Spare Parts Inventory Management 2026 – Deciding Stock by Downtime Cost

“We put in a spare parts inventory management system and the inventory value has not moved.” You hear this often in Japanese-owned plants in Thailand. The problem is rarely the system. It is the yardstick. Spare parts demand does not appear every day. Nothing moves for months, and then one item moves once. Apply production-material metrics such as turnover or days of inventory to that pattern and you end up scrapping the parts you should be holding and holding the parts you should be scrapping. This article works through three axes – how demand occurs, what downtime costs, and how long procurement takes – using the numbers of an 80-machine model plant to answer a single question. How much maintenance stock should you actually carry?

Why a spare parts inventory management system ends in “we installed it and nothing shrank”

Start with the model plant used throughout this article. It is a Japanese-owned metalworking and assembly site in Thailand with 80 machines, 1,200 spare parts SKUs, spare parts inventory valued at 4,800,000 THB, and annual maintenance parts consumption of 2,400,000 THB. That is an inventory turnover of 0.5 times per year. Items with no issue for 24 months or more – dead stock in the usual sense – account for 38% of SKUs and 27% of value, which is 1,300,000 THB.

Show those numbers to a management team and the first reaction is almost always the same. Turnover of 0.5 times a year is far too low. By production-material standards it is low. Inventory that turns half a time a year would be flagged for corrective action in any materials department. But the moment you pass that judgement on spare parts, the design starts moving in the wrong direction.

The reason is simple. The fastest way to raise spare parts turnover is to scrap the parts that do not move. Sweep mechanically through the 1,200 SKUs, remove the slow movers, and the denominator falls and turnover rises. Except that mixed in among those slow movers is a part that fails once a decade and takes an A-tier machine down for 72 hours when it does. Turnover cannot tell that part apart from a duplicate someone ordered by mistake. Both sit in the same column labelled “no issue in two years”.

The value of spare parts inventory is not in being consumed. It is in sitting there unconsumed. It behaves like insurance, and calling the premium for a year you never had to claim on “money wasted” tells you nothing. What deserves measurement is the comparison between the expected loss that occurs when the part is missing and the cost of continuing to hold it.

Holding cost also looks small when you count only the interest. This model plant uses a holding cost rate of 22% per year, made up of 8% cost of capital, 2% storage, 10% obsolescence, and 2% insurance and stocktaking. Obsolescence is why the rate runs high for spare parts. Replace a machine and its parts have nowhere to go, and electronic components lose their value to EOL, or End of Life. Holding 4,800,000 THB of stock therefore means paying 1,056,000 THB every single year.

So the first decision in a spare parts inventory management project is not a feature list and not a product name. It is which yardstick you put at the centre. A system configured around turnover or days of inventory targets will push the shop floor to scrap parts it should keep, no matter how capable the software is. The next sections set out the three axes that belong there instead.

Spare parts demand is intermittent – why no formula gives you the right stocking level

Safety stock formulas for production materials assume demand arrives every day from a stable distribution. You can compute a mean and a standard deviation, approximate them with a normal distribution, and work backwards from a target service level. For materials that move daily, that assumption is practical enough.

For spare parts it collapses. Zero issues this month, zero next month, then several at once half a year later. Demand shaped like that is called intermittent demand. Take the mean of intermittent demand and you get a fraction of a unit, and there is no month in which that fractional quantity actually occurs. Neither the mean nor the standard deviation describes anything that happens in reality.

Run a safety stock formula on top of that and it misses in both directions. Parts with few issues come back as “you barely need this”, and parts that happened to move recently come back as “hold more”. The first group contains expensive items that exist as insurance, and the second contains consumables you can buy from a local distributor tomorrow. That is why you cannot derive the right spare parts stocking level with the same arithmetic you use for production materials.

SBC classification – four quadrants from ADI and CV²

The starting point for handling intermittent demand is classifying the demand pattern itself. The common approach follows Syntetos, Boylan and Croston (2005), and is usually shortened to SBC classification. It uses only two measures.

ADI, or Average Demand Interval, is the average gap between demand events. It expresses how many periods pass between one month with an issue and the next, so a larger value means “rarely moves”. CV² is the squared coefficient of variation of demand size, expressing how much the quantity varies when demand does occur. A part that always moves in the same quantity has a small CV², and a part whose quantity swings wildly has a large one.

Put thresholds on both and you get four quadrants. The cutoffs are ADI = 1.32 and CV² = 0.49, and the four names are smooth, intermittent, erratic and lumpy.

