Capacity Planning Under Variability and Uncertainty

Introduction

Capacity requirement planning (CRP) is often introduced as a comparison between required workload and available productive hours. That comparison is necessary, but it is not sufficient when production systems operate under uncertainty. Demand fluctuates, processing times differ between jobs, machines fail, employees are absent, materials arrive late, and output may be rejected because of quality losses. A plan that is feasible under average conditions may fail repeatedly in actual operation.

Capacity planning under variability therefore concerns not only how much capacity is required, but also how much capacity must be protected against uncertainty, where that protection should be located, and how much risk management is economically justified. The central managerial problem is to balance utilization, responsiveness, cost, reliability and resilience.

This chapter develops a risk-based approach to capacity planning. It combines deterministic workload calculations with probability distributions, scenario analysis, sensitivity analysis, safety capacity, queuing concepts, simulation and supply-disruption analysis. The emphasis is on postgraduate-level interpretation: managers must understand not only the calculations, but also the assumptions, model limitations and organizational consequences of capacity decisions.

Learning Objectives

After studying this chapter, the reader should be able to:

1. distinguish variability, uncertainty, risk and ambiguity in capacity planning;
2. calculate effective capacity after considering availability, performance, quality, absenteeism and yield loss;
3. evaluate demand, processing-time, downtime and supply-disruption scenarios;
4. explain why high utilization increases waiting time and reduces operational responsiveness;
5. determine safety capacity using service-level and risk-based reasoning;
6. apply sensitivity analysis and basic simulation logic to capacity decisions;
7. formulate an integrated capacity plan and assess its managerial, ethical and governance implications.

1. The Nature of Variability and Uncertainty

Variability refers to observable differences in events or operating conditions. Examples include different processing times for different products, daily fluctuations in demand and variation in machine repair duration. Variability may be represented statistically when historical data are available.

Uncertainty refers to incomplete knowledge about future conditions. A new product may have no reliable demand history, or the probability of a geopolitical supply interruption may be difficult to estimate. Risk exists when uncertain outcomes and their probabilities can be described sufficiently to support decisions. Ambiguity exists when alternative interpretations, models or probability estimates are plausible.

Capacity decisions are affected by several interacting sources:

A key insight is that sources of uncertainty do not simply add independently. For example, a material shortage can create idle labor, while an urgent demand peak can increase overtime, fatigue and quality losses. Consequently, capacity should be modeled as a system rather than as a single number.

2. From Rated Capacity to Effective and Risk-Adjusted Capacity

2.1 Basic capacity definitions

Design capacity is the theoretical maximum output under ideal conditions. Effective capacity is the maximum output under stated operating conditions, including planned breaks, setups, maintenance and normal organizational constraints. Actual output is the output achieved in a particular period.

If a resource has N identical machines, H scheduled hours per machine, and r units per hour, its theoretical time capacity is:

where C_d is design capacity in units per planning period. This equation assumes that all machines are available, product mix is constant, quality is perfect and the rate is stable. These assumptions are rarely realistic.

A time-based representation is usually safer for mixed products. If job type i requires standard time t_i and planned quantity Q_i, total workload is:

where W is required workload in labor-hours or machine-hours, m is the number of job types and S is setup and other planned time in the same units. Capacity is feasible only when available productive hours are at least equal to workload, subject to timing and skill constraints.

2.2 Effective capacity with losses

Let scheduled hours be H_s. If planned availability is A, performance relative to standard rate is P, and first-pass yield is Y, then effective productive hours expressed as good-output-equivalent hours are:

Here, A, P and Y are proportions between 0 and 1. Availability excludes planned or unplanned time during which the resource cannot run; performance captures speed losses; yield captures the proportion of output accepted without requiring replacement or rework. The formulation is conceptually related to overall equipment effectiveness, although exact definitions must be aligned with the organization’s measurement system.

If the standard output rate is r, good output capacity is:

Units are consistent: hours × units/hour × dimensionless factors = good units. The equation assumes that losses are multiplicative and sufficiently separable. If downtime, speed and quality are statistically dependent, direct multiplication may misstate capacity.

Worked Example 1: Effective machine capacity

A machine is scheduled for 2 shifts of 8 hours on 20 days, giving H_s = 320 hours per month. Planned and unplanned availability is 0.90, performance is 0.95 and first-pass yield is 0.98. The standard rate is 10 units per hour.

