Choose your reading route
eduKateSG · Why Mathematics? · Spare Parts Planning: Poisson Demand, Lead Times, Service Levels and Stockout Risk
Start with the question, audit the worked example, or jump to the assumptions and practice. Every chapter stays open for uninterrupted reading.
Understand the question
Define the quantity and follow the mechanism.Route 2 · Chapters 8–14
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 15–21
Check assumptions
See where design and calibration matter.Route 4 · Chapters 22–28
Explore applications
Connect the method to a concrete investigation.Route 5 · Chapters 29–48
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
Why is mathematics important in spare-parts planning? It converts uncertain demand and replenishment lead time into a distribution of possible need, then connects that distribution to clearly named service measures, stockout consequences and obsolescence. The goal is a transparent decision, not a promise that one forecast will occur.
Did You Know? A “95% service level” can mean several different things. Probability of no stockout and fraction of units filled immediately are different measures.
Full chapter index
Chapters 1–7 · Understand the question
Chapters 8–14 · Reproduce the example
Chapters 15–21 · Check assumptions
Chapters 22–28 · Explore applications
Chapters 29–48 · Study and verify
- Guidance for students and parents
- Common misconceptions
- Connection to recent mathematics articles
- Practice questions
- Frequently asked questions
- A reporting checklist
- A final audit
- Useful next reading
- Inventory position versus stock on hand
- Order quantities and review periods
- Forecast evaluation
- New parts with little history
- Common-cause supply disruption
- Emergency procurement
- Service equity
- Carbon and waste
- Dashboard design
- An end-to-end stock study
- Demand during quality holds
- Revision and interchangeability control
CHAPTER 1 OF 48 · Understand the question
1. Why spare-parts planning needs mathematics
A spare on a shelf prevents downtime only if it is the right item, in the right place, before demand arrives. Too little stock creates backorders and service risk; too much ties up money, space and ageing inventory. Mathematics connects demand counts, procurement or repair lead time, service measures and uncertainty. It turns an argument about a single forecast into a range of possible lead-time demand, while preserving engineering questions about criticality, substitution and obsolescence.
A demand may mean an issued part, an approved requisition, an actual failure replacement or a replenishment order. These are not interchangeable. Emergency transfers, duplicate requests and returns can inflate transaction counts. Define the unit of issue, location, asset population and time stamp. Separate consumable use from repairable exchange. A clean demand definition is the foundation of every rate, distribution and service calculation.
Suppose 24 qualifying demands occur over 12 months of stable exposure. The observed mean is two demands per month. This is an estimate, not a promise of two every month. Record fleet size, operating hours and policy changes because calendar demand may rise when more assets enter service. An exposure-based rate can be more portable than a calendar average. Report the observation window and uncertainty.
Stock must cover demand while replenishment is unavailable. With constant rate lambda per month and deterministic lead time L months, a simple Poisson model has mean lead-time demand lambda L. At two demands per month and 1.5 months, the mean is three. If lead time is variable or demand depends on it, the distribution is a mixture rather than a single Poisson with the mean inserted casually.
A Poisson count model assumes events occur independently at a constant rate over the defined exposure, with probability governed by the interval length. It can be a useful baseline for low-count demand. It is unsuitable when failures cluster, the fleet grows, planned campaigns create bursts or one event consumes several parts. Compare mean and variance, plot time order and examine operational causes before relying on it.
For Poisson mean three, P(D=0)=e^-3, approximately 0.0498. This means about a five percent modelled chance of no qualifying demand during the lead-time interval, under the assumptions. It does not mean a five percent chance of stockout. Stockout depends on the stock position and the whole demand distribution. State the event clearly whenever reporting a probability.
| Quantity | Value |
|---|---|
| Demand rate | 2 per month |
| Lead time | 1.5 months |
| Mean lead-time demand | 3 |
| P(no demand during lead time) | e^-3 = 0.0498 |
If the relevant inventory position is S, a simple cycle-service calculation uses P(D≤S) for lead-time demand D, with boundary conventions matched to the policy. For mean three, P(D≤5) is about 0.916. This is an illustrative probability of demand not exceeding five, not a universal fill rate. Recalculate with the actual reorder rule, outstanding orders, backorders and issue quantities.