These cutoffs are not absolute boundaries. They are a guide for choosing a statistical forecasting method. If an item sits close to a threshold and your conclusion flips depending on which side you file it under, treat that as a signal to decide it on downtime cost and lead time instead, using the two sections that follow.

The stocking policy changes in every quadrant

What matters more than the classification is that the response differs by quadrant.

Smooth covers filters, O-rings, lubricants, belts and similar items replaced on a planned maintenance cycle. Demand here is predictable, so you can set a reorder point and an order quantity and put them on automatic replenishment. This is the only part of the spare parts population where production-material thinking transfers cleanly.

Intermittent covers parts with gaps between events that move one or two units when they do. General purpose sensors, relays and small motors are typical. Because the quantity is stable, the decision is whether to keep a small fixed stock and replenish on use, or to lean on distributor stock, and lead time settles that question. This is not territory for statistical reorder point calculations.

Erratic covers items that move in almost every period in unpredictable quantities, such as heavily worn jig components, cutting tools and seals. Here linking parts to equipment matters more than the stock number, because once you know which machine consumes how many, you can read the requirement forward from the production plan.

Lumpy is the hardest quadrant, and statistical forecasting barely works in it. Control units, servo amplifiers and dedicated drive components live here. Investing in better forecast accuracy for this quadrant does not pay back. The decision rests not on a forecast but on how much you lose when the part fails. The formula in this article is primarily a tool for this quadrant.

Why days of inventory and turnover are the wrong KPIs for parts stock

Taken together, the problem with turnover and days of inventory as KPIs becomes clear. These measures work for smooth items. But for the intermittent and lumpy parts that make up much of the 1,200 SKUs by count, all they can say is “rarely moves, therefore bad”. Structurally, the metric argues against insurance stock.

Three KPIs belong there instead. The first is equipment downtime caused by waiting for parts, particularly on A-tier machines. The second is the number of emergency air freight shipments, which is what a failed stocking design looks like once it turns into cash. The third is dead stock value, meaning the value of stock with no issue for 24 months. What matters here is reading that value in two halves. Stock that downtime cost justifies – insurance you are holding deliberately – is dead stock you are entitled to keep, and everything else is dead stock you should be reducing. Only the second half gets a reduction target. Hold it as a value rather than a rate, because a rate can be improved by adjusting the denominator.

For production materials that move daily, turnover and days of inventory remain perfectly valid. The two populations should be designed as separate management systems. The production-material side is covered in optimal inventory level management 2026, and reading it as a contrast makes the difference easy to see. Lump both under the single word “inventory” and run them on the same system settings, and one of the two will always break.

The second axis – sorting equipment into three tiers by downtime cost

Demand pattern alone does not decide whether to hold a part. An item that “rarely moves” still deserves stock if a stockout stops the whole line the moment it happens, and it does not if the line can route around the machine. That is where the second axis comes in – what a stoppage costs.

Unplanned downtime is recognised globally as a large number. The Siemens and Senseye analysis published as The True Cost of Downtime 2024 puts the losses of the world’s 500 largest companies from unplanned downtime at around 1.4 trillion USD a year, equivalent to 11% of revenue. In automotive it cites up to 2.3 million USD per hour. On the improvement side, the same analysis reports monthly stoppage counts falling from 42 to 25 and downtime hours falling from 39 to 27. These are figures for the world’s largest 500 firms and do not transfer directly to a mid-sized plant in Thailand, but the underlying idea – managing downtime in money – works at any scale.

The model plant sorts its 80 machines into three tiers by the loss per hour of stoppage.

The important thing about this table is not the absolute amounts but the ratio between tiers. Tier A and Tier C are 37.5 times apart. The same part at the same unit price can be justifiable as stock when it sits on a Tier A machine and indefensible when it sits on a Tier C one. Classifying by part number or unit price will never surface that difference. The point is to classify by the equipment the part sits on, not by the part.

Building the tiers is not difficult. Loss per hour is the production volume lost when the machine stops, multiplied by the contribution margin per unit. If overtime or weekend work can recover the output, substitute the cost of that recovery. Rigorous cost accounting is unnecessary, and judgements do not change as long as the order of magnitude is right. Letting the exercise stall for six months in pursuit of precision costs more than the precision is worth. Manufacturing and finance can assign values to all 80 machines in a day, and that is enough for a first version.