Thus, practical good-output capacity is approximately 2,680 units, not the design value of 3,200 units. The combined loss is approximately 16.3% of theoretical output. A plan based on 3,200 units would therefore contain a substantial hidden shortfall.

2.3 Labor capacity and absenteeism

For a labor group with N_e employees, scheduled hours per employee h, attendance rate a, and skill qualification factor q, available qualified hours are:

The qualification factor is useful when not every employee can perform every task. It should not be used to conceal poor cross-training; its definition must be documented. If absenteeism differs by day or skill, an average attendance rate may be inadequate, and scenario or distribution-based analysis is preferable.

2.4 Yield loss and input requirements

If the required saleable quantity is Q_g and yield is Y, the required input quantity is:

This applies when yield is expressed as the proportion of input becoming acceptable output and when rework is not already included in the processing-time standard. For example, producing 1,000 good units at a yield of 0.95 requires approximately 1,052.6 input units. If rework consumes capacity, both additional material and additional processing time must be modeled.

3. Modeling Demand and Workload Uncertainty

3.1 Demand distributions and forecast error

A forecast is not a commitment that demand will equal one number. It is an estimate with an error distribution. If demand in period t is D_t and the forecast is F_t, forecast error may be defined as:

The mean absolute deviation is:

The root mean square error is:

MAD is less sensitive to extreme errors than RMSE. Neither metric alone proves that future errors will follow the historical pattern. Structural changes, promotions and new competitors can invalidate historical estimates.

When demand is approximately normal with mean μ_D and standard deviation σ_D, a capacity level covering probability α may be estimated as:

where z_α is the standard-normal quantile satisfying P(D ≤ C_α) = α. This approximation is inappropriate for highly skewed, intermittent or bounded demand unless a suitable transformation or empirical distribution is used.

3.2 Workload rather than unit demand

Capacity risk depends on workload, not merely units. With several products, workload is:

If demand quantities are random, workload is random as well. Under independence between product demands, its variance is:

If demands are correlated, covariance terms must be included:

Ignoring positive correlation understates capacity risk. For example, several products may peak simultaneously during the same market event.

4. Downtime, Absenteeism, Yield and Supply Disruption

4.1 Downtime and reliability

For a repairable machine, availability can be approximated by:

where MTBF is mean time between failures and MTTR is mean time to repair. This steady-state relationship assumes stable failure and repair behavior, a consistent definition of failure, and no major effects from preventive maintenance or spare-part shortages. If planned downtime is included, the numerator and denominator must be defined consistently.

Reliability analysis should distinguish random failure from common-cause failure. Ten machines may appear to provide redundancy, but a shared electrical system or common supplier can cause simultaneous loss of capacity.

4.2 Absenteeism and skill bottlenecks

Average labor capacity can be misleading when absenteeism affects a scarce skill. A 5% overall absence rate may be manageable, while a 5% absence rate among certified welders may stop an entire production cell. Capacity plans should therefore model skill pools, cross-training and substitution rules.

A practical approach is to create scenarios such as:

• normal attendance: 95%;
• adverse attendance: 88%;
• critical-skill absence: one or more key qualifications unavailable;
• recovery condition: overtime or temporary labor available with stated lead time.

The analysis should report both total lost hours and the number of orders affected by skill-specific shortages.

4.3 Yield loss and quality capacity

Quality losses reduce capacity in two ways. Scrap requires replacement production, while rework consumes time that could have been used for new orders. If s is scrap proportion and r_w is the additional rework time per defective unit, the expected workload may be represented as:

where Q is planned saleable output and t is standard processing time per unit. This expression assumes a stable defect rate and one defined rework path. If rework has a different routing or queue, it should be modeled as a separate workload stream.

4.4 Supply disruption and effective capacity

A machine’s nominal capacity has little value when required materials are unavailable. Supply disruption should be represented through at least three dimensions: probability, duration and consequence. A simple expected-loss measure is:

where p_k is the probability of disruption scenario k and L_k is the capacity loss under that scenario. Expected loss alone is insufficient for low-probability catastrophic events; tail consequences and recovery time must also be considered.

Useful mitigations include dual sourcing, strategic inventory, substitute materials, supplier capacity reservations, postponement, design standardization and contractual recovery provisions. Each mitigation creates cost and sometimes new risk, such as quality variation from an unqualified second source.

5. Scenario Planning and Sensitivity Analysis

Scenario planning evaluates coherent combinations of assumptions rather than changing one variable in isolation. A capacity scenario may specify demand, mix, uptime, attendance, yield, material availability, overtime and recovery policy.

