How to calculate spare parts stock for 30 branches

Keeping the same set of parts at each of 30 branches is almost always more expensive and slower than managing a shared pool. Calculate spare parts stock for each compatible device group, then split it between the central warehouse and regional locations so a part can reach the site before the RTO expires.

Failure rate answers only how many replacements you will need on average. It says nothing about where the next failure will occur, whether two requests will coincide, or whether a shipment will make it past a closed road. The calculation therefore provides a baseline, while criticality, geography, and common causes add a deliberate reserve.

Stock planning starts with RTO, not the catalog

First separate devices by tolerable downtime, or expensive servers and replaceable office PCs will end up in the same queue. RTO sets the maximum time for restoring a function, not the time in which an engineer must start work. If a server must return to service in four hours, delivery of a part in eight hours misses the objective even when the supplier has formally met its contract.

Three classes work well for a network of 30 branches. Class A covers devices whose outage blocks payments, clinical work, access to a central system, or the operation of an entire branch. Their internal RTO may be measured in hours. Class B includes workstations and nodes with a limited workaround: move an employee to another desk, reroute a print queue, or temporarily use another terminal. Class C allows scheduled replacement after several days.

Do not assign a class based on device price alone. An inexpensive router power supply can stop a branch, while an expensive workstation may have an unused backup in the next room. Criticality belongs to the function and its available workaround.

Record four times for every group:

• detection and correct diagnosis;
• release of the part from stock and preparation for shipment;
• transportation to the specific branch;
• physical replacement, configuration, and verification.

Their total must be shorter than the RTO. If diagnosis and repair already consume three hours out of four, a central warehouse in another city will not help. You need a part on site, a complete standby node, or automatic failover. Inventory cannot fix an architecture that left no time for delivery.

NIST SP 800-34 defines RTO as the maximum time a resource can remain unavailable before the effect on business processes becomes unacceptable. The guide identifies spare equipment and support contracts as alternative strategies for meeting an RTO. I would add one practical qualification: count a contract as stock only after you have tested its promised delivery time on your own routes.

Count replacement failures, not every support ticket

The calculation needs the share of devices that actually required a specific part during the period. Total support tickets overstate demand because a reboot, configuration error, damaged cable, and motherboard replacement do not consume warehouse stock in the same way.

Collect a 12 to 24 month log with these fields: device model and revision, branch, failure date, symptom, confirmed cause, replaced part, repair result, and return date for the removed part. Mark damage caused by power surges, dust, overheating, water, and transportation separately. You cannot safely mix these events with random hardware failures.

For each part, calculate the observed annual rate:

annual_rate = replacements / device_years
device_years = sum(active_devices * active_days / 365)

If 300 identical desktop computers operated for a full year and 18 needed a compatible power supply, the observed rate is 18 / 300 = 0.06 replacements per device-year. If the fleet changed during the year, use device-years rather than the count on the last day.

Distinguish a device failure from part consumption. One repair may require a power supply and a cable, while a wrong diagnosis may write off a working module. Track a second metric, the share of returned parts for which the fault was confirmed. A low share means stock is disappearing because of weak diagnosis rather than unreliable equipment.

When your own sample is small, use manufacturer data or experience from a similar fleet, but record the conditions. IEC 61709 explicitly warns that failure rates depend on temperature, electrical load, and other stresses; the standard describes how to convert data from reference conditions to actual conditions, rather than providing universal base rates. A borrowed number from a presentation without its conditions is worse than a cautious estimate from your log.

Use a range for new equipment. For example, calculate low, working, and stress cases with rates of 0.02, 0.05, and 0.10. Replace the assumption with actual device-month data after the first six months. The budget then shows the uncertainty instead of hiding it behind one precise-looking number.

Zero observed failures do not mean a zero failure rate. For a rough upper bound at 95% confidence, use the "rule of three": divide 3 by the accumulated device-years. If 60 nodes operate for half a year without a replacement, the sample contains 30 device-years and the upper estimate remains about 0.10 failures per device-year. The fleet is still too young to remove its reserve.