The distribution of machines matters too. Tier A holds 12 machines, or 15% of the fleet. Concentrate management effort there and run the 40 Tier C machines on a fix-on-failure basis. The same structure appears when you decide what to outsource, which is covered in equipment maintenance outsourcing 2026. Spare parts and outsourcing look like separate topics, but both are variations on one question – how much do you spend on which machine.

Tiers are not permanent. Change the product mix and a machine that was Tier C yesterday becomes the bottleneck. Reassign them at least once a year, in step with the review of the production plan. Refining the spare parts design on top of a stale tier map produces precision that means nothing.

The third axis – procurement lead time varies 50 times over in Thailand

The third axis is procurement lead time, and this is where managing spare parts from a Thai site clearly departs from doing it in Japan. The same act of “ordering one part” differs by an order of magnitude depending on the route.

Five days if the local distributor has stock, 30 to 55 weeks for a control unit with semiconductors in it. The spread exceeds 50 times. Reading this table is straightforward – ask whether procurement lead time exceeds the time you can afford to leave that machine down. If it does not, you do not need to hold the part. If it does, you hold it or you arrange something else. Unit price and failure rate come after that test, not before.

The bottom row reflects specific conditions. GlobX, in Electronic Component Lead Times 2026 updated on 17 July 2026, reports MCUs at over 30 to 55 weeks, FPGAs and CPLDs at 40 to 52 weeks, power semiconductors and analogue ICs at 26 to 40 weeks, passives such as MLCCs at 12 to 26 weeks, and standard logic and discretes at 10 to 20 weeks. Normal conditions of 8 to 16 weeks have stretched to 26 to over 55 weeks as of 2026. The plants affected are not the ones buying bare semiconductors. They are the plants buying the control units, inverters and amplifiers those semiconductors go into. When the manufacturer cannot secure components, the delivery date of the finished unit slips by the same amount.

What that means in practice is that “order it when it breaks” no longer works for control equipment. No plant can leave a machine down for 40 weeks. The result is cannibalising parts from other machines, hunting for used units, or retrofitting an alternative model, and each of those piles maintenance hours and downtime on top of each other. That is why control components deserve insurance stock even when their failure rate is low.

One more condition changed in 2026. From 1 January 2026, Thailand abolished the 1,500 baht de minimis exemption under Customs Notification No. 219/2568. Shipments are now subject to 7% VAT and duty regardless of value. On 30 March 2026, tariff increases focused on consumer goods were also announced. Spare parts imports were always taxable, but an operation that used to send small sensors and connectors individually by air now carries customs handling and cost on every single shipment. Deciding that “it is a small part, so we will just order it each time” has become distinctly less attractive.

In practical terms, value now sits in designs that reduce the number of emergency shipments. For the same annual import value, consolidating 32 shipments into 11 cuts customs fees and internal processing effort alike. This feeds straight into layer 4 of the five cost layers discussed later.

Five stocking patterns that fall out of the three axes

Demand pattern, downtime tier, procurement lead time. Once those three are fixed, spare parts stocking resolves into five patterns. The work is not examining 1,200 SKUs one at a time. It is deciding which of the five each item belongs to.

Type 1, stock to a reorder point, is the ordinary approach. Fall below the set quantity and replenishment triggers automatically. It suits smooth items and inexpensive high-consumption erratic items. This is the area you can automate, so system configuration earns its keep here.

Type 2, a single insurance unit, applies to parts that rarely move but are fatal when missing. Set no reorder point and no order quantity. Keep exactly one unit on the shelf and replenish without fail after use. Lumpy items on Tier A machines with long lead times are typical, and the formula below is what justifies them. You do not need quantity. A second unit barely reduces the probability of a stoppage while doubling the holding cost.

Type 3, consignment stock at the distributor, means the distributor carries the stock and you pay for what you use. VMI stands for Vendor Managed Inventory, an arrangement in which the supplier manages the stock. It never becomes your asset, so no holding cost attaches, and you can still draw on it the same day. In the industrial estates around Bangkok this is a realistic option for general purpose items. But distributor stock is not reserved for you. A neighbouring plant can claim the same part first, so avoid consigning everything that sits on the critical path of a Tier A stoppage.

Type 4, sharing across sites, means pooling stock with other plants in the group or other sites in Thailand. Where identical machines exist at several sites, one unit per site is duplication. This only works on one condition – that you can see what the other sites hold, immediately. If finding out means phone calls and emails, buying your own will be faster than the search. That condition translates directly into a system requirement.

Type 5, do not hold it, covers parts the distributor stocks where downtime cost is low as well. Most general purpose Tier C items belong here. So do the majority of parts sitting on shelves “just in case”.