Cycle service level is commonly the probability of no stockout during a replenishment cycle. It treats a cycle with one unit short and a cycle with many units short as the same failure. That can be appropriate for some decisions and incomplete for others. Define the cycle and inventory position. Do not quote a service percentage without naming the measure, because different formulas can produce similar-looking numbers that answer different questions.
Fill rate measures the fraction of demand filled immediately from stock, usually in units. It depends on expected shortage, order quantity and policy. A system can have a modest probability of at least one stockout but still fill most units if shortages are small. Conversely, a high cycle service level can coexist with severe rare shortages. Choose the measure from user consequence and explain the denominator.
A backorder records demand that waits; lost sales or service cancellations may disappear from the ledger. Measure time to fulfil as well as count, especially for critical repairs. Priority rules can make average delay look good while a small class waits too long. Queueing, supplier lead time and repair capacity can interact. Inventory service is a time and consequence problem, not only a shelf count.
Supplier lead time includes ordering, manufacture, transport, receiving and inspection. Use its distribution when variability is material. Demand during a long lead time is naturally greater, so simulating demand conditional on each sampled lead time preserves the relationship. A mean lead time of 1.5 months can hide rare six-month delays that dominate stockout risk. Track promised and actual dates and separate controllable process delay from external disruption.
Many service parts have long runs of zero demand separated by nonzero bursts. Forecast occurrence and size separately when appropriate. Classical moving averages can lag after a fleet change and give fractional forecasts that are expectations rather than order quantities. Compare simple, Croston-type and count models using rolling evaluation. No method removes the need to understand whether zeros mean true absence, missing records or suppressed demand.
When count variance greatly exceeds the mean, a Poisson model may understate the tail. Common environments, maintenance campaigns, supplier defects or cascading failures can cluster demand. Negative-binomial, compound or scenario models may fit better. Preserve event links so a campaign consuming ten units is not mistaken for ten independent failures. Tail risk matters most when stockouts have large consequences.
| Measure | Question |
|---|---|
| Cycle service level | probability of no stockout in a replenishment cycle |
| Fill rate | fraction of demand filled immediately from stock |
| Backorder probability | chance an arriving demand waits |
| Expected shortage | average units short in the defined period |
Demand changes as installed base, utilisation and age change. A rate per active unit or operating hour can be combined with a fleet forecast, but the forecast has uncertainty. New designs may use different parts while older assets remain supported. Cohort models help distinguish ageing from expansion. Do not extrapolate a stable twelve-month count across a planned rollout without adjusting exposure and configuration.
A repairable spare cycles through installed, failed, transport, workshop, test and ready states. Pool size depends on failure arrivals and repair turnaround, not only purchase lead time. Condemnations create replenishment demand. A Markov, queueing or simulation model can represent the pipeline and shared repair resources. Treating a rotatable as a consumable ignores returns and can seriously distort stock needs.
Temporary substitution or cannibalisation may restore service but creates tracking, configuration and later replacement obligations. Compatibility must be documented by authorised engineering, not inferred from similar names. Model approved substitutes as alternative supply paths with their own lead times and risks. Unrecorded cannibalisation makes one location appear well supplied while moving the shortage elsewhere.
Central pooling can reduce total safety stock when demands are not perfectly correlated and transport is fast enough. Local stock can reduce response time and disruption risk. Compare network policies with demand correlation, transfer time, customs and priority rules. A pooled total does not ensure fair access. Simulation can show location-specific waiting and emergency transfers rather than only aggregate inventory.
A low-cost part can be critical if its absence stops a high-value or safety-related function. Classify parts by consequence, redundancy, approved substitution, detectability and replenishment risk. Criticality informs service objectives and review effort; it does not mechanically determine a quantity. Mandatory safety and regulatory controls remain constraints. Price alone is an inadequate priority score.
