Once you have decided a spare is worth stocking, the next question is precise: exactly how many should sit on the shelf to be confident of covering demand until a replacement can be procured? The standard way to answer this for randomly failing parts uses the Poisson distribution, which describes how many rare, independent events occur in a fixed span. This page walks through the Poisson-based spares calculation, showing how a part's failure rate, the population in service, and the procurement lead time combine into an expected demand, and how you then pick a stock quantity that meets a chosen confidence level. It is the concrete companion to the broader question of optimizing critical spares.
Spares quantity from failure rate in one line: You calculate spares by first finding the expected number of failures during the replenishment lead time, which is the failure rate per unit time, multiplied by the number of units in service, multiplied by the lead time. That expected demand is the mean of a Poisson distribution, and you then choose the smallest stock quantity for which the Poisson probability of demand being at or below it reaches your target confidence level. Higher failure rate, more units, longer lead time, or a higher confidence target all raise the number of spares required.
The starting point is expected demand during the window you must cover, which is the procurement lead time. If a single part fails at a constant rate, and you have several identical parts in service, the expected number of failures over the lead time is the failure rate multiplied by the number of units and by the length of the lead time. That product is a single number, often written as lambda in the Poisson context, and it is the average number of replacements you would expect to need before a new order can arrive. For a rarely failing critical spare this number is usually well below one, which is exactly why the ordinary average is not a stock quantity by itself.
Getting the units consistent is where most errors creep in. The failure rate has to be expressed over the same time base as the lead time, so if failure rate is quoted per year and lead time is in weeks, one of them must be converted before multiplying. Failure rate itself often comes from the inverse of a mean time between failures, or from a FIT rate, or from field records of how many of these parts failed across the fleet in a known period. Whatever the source, the aim is one clean figure for expected demand over the lead time, because the whole Poisson step depends on that figure being right.
It is worth being explicit about the assumptions. This method treats failures as random and independent, occurring at a roughly constant rate, which fits electronic components and many mechanical parts in their useful-life period. It does not fit parts that are wearing out, where failures cluster as the population ages, nor batch defects where one failure signals others. For wear-out behavior a Weibull-based analysis is more appropriate, and the constant-rate Poisson result should be treated as a reasonable model for the random-failure regime rather than a universal law.
With expected demand in hand, the Poisson distribution gives the probability of seeing exactly zero, one, two, or more failures during the lead time. The spares question is then framed as a confidence level, sometimes called a protection or service level: what probability do you want that the number of spares on hand will be enough to cover whatever demand actually occurs before replenishment? A confidence of ninety-five percent, for example, means you want the stock to cover demand in ninety-five out of a hundred comparable lead-time windows, accepting a stockout in the other five.
To find the quantity, you accumulate the Poisson probabilities from zero upward until the running total, the cumulative probability, first reaches or exceeds your target confidence. The smallest stock level at which that happens is the number of spares to hold. Because expected demand for a critical spare is often small, the answer is frequently just one or two units even at a high confidence level, but the calculation makes that defensible rather than a hunch. Raising the confidence target, or the expected demand, will at some threshold tip the required quantity from one to two, and the method shows exactly where that tipping point lies.
A short worked shape makes it concrete. Suppose expected demand over the lead time comes out small, well under one. The probability of zero failures is already high, so a single spare may push cumulative coverage past a ninety-five percent target and one unit suffices. If instead you run several units with a higher failure rate and a long lead time, expected demand might approach or exceed one, the probability of needing two or more becomes non-trivial, and the calculation will call for two spares to keep the same confidence. The value of doing the arithmetic is that it replaces argument with a number you can defend to whoever controls the inventory budget.
The calculation is only as trustworthy as the failure rate and lead time fed into it, and both are best drawn from real operating history rather than assumptions. The failure rate is most credible when it comes from your own fleet: how many of these parts were in service, over how many operating hours, and how many failed. Dividing failures by accumulated part-hours gives a field failure rate that reflects your duty and environment, which can differ substantially from a datasheet figure. The population and hours needed for that ratio are exactly the kind of data a control or SCADA system records as runtime and event counts.
A cloud SCADA platform such as Merobix helps by pulling runtime, start counts, and fault history together across many sites, so the accumulated operating hours and the failure events for a class of part are visible in one record rather than scattered across local panels and paper logs. That central history also captures the actual replenishment lead times of past orders, which is a far better input than a vendor's nominal quote, especially for long-lead items where reality and the catalogue often diverge. Feeding measured lead time into the Poisson step keeps the confidence calculation honest.
Because the inputs drift over time as equipment ages and suppliers change, the calculation is worth revisiting rather than treating as a one-time exercise. Trended fleet data lets you notice when a part's field failure rate is creeping up, which raises expected demand and may justify holding an extra spare, or when a supplier's lead times have shortened enough to safely reduce stock. Keeping the failure rate and lead time current, using the operating record the SCADA platform already holds, means the spare quantity stays matched to how the assets are actually behaving rather than to a snapshot taken years earlier.
The Poisson distribution describes how many rare, independent events happen in a fixed span when they occur at a roughly constant average rate, which matches how many low-usage parts fail during a lead time. It lets you turn an expected number of failures into the probability of seeing zero, one, two, or more, which is exactly what a spares confidence calculation needs. It fits the random-failure period of a part well, though not wear-out behavior, where a Weibull approach is better.
The confidence level should reflect the consequence of a stockout for the asset the spare protects, so a part whose failure stops the plant or a safety function warrants a higher target than one backing a redundant, low-impact service. Values such as ninety or ninety-five percent are common for important spares, while less critical items can tolerate lower coverage and less stock. The right level is a business decision about acceptable stockout risk, and the Poisson method then converts that choice into a quantity.
With little or no field history you can start from a manufacturer datasheet, a FIT rate, or a published reliability figure, and use the inverse of a mean time between failures as the rate. These sources tend to be optimistic, so treat the resulting spare quantity as a first estimate and revise it as your own operating hours and any failures accumulate. Capturing runtime and fault data from the start, through a SCADA or maintenance system, lets you replace the borrowed rate with a real one as soon as you have enough evidence.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.