Diagnostic data should change the repair process as well as the numerator. If an engineer often takes two possible parts to avoid a repeat trip, count both as temporarily unavailable until they return. For Class A, prepare a short diagnostic tree that distinguishes the power supply, board, and drive. Better diagnosis can reduce the required rotating pool more than another purchase.

The coverage horizon includes delivery and review

Stock must cover failures until replenishment is available again. Add the order review interval, receiving, and any delay that occurs regularly to the delivery time.

With continuous review, the warehouse orders a part as soon as it issues one. The baseline horizon is then close to the full replenishment time: order confirmation, supplier picking, transportation, customs procedures where applicable, incoming inspection, and put-away. With a review every two weeks, a failure may happen the day after an order, so stock must survive almost the entire additional two-week interval.

Use the required percentile, not the average lead time. If nine shipments arrive in three days and the tenth in twelve, the 3.9-day average is not suitable for a commitment to restore a critical system. For Class A, it is safer to calculate with the 90th or 95th percentile of delivery history and examine rare extreme events separately.

The baseline formula for expected consumption over the horizon is:

mean_failures = installed_base * annual_rate * coverage_days / 365
coverage_days = lead_time_percentile + review_interval + receiving_time

Compatibility changes installed_base. Do not combine devices merely because both are called "servers." Revisions within one series may use different power supplies, connectors, controller firmware, or supported memory lists. A stock group exists only when one part can restore any device in the group without breaching support conditions.

Repair time for the removed part also affects replenishment when you use a rotating pool. The horizon then ends when a tested module returns to available stock, not when a new part arrives. Do not call it a "five-day repair" when the service center takes twenty days end to end, including two shipments and receiving.

Trigger an order from inventory position, not the number of boxes on the shelf:

inventory_position = on_hand + confirmed_inbound - reserved - backorders
reorder_when inventory_position <= expected_lead_time_demand + safety_stock

A part reserved for an overnight repair is already unavailable even though it remains physically in the warehouse. Do not count an order that the supplier has not confirmed as inbound stock for Class A. These two rules remove a common illusion of sufficient inventory.

The Poisson distribution turns the average into a quantity

Rounding expected consumption upward is not enough because the average does not cover coincident failures. For rare independent events, the Poisson distribution provides a simple way to choose the minimum number of parts for a target probability of avoiding a shortage.

Let the mean number of replacements during the horizon be μ. Choose a service level, such as 95% for Class B and 99% for Class A. Then find the smallest integer s for which the probability of no more than s requests during the horizon reaches the target:

P(X <= s) = sum(exp(-mu) * mu^k / k!, k = 0..s)
choose smallest s where P(X <= s) >= service_level

A spreadsheet can perform the check with one function:

=POISSON.DIST(spares, mean_failures, TRUE)

This calculates safety stock for demand variability; it does not predict exact dates. If μ = 0.94, one part covers only about 76% of horizons, two cover about 93%, and three cover about 98%. Rounding the mean to one part leaves too many periods with a shortage.

The model needs a reality check. Failures after a thunderstorm, a batch-wide defect, and simultaneous wear in identical drives are not independent. The Poisson distribution understates the tail risk in those cases. Seasonality also needs separate rates because summer heat in a poorly cooled room and winter logistics can raise consumption and replenishment time at once.

Do not set 99.9% for every screw. As the target rises and the mean falls, inventory grows sharply relative to consumption. The service level must follow from the RTO, available workaround, and shortage cost. Class C may permit ordering after failure; Class A may need a part at every critical site even when network-wide demand asks for only two.

A 30-branch example produces three different answers

One network calculation shows why a single rule such as "one part per ten devices" fails. Suppose 30 branches use 60 server nodes, 300 desktop computers, and 90 all-in-one computers. The figures below are illustrative and must be replaced with your organization's failure log and actual lead times.

Mean demand for the server power supply is low: 60 × 0.08 × 10 / 365 = 0.13. One part provides about a 99% probability that demand will not exceed stock. But a Class A server with a four-hour RTO cannot wait ten days for central delivery. Placement decides the final answer: keep one compatible module at each site without a standby node, or test failover to the standby system in advance.