Holding more stock raises exposure to design change, expiry, corrosion and unsupported technology. Include expected write-off, preservation inspection and storage conditions. A last-time buy is a lifecycle decision involving future fleet plans, repair capability and redesign options. A demand model fitted to history cannot know that a platform will retire unless the scenario includes it. Report horizon and configuration assumptions.
Observed issues can understate true demand when the warehouse has no stock: users may wait, borrow, substitute or stop requesting. This is censored demand. Link requisitions, issues, backorders, cancellations and maintenance records. Correct duplicate and reversed transactions without erasing the audit trail. Mark periods when policy or catalogue numbers changed. Forecast accuracy cannot exceed the meaning of the underlying ledger.
The U.S. Government Accountability Office has repeatedly reported that inaccurate demand and lead-time estimates can produce both shortages and excess spare parts. Read a GAO report on spare-parts planning and GAO work on lead-time estimates. These reports support the planning principle; local quantities still require current data, mission context and authorised decisions.
Parameter uncertainty, policy change and supply disruption all affect stock. Generate predictive distributions rather than a point forecast, and show scenarios for demand rate, lead-time tail and fleet size. A percentile stock target reflects a chosen service trade-off, not certainty. Sensitivity analysis identifies whether better demand data, supplier improvement or approved substitution would reduce risk most effectively.
Inventory models can minimise holding, ordering, shortage and downtime costs, but the inputs and constraints encode values. For critical service, impose service or risk requirements first. Include budget, shelf capacity, minimum order, package size and repair capacity. A mathematical optimum outside operational feasibility is not a solution. When cost estimates are controversial, display a Pareto frontier of stock and service rather than one hidden weighted total.
Simulate demand times and quantities, lead times, order rules and backorders. Validate the simulation against Poisson formulas in a simple deterministic-lead-time case. Then add variable lead time, clustering or repair returns. Track cycle service, fill rate, expected shortage, waiting time, emergency shipments and expired stock. Use common random numbers when comparing policies so differences are easier to detect.
Use fictional spare demands for a school device fleet. Estimate a monthly rate, calculate Poisson lead-time probabilities and compare stock levels. Then introduce a campaign that creates clustered demand and see how the Poisson assumption changes. No real purchase should follow from the exercise. The lesson is that a forecast is a distribution tied to a mechanism and service definition, not a magic number.
Create monthly rows with qualifying demand, active fleet, operating exposure, stockouts and actual lead times. Compute rates and a lead-time-demand table. Use POISSON.DIST carefully for exact and cumulative probabilities and label the inventory boundary. Add a service-measure sheet that distinguishes cycle service and fill rate. Graph demand, variance and supplier delay. Keep configuration and policy changes visible.
Validate transaction states and build an event ledger. Fit baseline Poisson and overdispersed models, then perform rolling forecast evaluation. Simulate replenishment rules with fixed seeds and unit tests for zero demand, simultaneous orders and long lead times. Return distributions for service and inventory, not only means. Store data version and policy parameters. Review outputs with logistics and engineering specialists before action.
Spare-parts mathematics appears in transport, aviation, healthcare technology, utilities, manufacturing, defence and computing infrastructure. Roles combine statistics, supply chain, operations research, engineering and information systems. Mathematics expands understanding but does not guarantee a job or salary. Students can explore probability and simulation, then choose education pathways that fit their strengths and current verified admissions requirements.
Begin with the meaning of one demand and one lead time. Students should calculate lambda L and e^-3 by hand before using a spreadsheet. Parents can ask why a service level needs a name and why a mean is not a stock quantity. Use simulated inventories. Encourage the learner to explain both shortage and excess risk; balanced reasoning is more valuable than chasing a single high percentage.
Average monthly demand is not the amount needed every month. P(D=0) is not stockout probability. Cycle service level and fill rate are not interchangeable. Mean lead time does not capture its tail. Zero issues can reflect stockouts rather than no demand. Repairable pools are not ordinary consumables. Expensive parts are not automatically most critical. More stock does not remove obsolescence or common-cause risk.