Sensitivity analysis changes one assumption while holding others constant, or uses a structured design to identify interactions. If required capacity is R and effective capacity is C, utilization is:

A sensitivity table can show how utilization changes as demand or availability changes. The most influential variables deserve improved measurement, preventive action or explicit management reserves. Sensitivity is not the same as probability: a variable can have a large effect but low likelihood, or a small effect but high likelihood.

6. Safety Capacity and Service-Level Planning

Safety capacity is capacity deliberately retained to absorb uncertainty. It may take the form of spare machines, flexible labor, overtime rights, cross-training, reserved supplier capacity, time buffers or modular equipment.

If required workload is approximately normal with mean μ_W and standard deviation σ_W, and effective capacity is treated as a fixed level, a capacity target with service probability α is:

The safety capacity above the mean is:

This calculation assumes a stable distribution and a clearly defined service event, such as “total workload does not exceed capacity in the planning period.” It does not automatically guarantee on-time delivery because timing, sequencing and bottleneck location also matter.

A risk-based decision compares the cost of additional capacity with the cost of shortage. If the marginal cost of one unit of safety capacity is c_C and the expected shortage cost avoided is c_S, additional capacity is economically justified when the expected avoided shortage cost exceeds its cost. In practice, shortage cost may include lost contribution, expediting, penalties, reputational damage and safety consequences, not merely overtime.

7. Queuing Concepts and the Utilization Paradox

A resource can have average capacity greater than average demand and still create unacceptable waiting. Queuing theory explains why. As utilization approaches 100%, waiting time can increase sharply because arriving jobs encounter work already in process and because variability accumulates.

For a simple single-server M/M/1 queue, arrivals are Poisson with rate λ, service times are exponential with rate μ, and λ < μ. Utilization is:

The expected time in the system is:

and expected number in the system is:

These results rely on the stated queue assumptions and should not be applied literally to every manufacturing system. They nevertheless demonstrate the utilization effect. In a more general queue, the Kingman approximation for a single-server, first-come-first-served process is:

where W_q is approximate mean waiting time, c_a is the coefficient of variation of interarrival times, c_s is the coefficient of variation of service times, and E[S] is mean service time. The coefficient of variation is standard deviation divided by mean. The approximation assumes a stable queue and is most useful for managerial insight rather than exact scheduling.

The implication is important: reducing variability can be as valuable as adding nominal capacity. Smaller batches, reliable material supply, preventive maintenance, standardized work and better sequencing may reduce waiting without purchasing another machine.

8. Simulation for Capacity Risk

Simulation is appropriate when interactions, distributions and operating rules are too complex for closed-form equations. A discrete-event simulation may represent arrivals, routing, setup, machine failures, repairs, labor calendars, quality checks, queues and dispatching priorities.

A basic Monte Carlo procedure is:

1. define the planning horizon and performance measures;
2. assign probability distributions or empirical samples to uncertain inputs;
3. generate random observations for demand, processing time, downtime and yield;
4. execute the production logic for one replication;
5. record output, utilization, queue time, shortages and service performance;
6. repeat for many replications;
7. summarize the distribution of outcomes and compare alternatives.

Simulation outputs should include confidence intervals, not just one average. Initial transient conditions should be handled through a warm-up period where appropriate. Random-number seeds, input data, validation tests and model version control are governance requirements.

Simulation is not a substitute for knowledge. A highly detailed model with poor assumptions can produce precise but misleading results. Validation should include face validation by operational experts, comparison with historical performance and sensitivity tests for key assumptions.

9. Integrated Capacity-Planning Framework

A robust planning process can be organized into eight stages:

Stage 1: Define the decision and service promise

Specify the horizon, products, customer service requirement, bottleneck resources and consequences of failure. “Enough capacity” is meaningless unless the required service level is defined.

Stage 2: Build the workload model

Translate demand and product mix into processing, setup, inspection, material-handling and rework requirements. Use standard times that are current and traceable.

Stage 3: Estimate effective capacity

Adjust scheduled hours for availability, performance, quality, attendance, skills, planned maintenance and calendar restrictions. Identify bottlenecks by period, not only in aggregate.

Stage 4: Characterize uncertainty

Use historical distributions, forecast-error analysis, supplier evidence, engineering judgment and scenario workshops. Document assumptions and distinguish measured data from subjective estimates.

