assumptions fails.

Lumpy demand

When a component is consumed in batches, such as nothing for nine days and then 4,000 units for one work order, the daily average d and its swing σ_d describe no day the plant will actually see. The resulting buffer is too small for the batch and too large between batches. These parts are better planned from the production schedule through MRP. Where a buffer is held, it should be sized from a Poisson or negative binomial quantile rather than Z × σ_d, since the normal curve mis-stocks such parts. Two common cutoffs sort them: an average gap between demands above 1.32 periods marks a part as intermittent, and a squared coefficient of variation of demand sizes above 0.49 marks it as lumpy as well.

Shared capacity

When several parts run on the same machine at one supplier, a single bad month makes all of them late at once. Each part’s safety stock looks adequate on its own, yet the set leaves the build exposed, because the formula treats each lead time as a separate draw. The buffer is better sized on the part that gates the build, with the remaining parts planned to arrive alongside it.

Quotes that aren’t independent

The formula also assumes that one late delivery says nothing about the next. In practice, lead times move in runs: when a supplier falls behind, the next few POs often arrive late as well, and when demand rises across the market, consumption and supplier lead time rise together. Where received dates show runs of late POs, σ_LT should be sized from the bad stretch, or the 90th-percentile lead time used as LT with the demand-only formula on top. Where busy periods and late deliveries go together, add the two single-source formulas instead of combining them under one root: SS = Z × σ_d × √LT + Z × d × σ_LT. For the example, that gives 915 + 3,447 = 4,363 units, a safe upper bound on the combined 3,567.

Skewed and late-only lead times

A supplier that delivers late or later, and seldom early, produces a long right tail that the normal curve misses. When the P95 lead time from receipt history sits well above LT + 1.645 × σ_LT, the empirical percentile is the better basis: reorder point = d × P95 lead time, plus a small demand buffer.

Daily swings scaled up to the lead time

Multiplying σ_d by √LT assumes that average use is known exactly. In practice it is estimated, and the buffer comes out too small, with service below target, most of all when the average rests on a few weeks of data. Measuring the swing on weekly buckets, or on lead-time demand itself, avoids the problem. The bucket also changes the coefficient of variation, so it should be fixed before any XYZ cutoff is set.

When timing is the risk

Where receipts arrive late but consumption holds steady, safety lead time can outperform safety stock: the order goes out a few days earlier and the quantity stays the same. Short shipments are a different risk. They are a yield problem, and σ_LT does not capture them.

Risk that history cannot see

Single-source parts and suppliers in disruption-prone regions carry risk that no PO history reveals. That risk is better covered by a strategic buffer, held as a separate layer above safety stock and sized in days of cover, so that the time the plant can survive a disruption is at least the time the supplier needs to recover. Keeping it out of the statistical safety stock makes the cost of that resilience visible to the business.

An answer that looks too large

A result measured in weeks of stock usually points to an input problem rather than a formula problem. Common causes include a single PO that waited on a drawing change, a lead time in calendar days set against use in working days, and a σ_LT drawn from a history that mixed two suppliers. If the number holds up under that scrutiny, a tighter delivery promise, a second source or a supplier that confirms and expedites will cost less than the additional stock.

MOQ and order size

The MOQ determines how much arrives, not when to reorder. A large MOQ raises average stock, yet stock still falls to the reorder point before each new order lands, so the safety stock requirement is unchanged. Order quantity is a separate decision, best made with the MOQ and price breaks in view.

Reorder point from your result

The reorder point adds the buffer to expected use during the lead time: average daily use × average lead time + safety stock. It should be compared with inventory position, meaning on-hand stock plus open POs minus allocations. On-hand stock alone misses both what is already on order and what is already committed to jobs. Reorder point formula covers inventory position, MOQ at arrival and components with several parents.

How often to recalculate

For most parts, a quarterly review is sufficient. A supplier quote change, a parent product ramping up or ending, or received dates drifting from promised ones should each trigger an immediate recalculation. Lead time tends to move first, so it is the input to keep most current.

Actual lead times stay current only when someone records the promised and received date for each PO. Cartograph follows up on open POs: Scout Agent emails and calls suppliers to confirm dates and expedite late lines, and Cartograph records the promised and actual dates on each PO line. Get Cartograph free to see it on your own POs.

Safety stock questions