The four most recent companions are Renewal Theory, Maintainability Engineering, Availability Engineering and Reliability Allocation. Each answers a neighbouring question, so use the one whose mechanism matches the decision. The Mathematics Learning Hub connects these ideas to probability, statistics, algebra, computing and responsible modelling.
1. Define a qualifying demand. 2. Calculate 24/12. 3. Find lambda L for 2 and 1.5. 4. Explain e^-3. 5. Distinguish cycle service and fill rate. 6. Give one overdispersion cause. 7. Explain censored demand. 8. Describe a rotatable pipeline. 9. Name one obsolescence cost. 10. Propose a variable-lead-time simulation.
### Is Poisson always suitable for spare demand? No. Check rate stability, independence, dispersion and operational causes.
### Is a 95% service level unambiguous? No. Name the service measure and denominator.
### Why model lead-time demand? Replenishment cannot cover demand that occurs before the order arrives.
### Can more stock guarantee service? No. Compatibility, common causes, repair capacity and logistics also matter.
Define part, unit of issue, fleet, location, demand event and horizon. State demand model, exposure, lead-time distribution, inventory policy and service measure. Report data gaps, stockout censoring, repair returns, substitutes, criticality and obsolescence. Show uncertainty and scenarios. Recalculate exact probabilities and preserve units. Link authoritative evidence accurately. Avoid presenting an illustrative stock level as procurement advice.
Trace each count to a transaction and maintenance need. Verify 2 times 1.5 equals 3, e^-3 is about 0.0498 and P(D≤5) is about 0.916. Check cumulative-probability boundaries. Inspect long supplier delays and clustered demands. Reconcile repair returns and condemnations. Test simulation edge cases. Confirm service claims use the correct measure and that no unsupported purchase recommendation remains.
Continue with GAO reports on spare-parts forecasting and lead time, then review NIST probability guidance. Read renewal theory for repeated replacements, maintainability for repair delays and availability for service impact. The Mathematics Learning Hub connects Poisson probability, forecasting, optimisation and simulation.
Stock on hand counts physical ready units. Inventory position usually adds outstanding replenishment and subtracts backorders. Reorder rules commonly act on position, not shelf count. Mixing the two can trigger duplicate orders or delayed action. Repairables need additional states for unserviceable and work-in-progress units. Define every inventory state and reconcile it to transactions. A probability calculation using S is meaningful only when S matches the policy boundary.
Continuous review can order when position crosses a point; periodic review waits until the next review date. Periodic review must cover demand during lead time plus the review interval. Minimum order quantities and pack sizes create overshoot. Compare policies with the actual calendar rather than a textbook reorder point alone. A monthly review can dominate a short supplier lead time, so omitting it understates risk.
Evaluate forecasts on rolling origins that mimic the decisions they would have supported. Use count likelihood, scaled errors or service outcomes suited to intermittent demand; percentage errors fail when actual demand is zero. Compare against a simple baseline. A method that predicts means well may still underestimate tail stockouts. Separate forecast error from inventory-policy performance so the team knows which component needs improvement.
A new part may have no direct demand record. Use engineering analogy, bill-of-material exposure, test evidence and scenario ranges, while documenting uncertainty. Pooling from similar parts requires a reasoned similarity definition. Update quickly as field data arrive. Avoid assigning a precise Poisson rate from one event. For critical items, design and supplier risk may justify temporary protection while evidence matures, subject to authorised review.
A single supplier, material or transport route can affect several parts at once. Independent lead-time models miss this correlation. Create scenarios for factory interruption, customs delay or quality hold and identify approved alternatives. Extra stock can buffer some disruptions but not wrong configuration or systemic defect. Portfolio risk matters when many individually modest exposures share one source. Diversification decisions need engineering qualification as well as arithmetic.
Expediting can shorten shortage duration but adds cost, uncertainty and sometimes quality risk. Model it as an alternative action triggered by stock position or priority, with a distribution of delivery time. Historical expedite data may be selected toward the hardest cases, so interpret carefully. Measure frequency and root causes; routine emergency buying can hide poor planning. Ensure incoming inspection and configuration control are not bypassed.