The mapping between the three axes and the five patterns looks like this.

Use this table in one direction – lead time first, then tier, then demand quadrant. Reverse the order and every unpredictable part becomes insurance stock, and inventory swells. You do not need to file all 1,200 items neatly. Assign the top 20% by value and every part sitting on a Tier A machine, and you have covered most of the money.

How expensive a part can justify – working out the break-even unit price for maintenance parts

This is the heart of the article. Here is the formula that justifies Type 2 insurance stock.

Maximum justifiable unit price = annual failure rate λ × (hours of stoppage directly attributable × cost per hour of stoppage) ÷ holding cost rate

Put differently, the test is not turnover but whether avoided expected loss exceeds holding cost. λ, lambda, is the annual failure rate, meaning how many times a year that part fails on average. Set it from issue history, the manufacturer’s recommended replacement interval, and experience with identical machines. “Hours of stoppage directly attributable” is not the total time from failure to recovery. It is the extra hours the machine stays down because the part was not on hand. Confuse the two and the numbers come out far too large.

Example 1 – a servo amplifier on a Tier A machine

Unit price 180,000 THB, annual failure rate λ = 0.15 per year, procurement lead time 8 weeks. Assume that even without the part on hand, workarounds such as cannibalising another machine, renting, or repairing shorten the stoppage to 72 hours. The machine does not sit idle for the full eight weeks.

Loss per event is 72h × 45,000 = 3,240,000 THB. Avoided expected loss is 0.15 × 3,240,000 = 486,000 THB per year. Holding cost is 180,000 × 22% = 39,600 THB per year. The ratio is 12.3 times, which is an unambiguous hold. The pattern is Type 2, a single insurance unit. When someone asks why you keep a 180,000 THB part that gets used 0.15 times a year, this calculation is the answer.

Example 2 – a general purpose photoelectric sensor on a Tier C machine

Unit price 4,500 THB, λ = 0.8 per year, held in stock by the local distributor and delivered in two days. A Tier C machine is a standalone process with alternatives, so production itself carries on for those two days even without the part. What actually stops is the four hours the changeover takes, making the loss per event 4h × 1,200 = 4,800 THB. Avoided expected loss is 0.8 × 4,800 = 3,840 THB per year against holding cost of 4,500 × 22% = 990 THB per year, a ratio of 3.9 times.

On the ratio alone it looks like a hold. But for this part the tolerable downtime, which runs past two days, exceeds the procurement lead time of two days. Production never fully stops before the replacement arrives, so putting the part on your own shelf adds almost nothing to the downtime you prevent. Type 3 consignment at the distributor, or Type 5, do not hold, is the right answer.

This is the trap where “judging on the ratio alone means holding everything”. Even at a ratio of 3.9 times, the absolute difference is only a few thousand THB a year, which does not repay the administrative effort one more SKU creates – assigning a bin number, counting it at stocktake, maintaining its reorder point. A ratio misleads you when the absolute amounts are small. So before you compute the ratio, check whether procurement lead time exceeds the tolerable downtime. If it does not, there is very little downtime reduction available to buy with your own stock in the first place. That is exactly why the previous section said to read lead time first.

Solve the same formula for unit price and you get a break-even price per tier and per λ. Above that price, holding the part yourself is not economically defensible at that failure rate.

Break-even unit price = λ × hours of stoppage directly attributable × cost per hour of stoppage ÷ 0.22

Here is how to read it. A part on a Tier A machine justifies stock up to 1,473,000 THB even if it fails once a decade, at λ = 0.1. In practice almost nothing sitting on a Tier A machine hits that ceiling, which is the real meaning of that row – when in doubt on Tier A, hold it.

Tier C, by contrast, justifies 2,200 THB for a once-a-decade item and 21,800 THB even for one that fails annually. If you are keeping a part worth more than 21,800 THB as insurance for a Tier C machine, that cannot be explained economically. Tier B sits in the middle, with decisions clustering in the 87,000 to 870,000 THB band. That is the tier that genuinely needs debate, which also means it is where your review time should go.

Two cautions. First, the formula uses hours that already assume workarounds shorten the stoppage, which is why Example 1 uses 72 hours. Calculate as though the machine sits idle for the full eight weeks and avoided expected loss grows by an order of magnitude, and the answer becomes “hold everything”. Second, λ is the failure rate for one part. The 0.15 per year in Example 1 is set as the value for a single machine carrying that part. If you have 12 identical machines and want the probability that one of them fails, λ scales with the fleet. Whether one insurance unit can cover 12 machines depends on the probability of simultaneous failure and on recovery time.