For PC power supplies, 300 × 0.06 × 19 / 365 = 0.94. Three parts in the central warehouse provide a service level above 95% under independent demand. If every branch can temporarily move a user to another desk, there is no reason to distribute 30 power supplies, one per city. A central pool of three plus working spare workstations in branches usually fits the requirement better.

For all-in-one boards, the mean is 90 × 0.04 × 40 / 365 = 0.39. Two boards clear 95% with room to spare. Yet the 40-day horizon points to another issue: the revision may change during that time, while an expensive board locks up cash. If the RTO allows one day, a complete replacement all-in-one in a regional pool is often more useful than a board. The engineer swaps the device faster and sends the difficult repair to the center.

Test how sensitive the result is to input error. If the PC power-supply rate is 0.09 rather than 0.06, the 19-day mean rises from 0.94 to 1.41 and the 95% target requires four parts. If delivery falls to five days with daily review and one day for receiving, the horizon falls to six days, the mean becomes 0.44, and baseline stock falls to two. This comparison shows when buying boxes is worthwhile and when faster replenishment is the better purchase.

Now add placement. Suppose six remote branches cannot be served reliably in less than a day, while the rest receive a shipment from the central warehouse within 6 to 12 hours. Establish a regional minimum for critical nodes only at the six remote sites, or at two or three hub cities if final delivery fits the RTO. Keep the remainder shared. This may raise the server module count from one statistical unit to several operational units, but it buys recovery time rather than an abstract percentage.

Central stock and regional kits cover different risks

A central pool absorbs random demand better because one part is available to all 30 branches. A local kit wins only on time. A mixed arrangement usually produces fewer shortages for the same budget: a minimum near critical devices and the rest in a shared pool.

Split inventory into three layers. Keep on site whatever is physically necessary to meet the local RTO. Store compatible parts for several nearby branches at a regional warehouse. Leave expensive, low-demand items, the repair pool, and surge stock at the center.

Do not divide a calculated three parts proportionally among 30 cities. You cannot execute a fractional answer, and rounding upward creates 30 units. Set mandatory local minimums first, then recalculate the demand left for the central pool. A local part cannot be counted as available to the entire network at the same time.

Maintain a decision table for every item:

A warehouse without an owner becomes a locked cupboard. A part can appear in stock while sitting in a locked office, already installed without an issue transaction, or missing its mounting kit. Round-the-clock support requires 24-hour access, labeling, packaging for onward shipping, and clear authority to open the reserve at night.

Do not use disassembly of a working device as a permanent substitute for stock. Cannibalization can save an RTO, but it creates a second unfinished repair and hides actual part consumption. When it happens, open two linked tickets: one for restoring the critical function and another for returning the donor device. Otherwise the records show one successful repair a month later while the fleet has lost a working unit.

A tested substitute can combine several small pools, but you must prove compatibility before an incident. Install the part in every listed revision, update firmware where needed, run a load test, and save the result with the item record. A supplier's statement that it "should fit" does not reduce the number of stock groups.

A common cause breaks clean statistics

You need a reserve above the Poisson result when one event can disable several identical devices or cut several branches off from supply. This is not a percentage added for comfort. It is a separate scenario with a cause, scale, and protection.

Test at least four scenarios: a power surge damages several supplies at one branch; heat raises the number of fan failures; a defective batch begins failing after a similar service life; severe weather closes the main delivery route. Estimate the largest plausible simultaneous consumption and the time needed to restore logistics for each one.

Duplicating the same part does not always provide protection. If two power supplies connect to one faulty uninterruptible power supply, the replacement may burn out immediately after installation. Remove the common cause before consuming the reserve. For a batch defect, storing five modules from the same batch creates false confidence; use another batch or a tested compatible replacement.

Place part of the critical pool outside the common risk zone. If flooding or a road closure can affect the warehouse and its branches at the same time, the second stock location needs a different route. NIST advises separating an alternate storage site far enough from the primary location to reduce exposure to the same threats. The same reasoning applies to spare parts.