Stage 5: Test scenarios and sensitivities

Evaluate base, upside, downside and combined-stress conditions. Identify threshold points at which overtime, subcontracting, rescheduling or capital investment becomes necessary.

Stage 6: Select capacity protection

Options include flexible shifts, overtime, cross-training, temporary labor, preventive maintenance, spare parts, subcontracting, dual sourcing, finished-goods inventory and modular capital. The choice should consider reversibility and lead time.

Stage 7: Simulate or pilot critical decisions

Use discrete-event simulation for complex networks and pilot tests for new work methods. Compare not only average output but also tail risk, queue time and recovery performance.

Stage 8: Monitor and update

Track leading indicators such as forecast bias, processing-time dispersion, absenteeism by skill, failure intensity, supplier confirmation reliability, queue length and first-pass yield. Capacity planning is a control process, not a once-a-year calculation.

10. Illustrative Case: Northstar Components

The following case is fictional and is intended for analytical instruction.

Background

Northstar Components produces precision pump housings on a machining cell. Monthly customer demand is forecast at 4,800 good units. Each unit requires 0.50 machine-hours. The cell has three machines, each scheduled for 176 hours per month. Historical availability is 0.88, performance is 0.94 and first-pass yield is 0.97. Demand standard deviation is 420 units. The operations director is considering whether to accept a new contract that would add 12% to demand.

Problem

The average calculation suggests that three machines may be adequate, but the plant has experienced late deliveries during demand peaks and material shortages. Management needs to evaluate average feasibility, service-level capacity and practical risk.

Analysis of current demand

Scheduled machine-hours are:

Effective good-output-equivalent hours are:

Good-output capacity is therefore:

The preceding expression exposes a unit mistake: 423 hours divided by 0.50 hours per unit equals 846 units, which is inconsistent with the likely output from three machines. The correct interpretation is that 0.50 hours per unit applies to one machine, so 423 effective machine-hours produce:

This reveals that the stated standard is incompatible with the demand and asset count if interpreted literally: three machines at 176 hours and 0.50 hours per unit have design capacity 3,168 units, not 4,800. The data must therefore be clarified before a decision. If the intended processing time is 0.10 machine-hours per unit, effective capacity becomes approximately 4,230 units, still below the 4,800-unit forecast. This diagnostic is itself a major CRP finding: inconsistent master data can invalidate sophisticated analysis.

Northstar’s engineering review confirms that the correct standard is 0.10 machine-hours per unit, and that the forecast includes 12% subcontracted output not requiring the cell. Internal demand is therefore 4,224 units. Effective capacity of approximately 4,230 units leaves almost no average margin.

Risk analysis

Assuming internal demand is approximately normal, a 95% period service target requires:

The current effective capacity of 4,230 units is far below this risk-adjusted requirement. The contract would increase internal demand by 12%:

That level is above average capacity and close to, but still below, the estimated 95% requirement before considering supply disruption.

Approach and solution

Northstar evaluates four alternatives:

1. add a fourth machine, with a six-month acquisition lead time;
2. authorize weekend overtime equivalent to 60 machine-hours monthly;
3. qualify a second supplier for a critical casting;
4. reduce performance losses through tool-condition monitoring and standardized setup.

The recommended immediate plan combines supplier qualification, 40 machine-hours of flexible overtime and setup reduction. The fourth machine is approved conditionally, triggered if rolling three-month demand exceeds 4,600 units or if the 90th-percentile queue-time target is missed for two consecutive months. This staged decision avoids treating a forecast as certainty while preserving a path to expansion.

Outcome and learning points

The case demonstrates that capacity analysis begins with data validation. It also shows that average capacity can conceal service failure, that subcontracted demand must be separated from internal workload, and that a portfolio of flexible and structural responses is often superior to a single large investment. Finally, risk-adjusted capacity should include supply and quality conditions, not just machine hours.

11. Technology, AI and Contemporary Practice

Modern manufacturing execution systems, industrial internet platforms and advanced planning systems can provide near-real-time data on cycle time, downtime, queue length and quality. Digital twins can test alternative routings, maintenance policies and staffing plans before implementation.

Machine-learning models may improve demand forecasts, failure prediction and processing-time estimation. However, predictive accuracy does not guarantee decision quality. Models can fail when operating conditions change, when data omit unplanned work, or when historical schedules encode unfair labor assumptions. AI-supported capacity planning should therefore include human review, data lineage, uncertainty intervals, drift monitoring and override authority.