Seven functions a spare parts inventory management system has to have

Once the three axes settle the design, you need the machinery to run it day to day. What you ask of a spare parts inventory management system looks quite different from the feature list of a general inventory package. Seven functions are non-negotiable.

Equipment BOM, the link between machines and parts, comes first. It is the structure showing which part is used in which unit of which machine, and how many. Without it, you cannot tell which spare parts become redundant when you replace a machine, and dead stock accumulates. With it, you can identify the parts a planned machine replacement will orphan before it happens. This is the single most common omission in general inventory software.

Linking issues to equipment and work orders comes second. When a part leaves the shelf, record which machine and which job consumed it. Without that record you can never compute λ. Whether the formula above is usable at all depends on whether this function survives contact with daily operations. Conversely, keep it up for one year and from the following year you can decide on real data.

Reorder points and minimum order quantities come third. This is the basic capability for putting Type 1 items on automatic replenishment, but in Thailand MOQ and lot constraints bite. Setting the reorder point of a lot-constrained part to a minimum simply means stock climbs by a full lot every replenishment. Being able to hold reorder point and order quantity independently is the requirement.

Alternates and interchangeable parts come fourth. Hold the relationships between functionally equivalent parts from different manufacturers, successor models, and superseded part numbers. Without it, a usable part sits on the shelf undiscovered because the number does not match, and an emergency shipment goes out. This matters most for electronic components, where EOL is frequent.

Storage location and conditions come fifth. Beyond a bin number, hold storage conditions as attributes – dry cabinets, ESD-safe shelving, shelves with expiry management for lubricants. Under the Thai climate conditions discussed below, whether you manage this changes how long parts last.

EOL and discontinuation flags come sixth. Mark parts the manufacturer has announced as discontinued, and decide whether to secure the quantity needed until that machine is replaced or switch to an alternate. With control equipment running 30 to 55 weeks, missing an EOL notice forces a machine replacement on someone else’s schedule.

Cross-site stock enquiry comes seventh. It is the precondition for Type 4 sharing – other sites’ stock has to be visible on the spot. If it is not, Type 4 is unavailable to you.

All seven sit on the same ground as the maintenance work history. Linking issues to work orders means connecting to how jobs are raised in the CMMS, the Computerized Maintenance Management System. How granular the register should be and how jobs should be raised is covered in how to choose an equipment maintenance management system 2026. Try to close the loop inside the spare parts module alone and issue records never get past “who took what”, and λ never appears.

The five cost layers, and how far to systemise maintenance parts inventory

Compare only the price of the parts and the price of the software when quotations arrive and you will decide badly. Spare parts management costs fall into five layers. Here are first-year figures for the model plant.

The key to reading this table is that layers 4 and 5 are what goes out of the business every year, so layer 3 cannot be judged expensive or cheap on its own. Companies hesitate over spending 420,000 to 1,200,000 THB on a system while paying 580,000 THB a year on layer 4 emergency shipments and 260,000 THB a year on layer 5 obsolescence. Comparing a layer 3 investment that lands mostly in year one against 840,000 THB leaving the business every year is a perfectly sound comparison. Layer 3 spans a wide range because the range is set by how much initial data cleanup you buy. Outsource the stocktake and part number rationalisation for 1,200 items, or push it through your own maintenance team on night shifts, and you move from one end of that range to the other.

So how far does Excel go? The line is clear. Excel is enough while you have a few hundred items at most, a single site, and no need to keep issue records on paper. If all you hold is bin number, quantity and unit price, Excel is faster and cheaper.

Excel stops working when any of three things happens. The first is wanting to link issues to equipment and work orders, because attaching machine, job, technician and timestamp to every issue line breaks aggregation in a spreadsheet structure. The second is multiple people editing at once, because three maintenance staff updating the same file across night and day shifts will overwrite each other and the quantities will stop reconciling. The third is wanting to query stock across sites, which file sharing cannot deliver.

The model plant has 1,200 items, 12 Tier A machines, and cross-site sharing on the table, so it belongs on the system side. That does not mean installing everything at once. Put in equipment BOM and issue linkage first, and configure reorder point automation only after a year of data has accumulated. Sequenced that way, layer 3 can start near the bottom of its range at 420,000 THB.

Five issues specific to a site in Thailand

Everything above applies to a factory in any country. A site in Thailand adds five more issues.