When stock is scarce, allocation rules determine who waits. First-come-first-served, criticality priority and reserved pools produce different location outcomes. Report waiting and fill by user class rather than only network average. A high aggregate fill rate can conceal persistent deprivation of a small site. The policy should be explicit, reviewable and consistent with service obligations. Mathematics can reveal distributional effects without deciding values silently.
Emergency air freight, excess manufacture and obsolete disposal have environmental costs. Include physical measures such as kilograms shipped, expedited distance or scrapped units alongside financial cost. Repair and pooling can reduce material use but may add transport. Scenario analysis makes these trade-offs visible. Do not claim an environmental optimum from a model that counts only holding cost. The system boundary should cover the outcomes used in the decision.
A useful dashboard shows demand exposure, backorders, lead-time distribution, service measure by name, repair pipeline and obsolescence risk. It displays cohort or configuration changes and data freshness. Avoid a single green percentage that combines incompatible parts. Provide drill-down to the transaction definition and uncertainty. Alerts should be tied to an action owner and threshold rationale. Visual clarity is part of mathematical accountability.
Define demand and service, clean transactions, estimate exposure-adjusted counts and lead-time distributions, then test Poisson assumptions. Calculate a baseline and validate it by simulation. Add variable lead time, clustering, repair returns, constraints and obsolescence. Compare policies on named service measures, cost and waiting. Review criticality and substitutes with engineering. Document the final stock decision, uncertainty and triggers for revision. This chain keeps a stock number connected to evidence.
A quality hold can make physical stock unavailable even when the ledger shows units on hand. Create a quarantined state and exclude it from ready inventory until authorised release. If a suspected defect affects installed and spare units together, demand may rise exactly when supply becomes unusable. Model this common-cause scenario rather than assuming stock and demand are independent. Record inspection capacity, release criteria and replacement source. This case demonstrates why serviceable condition and configuration matter as much as quantity.
Part numbers can split or supersede over time. A later revision may be backward compatible, conditionally compatible or prohibited for older configurations. Build an approved interchangeability graph from engineering records and attach effective dates. Forecast at the level that can actually satisfy demand, while retaining traceability to the issued item. Merging numbers without approval overstates supply; keeping every superseded number isolated can understate it. Inventory mathematics depends on an accurate technical identity.
Extended practice laboratory
Practice case 1: Estimate a rate
Setup: Use 24 qualifying demands over 12 months.
Task: Calculate the observed mean rate.
Checking target: Obtain 2 demands per month with uncertainty.
Practice case 2: Lead-time demand
Setup: Use rate 2 per month and lead time 1.5 months.
Task: Calculate Poisson mean lambda L.
Checking target: Obtain mean lead-time demand 3.
Practice case 3: Poisson probability
Setup: Use mean 3.
Task: Calculate P(D=0)=exp(-3).
Checking target: Obtain about 0.0498.
Practice case 4: Cumulative service check
Setup: Use a stock level and Poisson mean 3.
Task: Calculate P(D<=S).
Checking target: Interpret as a cycle-service probability under assumptions.
Practice case 5: Separate service measures
Setup: Compare probability of no stockout with fill rate.
Task: Explain why they differ.
Checking target: Do not label both simply service level.
Practice case 6: Add variable lead time
Setup: Simulate supplier lead time with demand.
Task: Form the lead-time demand distribution.
Checking target: Do not substitute mean lead time when variability matters.
Practice case 7: Handle intermittent demand
Setup: Plot zero and nonzero periods.
Task: Compare Poisson assumptions with observed dispersion.
Checking target: Use a richer model when counts are clustered.
Practice case 8: Include repairable spares
Setup: Model failed-unit return and repair turnaround.
Task: Track the repair pipeline.
Checking target: A rotatable pool is not ordinary consumable stock.
Practice case 9: Audit criticality
Setup: Classify consequence, substitution and procurement risk.
Task: Set different evidence and service goals.
Checking target: Do not let price alone define importance.
Practice case 10: Write the conclusion
Setup: State demand, lead time, service measure and horizon.
Task: Report uncertainty, obsolescence and alternatives.
Checking target: Avoid treating one forecast as certain demand.