Add the stress scenario to the figures separately:

final_stock = poisson_stock + local_RTO_minimum + common_cause_reserve

Do not add the components mechanically if they cover the same risk. A local minimum may already cover one failure in the stress scenario. Walk through the event in time: what fails, which part leaves first, when replenishment arrives, and where the next unit sits. This review quickly finds double counting.

Downtime cost sets the upper stock limit

The last unit is justified when the expected loss from a shortage exceeds its full ownership cost. Purchase price alone says little about the decision. An expensive, rarely used board may be more economical than a cheap part if the first stops a critical process and the second can arrive tomorrow.

Calculate annual ownership cost: purchase or depreciation, storage, inventory counting, testing, insurance, obsolescence, and disposal. Add repair and transportation for a rotating pool. For a shortage, estimate employee downtime, uncompleted transactions, urgent delivery, an engineer visit, and contractual consequences where applicable.

Compare neighboring options, such as three versus four parts, rather than "stock" versus "no stock." The fourth part reduces only the remaining probability of more than three demands during the horizon. Multiply that reduction by the loss from the corresponding downtime. If the result is lower than the fourth part's annual cost, keep three or change the recovery design.

A supplier contract does not automatically reduce warehouse stock to zero. It transfers some inventory and logistics risk to the supplier if the contract names a delivery time, compatible items, request hours, and responsibility when a part is unavailable. Check whether the clock starts when the ticket is logged or after remote diagnosis. The difference can consume a short RTO.

Compare three ways to restore one function: a local part and on-site repair, a replacement device followed by central repair, and urgent contract delivery. For each one, add the minutes for diagnosis, technician availability, configuration transfer, and data verification as well as the money. The cheapest warehouse option can be the most expensive way to restore service.

Be especially cautious when buying parts for equipment near retirement. A large last-time purchase is popular because it promises independence from the supplier. In practice, the stock often outlives the fleet: migration plans accelerate, boards age on shelves, and firmware and revisions diverge. Combine a limited final order with a replacement-device plan and a date for disposing of the remainder.

IEC 62550 treats spare-parts provisioning as a support activity that must sustain continuity of system operation. That framing is useful because you cannot optimize the warehouse separately from repair, logistics, and required availability. The lowest cost of boxes on shelves is not the lowest support cost.

Recalculation and drills keep stock from expiring

Recalculate Class A items monthly and the rest at least quarterly, as well as after any material change in the fleet, route, or failure rate. A formula without updates turns last year's accuracy into today's error.

At review, reconcile the installed base, device-years, actual replacements, percentage of confirmed faults, percentile lead time, current balance, open orders, and revision compatibility. Look separately for inactive parts and urgent purchases made while the system showed stock on hand. The second case almost always points to bad placement, access, or record quality.

Every number needs a source and a date. Take the installed base from the current asset register, warehouse issues from inventory records, confirmed causes from closed tickets, and lead times from order and receipt dates. If the systems disagree, define one precedence rule and correct differences at review. Otherwise the formula will merely multiply incompatible data with great care.

Run an overnight recovery drill for one critical node every six months. The on-call engineer must identify the part, obtain it, deliver it or take it locally, perform the documented replacement, and restore service within the RTO. Time every stage. Only this test will uncover a sealed box containing the wrong revision.

For a network in Kazakhstan, distance and seasonal road conditions make route testing part of technical support. GSE.kz combines hardware manufacturing in Kazakhstan, systems integration, round-the-clock support, and a nationwide service network, so fleet design can connect compatibility, supply, and recovery in one model without pretending that one warehouse number solves everything.

Approve not only a quantity for each item, but a reason: "two units for 95% of demand over a 19-day horizon," "one at the remote branch for a four-hour RTO," or "one from another batch against a common defect." If the owner cannot explain a unit's purpose, nobody can confidently replenish or safely dispose of it. A sound stock plan reads as a set of testable decisions rather than one number in an annual budget.