Import lead time and customs come first. The 4 to 10 weeks for direct supply from a Japanese manufacturer includes customs time. Documentation problems push you towards the top of that range. Confirming HS codes in advance, getting product descriptions right on invoices, and settling whether a certificate of origin is required – done in calm periods rather than in emergencies – is what actually shortens lead time. The more urgent the part, the sloppier the paperwork tends to be, and the later it arrives as a result.

The de minimis repeal comes second. From 1 January 2026, the 1,500 baht exemption is gone and shipments face 7% VAT and duty regardless of value. Sending small parts by air on demand has become more expensive. Designs that reduce the number of emergency shipments now pay off more than they used to.

The relationship with BOI privileges comes third. Under the Thailand Board of Investment machinery import exemption, machinery imported duty-free must be brought in within 30 months of the promotion certificate being issued, and must be used exclusively to produce the promoted products for more than five years. The point to note is that replacement parts bought in later years do not automatically fall inside that scope. The BOI page in question does not address spare parts explicitly. Whether spare or replacement parts qualify for exemption depends on the type of promotion, the wording of the certificate, and the timing of import. Each case needs confirmation with the BOI and with customs, and nothing in this article should be used as the basis for that judgement. It is not a question anyone can answer categorically.

Maintenance staff turnover and undocumented knowledge come fourth. Transfers and job changes among maintenance staff are said to happen readily in Thai manufacturing, and “only that person knows where that part is” turns directly into real losses. Basics work here – assigning bin numbers, verifying physical items with photographs, recording part names in Thai alongside English. Keeping the register in both Thai and English is extra effort, but it pays for itself at handover.

Heat, humidity and rainy season storage come fifth. Electronic components lose life to humidity, so control boards and amplifiers need dry cabinets or desiccators. Rubber, belts and O-rings degrade with age, so unless you record the purchase date and enforce first in, first out, you will create parts that are on the shelf and unusable. Lubricants and greases have expiry dates too. The 180,000 THB in layer 2 is what it costs to put those storage conditions in place. Drive the project on “cut the inventory value” alone and this layer gets deferred, and layer 5 obsolescence grows as a result.

Clearing out dead stock – what to do with 38% of your SKUs

In the model plant, 38% of SKUs have had no issue for 24 months or more, worth 1,300,000 THB. Working through them is not simply a matter of scrapping. It is the work of separating what can go from what is genuinely there as insurance.

The first thing to reconcile against is the equipment replacement plan. Parts for machines already removed, and parts for machines scheduled for replacement next year, are unambiguously redundant. With an equipment BOM you can extract them mechanically. Conversely, a part on a Tier A machine that will not be replaced for a decade is legitimate insurance stock even with no issue in 24 months. Make that call with the break-even prices from the earlier section. Deciding when to replace ageing machines in the first place is covered in replacing legacy equipment 2026. Where the replacement plan is unsettled, the spare parts cleanup stalls halfway too.

The second thing to reconcile against is the physical stocktake. Identifying dead stock is most efficient when it rides on a physical count, because you are handling every item anyway. Record the last issue date and the physical condition at the same time. Hardened rubber, corroded boards and expired lubricants only surface at that moment. How to run the count is covered in making physical inventory stocktaking efficient 2026. Spare parts counts differ from finished goods in having many line items and value spread thinly across them, so handhelds and labelling pay off.

The third element is how to book the write-down. This is where most plants stop. Writing off 1,300,000 THB of stock hits the period result. So the shop floor says “we might still use it” and finance leaves it on the books, and unusable parts keep occupying shelf space.

The workable approach is to write down in stages rather than all at once. Start with parts for machines already removed, then parts replaceable by an alternate, and leave the genuinely debatable items until last. Split across fiscal years, the impact on any single year’s result is diluted. In the model plant’s post-design target, dead stock falls from 1,300,000 THB to 450,000 THB. It does not reach zero. The remaining 450,000 THB is insurance stock that downtime cost justifies. Do not set a target of zero dead stock. Aim at zero and you will inevitably write off parts you should have kept.

Some parts can also be sold outside. Neighbouring plants running the same machines, used equipment traders, and manufacturer take-back schemes are the usual routes. The cash back is modest, but you avoid disposal costs and free up storage.

A 90-day sequence to get this running

Here is the order in which to turn the design into action. Cleaning up all 1,200 items before installing anything is a plan that never finishes.

