MRO & spare parts · inventory

Why your spare-parts forecast is wrong, and the stock pays for it.

Spare parts do not behave like products. Their demand is intermittent and lumpy, and the averaging methods most planning systems reach for quietly guarantee two bad outcomes at once: a stockout when the part is finally needed, and dead stock the rest of the year.

6 min read · November 2026

Spare parts wait, they do not sell

A finished good sells more or less continuously, so you can see a trend and lean on it. A spare part is different. It might be pulled three times in a year and sit untouched for the other forty-nine weeks, and then two come out in the same month because a line went down. That pattern, long gaps punctuated by small bursts, is what planners call intermittent or lumpy demand, and it is the normal shape of MRO, spare parts and long-tail SKUs, not an exception.

The trouble is that most planning systems were built for products that move, and they treat a spare part as if it were just a slow version of one. It is not. It is a different statistical animal, and the method that works for a fast mover actively misleads on a part that waits.

Why the average is the wrong tool

Put an intermittent part through a moving average or exponential smoothing and the maths smears those occasional bursts into a tiny, steady-looking number. Demand of two units, then zero for months, then three, becomes a forecast of a fraction of a unit per week. Set a reorder point on that and you will be short the day the part is actually needed, which for a critical spare means a machine or a line standing still while someone pays for emergency freight.

So planners do the sensible human thing: they override the system and hold extra, just in case. Now the opposite problem appears. Safety stock gets set by gut rather than by measured risk, and the shelves fill with parts that may not move for years. The forecast error here is not random noise that washes out; it is structural, and it costs twice, in the stockouts and in the dead stock.

There is a right method, and it is older than the ERP

The honest answer is that intermittent demand has had a proper method since 1972, when Croston separated the problem into two parts: how big a demand is when it happens, and how long the gap is between demands. Forecast those two things separately and you get a far more honest picture of a part that waits than any average can give. The Syntetos-Boylan Approximation, a later refinement, corrects a known bias in the original and is the standard tool today.

None of this is a black box. It is a deterministic, explainable calculation: feed it the same demand history and it returns the same answer, and you can read why. That matters for a spare part, because the person defending the stocking decision is often an operator or a controller, not a data scientist, and "the model said so" is not an answer either of them can use.

The cash hiding in the long tail

Spare parts are where working capital quietly goes to die. Industry estimates vary, but vendor studies commonly put somewhere between a fifth and two-fifths of MRO inventory as excess, obsolete or duplicated. Treat the exact figure with care, because the sources are consultant and vendor samples rather than one clean study, but the direction is not in doubt: a large share of what sits in the storeroom is cash that is not coming back.

The fix is not to hold less. A critical spare that prevents a line-down is worth holding even if it never moves, and cutting it to save carrying cost is a false economy. The fix is to hold the right parts: protect uptime on the critical few by sizing their buffers on real lead time and failure risk, and release the cash tied up in the slow, duplicated and dying many. That is a per-part judgement, and it is exactly the judgement an averaging system cannot make.

What to ask of a spare-parts method

Before trusting any system with the storeroom, put it through a few plain tests. Does it actually model intermittent and lumpy demand, or does it quietly fall back on an average? Does it size safety stock per part, on real lead time and criticality, rather than on a flat days-of-cover rule? Does it surface duplicates and parts sliding toward obsolescence before they are written off? And can it show, for any part, why the reorder point is what it is, in terms an operator and a controller can both read?

A method that passes those tests does the one thing the storeroom needs: it tells the critical few from the slow many, and it keeps the first available while it frees the cash trapped in the second. That is not a dashboard refinement; it is the difference between a line that keeps running and a balance sheet that stops bleeding.

Questions operators ask

What is intermittent or lumpy demand?

Demand that occurs in scattered, irregular bursts with long gaps of zero in between, rather than as a steady flow. It is the normal pattern for spare parts, MRO items and slow-moving long-tail SKUs, where a part may be needed only a handful of times a year.

Why do moving averages fail for spare parts?

Averaging smears occasional bursts of demand into a small, steady number, which sets reorder points too low to cover the moment the part is actually needed. Planners then over-hold to compensate, so the same method produces both stockouts on critical parts and dead stock on the rest.

What are Croston's method and the Syntetos-Boylan Approximation?

Croston's method, from 1972, forecasts intermittent demand by separating the size of each demand from the time between demands, which fits a part that waits far better than an average. The Syntetos-Boylan Approximation corrects a known bias in Croston and is the standard method for intermittent parts today. Both are deterministic and explainable.

How much MRO inventory is typically excess or obsolete?

Vendor and consultant estimates commonly range from about 20% to 40% of MRO inventory being excess, obsolete or duplicated, though these figures come from client samples rather than a single primary study, so they are best treated as indicative. The consistent point is that a large share of spare-parts stock is trapped cash.

The takeaway
A part you cannot forecast is not a problem to ignore; it is a method problem to fix. Hold the critical few, release the slow many.
Yasser El Amiri, Founder, Audacity Business Consulting
Keep reading
Industrial MRO · sector

Long-tail inventory in industrial distribution

Equipment · sector

Spare-parts inventory for equipment and machinery

Inventory · article

Why safety stock is really about error

Value in an hour, not a year

See it on your own data in 45 minutes.

Picture your next S&OP meeting: one number on screen that nobody argues with, and a trail behind it that an auditor could follow. See it free, on your own data, in 45 minutes; and if it does not hold up, we part as friends.