Days 0 to 30 cover only the three-tier split and the identification of parts on Tier A machines. Manufacturing and maintenance assign an hourly stoppage cost to all 80 machines and confirm the 12 Tier A units. Then build a parts list for the main assemblies on those 12 machines – not for all 1,200 items, only what sits on Tier A. In parallel, attach a last issue date to every item from two years of issue history. That much works even if the inventory data lives in Excel.

Days 31 to 60 apply the formula to Tier A parts. Unit prices already exist, so what you need is λ and the hours of stoppage directly attributable. Take λ from issue history and manufacturer recommendations, and stoppage hours from the maintenance team’s judgement. Rough is fine. This step surfaces both the parts you should hold and do not, and the expensive parts you hold and should not. At the same time, fill in the lead time table by asking your main distributors and manufacturers. For control equipment, ask about EOL status in the same conversation.

Days 61 to 90 are where you actually order the Type 2 insurance stock and stop replenishing the items assigned to Type 5. Doing both at once is the point. Do only one and either inventory value rises or downtime risk does. In parallel, start recording equipment and work order against every issue. Begin on paper and in Excel even if the system is not in yet. That record is the raw material for next year’s λ.

Ninety days gets you that far. Reorder point automation, cross-site stock enquiry and an alternates master come later. The post-design picture for the model plant is a 12-month position, not a 90-day one.

What deserves attention in this table is that inventory value falls 29% while Tier A downtime waiting for parts falls 62%. That comes from cutting in one place and adding in another at the same time. Cut inventory uniformly and the value falls the same way, but downtime gets worse first. Add stock out of fear of stopping and downtime improves while holding cost rises. Moving both at once is what designing on three axes buys you.

Five common failures in maintenance parts inventory management

Finally, five patterns you see often in plants in Thailand.

Targeting inventory value reduction alone is the first. When the target is money only, the shop floor cuts where cutting is easiest, which means expensive slow-moving parts. Those are the Tier A insurance items. Six months later stoppages rise, emergency shipments rise, and spending goes up. If you set a value target, always pair it with Tier A downtime waiting for parts.

Setting a reorder point on everything is the second. Calculate and set reorder points and safety stock across all 1,200 items and the intermittent parts get meaningless numbers. In particular, a part that happened to move twice in the past year gets a high reorder point and automatic replenishment starts running. Reorder points belong to smooth and erratic items only.

Cleaning up from data without looking at the physical stock is the third. The book may say “in stock” while the physical item has been ruined by humidity. That is not unusual in Thai storage conditions. Always pair the data cleanup with physical verification.

Holding stock yourself without checking distributor stock is the fourth. As Example 2 showed, holding a part the distributor already stocks is double investment. Once a year, ask your main distributors which part numbers they keep, in what quantity, and at which location. The reverse also happens – a part you assumed the distributor kept on the shelf has quietly been dropped from their range.

Finishing the design without starting to record λ is the fifth. Estimates are fine for the first pass, but unless you start linking issues to equipment, year two and year three will still be estimates. A design is never finished in one go. The more real data accumulates, the sharper the line between what you hold and what you do not. That is why the 90-day plan puts issue recording at the end.

Summary

Production-material stocking formulas do not apply to spare parts, because demand is intermittent. Evaluate on turnover or days of inventory and you will scrap the parts you should hold and hold the parts you should scrap.

Use three axes instead. Read the demand pattern through SBC classification to judge whether it can be forecast, sort equipment into Tiers A, B and C by downtime cost to put a number on the loss from a stockout, and check whether procurement lead time exceeds the tolerable downtime. From those three, stocking resolves into five patterns – reorder point, a single insurance unit, distributor consignment, cross-site sharing, and not holding at all.

The money decision rests on whether avoided expected loss exceeds holding cost. Multiply the annual failure rate λ by the hours of stoppage directly attributable and the cost per hour, then divide by the holding cost rate. Run the numbers through and the picture is that Tier A parts should be held when in doubt, Tier C parts should be left with the distributor, and Tier B is where judgement is required.

What you need from a system follows from the same design. Equipment BOM and issue-to-equipment linkage come first, reorder point automation later. Install a feature-rich system with no data behind it and λ never appears, so the formula stays unusable. Design Tier A in 90 days and start recording issues. From there, taking 12 months to bring inventory value and downtime down together is the realistic path.

Where to get help reviewing your spare parts management

Which parts to hold, which to leave with the distributor, and which to scrap. That decision cannot be made without looking at the actual register and the actual state of the equipment. With the same 1,200 items, the answer differs between a plant with 12 Tier A machines and one with more than twice that. Whether the goal is compressing inventory value or shortening downtime, the first move is identical – sort the equipment into tiers and link issues to machines. TOMAS TECH implements production management and maintenance systems in factories in Thailand, and that work often starts with taking stock of an existing Excel register. It is perfectly fine to be at the stage of working out where to begin. Please get in touch through our contact page.

Frequently asked questions

What is a spare parts inventory management system?

It is a system for managing the stock of parts used in equipment maintenance while holding the link to the equipment itself. What separates it from inventory software that holds only item, quantity and bin number is the equipment BOM, which records which part is used in which unit of which machine, and the linkage of each issue to a machine and work order. With those two in place, you can see which parts become redundant when a machine is replaced, and you can derive a failure rate per part from real data.

How much spare parts stock should we hold?

There is no universal guideline. The judgement is made part by part, on whether avoided expected loss exceeds holding cost. Avoided expected loss is the annual failure rate multiplied by the extra hours the machine stays down because the part is missing and by the cost per hour of stoppage, and holding cost is the unit price multiplied by the holding cost rate, which is 22% per year in this article. On a Tier A machine in the model plant, a part that fails once a decade justifies stock up to 1,473,000 THB. On Tier C the limit is 2,200 THB. The right stocking level for the same part varies that much depending on the machine it sits on.

How is maintenance parts inventory different from production material inventory?

The demand pattern differs. Production materials move daily, so mean and standard deviation carry meaning and safety stock formulas apply. Maintenance parts show intermittent demand, sitting still for months and then moving all at once, and the mean describes nothing that actually happens. The metrics therefore differ too. Turnover and days of inventory are valid for production materials, while maintenance parts should be measured by downtime waiting for parts, the number of emergency shipments, and dead stock value. Run both on the same system settings and one of them will break.

Can we manage spare parts in Excel?

Excel is sufficient while you have a few hundred items at most, a single site, and no need to record issues on paper. It stops working in three cases – when you want to link issues to equipment and work orders, when several people update simultaneously, and when you want to query stock across sites. Once you pass 1,200 items and run identical machines at several sites, quantities in Excel stop reconciling. That said, even before a system goes in, you can start recording the link between parts and equipment in Excel.

How should we dispose of spare parts that have become dead stock?

Separate what can go from what is needed as insurance first. Parts for machines already removed or scheduled for replacement are unambiguously redundant, while an expensive part on a Tier A machine remains legitimate insurance even with no issue for 24 months. After sorting, verify physical condition during the physical stocktake and book write-downs in stages. Process parts from removed machines first and leave the debatable items for later. Split across fiscal years, the impact on any single year’s result is diluted. Do not set a target of zero dead stock.

How long does it take to get spare parts from Japan to a plant in Thailand?

Two to five days if the local distributor holds stock, two to four weeks for sourcing within Thailand, and four to ten weeks for direct supply from a Japanese manufacturer including air freight and customs. Control units containing semiconductors are a separate case and can take 30 to 55 weeks. On top of that, the 1,500 baht de minimis exemption was abolished on 1 January 2026, so shipments now attract 7% VAT and duty regardless of value. Sending small parts by air on demand has become less attractive than it was.

How much does a spare parts inventory management system cost?

Looking at the cost of the software alone leads to a bad decision. Spare parts management costs split into five layers – parts cost, storage, register and system, procurement and customs, and obsolescence and write-off. In the model plant’s first year, register and system runs 420,000 to 1,200,000 THB, storage 180,000 THB, procurement and customs 580,000 THB a year, and obsolescence and write-off 260,000 THB a year. The wide range on the system layer comes from the scope of initial data cleanup, and whether you outsource the part number rationalisation for 1,200 items or do it in house moves you from one end to the other. Layers 4 and 5 recur every year, so weigh them against a layer 3 investment that lands mostly in the first year.

References

• The True Cost of Downtime 2024, a Siemens and Senseye analysis, AEMT
• Syntetos, Boylan & Croston (2005) On the categorization of demand patterns, Journal of the Operational Research Society, 56, 495-503
• Svetunkov, Intermittent demand classifications – is that what you need? (16 July 2024)
• Thailand Board of Investment, Benefits after promotion, duty exemption on imported machinery
• GlobX, Electronic Component Lead Times 2026, updated 17 July 2026
• LexBangkok, Thailand Customs Duty 2026
• Japan Institute of Plant Maintenance, FY2024 Maintenance Practices Survey Report, an ongoing survey of maintenance and spare parts spending