A useful governance principle is to separate prediction from policy. An algorithm may estimate a high probability of absenteeism, but management must still decide whether the response is cross-training, temporary labor, overtime or schedule redesign. Decisions affecting employee workload, safety or employment conditions require transparency and compliance with labor law and organizational policy.

12. Limitations, Ethics and Governance

Capacity models can encourage managers to maximize utilization at the expense of safety, maintenance and employee well-being. Persistent overtime may appear to solve a capacity gap while increasing fatigue, defects and injury risk. Ethical capacity planning treats legal working-time limits, safe staffing and humane scheduling as constraints, not optional costs.

Other governance concerns include:

• misrepresenting effective capacity by excluding rework or unpaid labor;
• using optimistic supplier probabilities without evidence;
• hiding uncertainty through a single deterministic number;
• allocating scarce capacity in ways that unfairly discriminate among customers or workers;
• allowing AI systems to make unreviewed employment or safety decisions;
• failing to document assumptions used in investment approvals.

A defensible capacity plan should preserve an audit trail: data sources, definitions, formulas, scenario assumptions, model validation results, decision thresholds and approval responsibilities.

Best-Practice Checklist

A professional capacity plan should:

1. calculate workload in time units as well as product units;
2. distinguish design, effective, actual and risk-adjusted capacity;
3. model bottleneck skills and materials explicitly;
4. include downtime, attendance, yield, setup and rework;
5. use distributions or scenarios rather than relying only on averages;
6. test correlated and common-cause disruptions;
7. examine queue time and service performance, not only utilization;
8. compare flexible, structural and supply-side responses;
9. validate data and simulation logic with operational experts;
10. define trigger points for escalation and capacity expansion;
11. report uncertainty and tail outcomes to decision makers;
12. protect safety, legal compliance, fairness and employee well-being.

Chapter Summary

Capacity planning under variability and uncertainty extends conventional CRP by recognizing that demand, processing time, availability, attendance, yield and supply are not fixed. Design capacity must be converted into effective capacity through explicit loss factors, while mixed-product requirements should be expressed as workload hours. Demand and workload distributions support service-level planning, but correlation and distributional assumptions must be examined carefully.

Scenario planning reveals how coherent combinations of adverse conditions affect feasibility. Sensitivity analysis identifies influential variables, while safety capacity provides protection against specified levels of uncertainty. Queuing theory explains why high utilization can produce disproportionate waiting, making variability reduction an important alternative to adding assets. Simulation is valuable when routing, failures, priorities and recovery rules interact, provided that the model is validated and its uncertainty is reported.

The professional capacity planner integrates quantitative analysis with staged decisions, operational judgment, data governance, supply resilience and ethical responsibility. The objective is not maximum utilization; it is reliable, safe and economically sustainable performance under plausible future conditions.

Key Terms

Variability: Observable differences in demand, time, quality or operating conditions.

Uncertainty: Incomplete knowledge about future events or their outcomes.

Risk: Uncertainty whose possible outcomes and probabilities are sufficiently characterized for decision analysis.

Design capacity: Theoretical maximum output under ideal conditions.

Effective capacity: Practical capacity after planned and normal operating losses.

Actual output: Output achieved during a specified period.

Workload: Required processing and related time expressed in labor-hours or machine-hours.

Availability: Proportion of scheduled time during which a resource is operational.

Performance: Actual operating speed relative to the applicable standard speed.

First-pass yield: Proportion of input producing acceptable output without rework.

Safety capacity: Deliberately retained capacity used to absorb uncertainty.

Service level: Probability or proportion of demand fulfilled within a specified requirement.

Sensitivity analysis: Examination of how an output changes when an input assumption changes.

Scenario analysis: Evaluation of coherent combinations of assumptions about future conditions.

Utilization: Required workload divided by available capacity.

Queuing: Waiting and service behavior arising when work arrives at constrained resources.

Coefficient of variation: Standard deviation divided by the mean.

Monte Carlo simulation: Repeated random sampling used to estimate a distribution of outcomes.

Common-cause failure: A disruption that affects multiple resources because they share a cause.

Resilience: The ability of a production system to absorb disruption and recover acceptable performance.

Review Questions

Short Answer Questions

1. Distinguish design capacity, effective capacity and actual output.
2. Why should mixed-product capacity requirements be expressed in workload hours?
3. State two limitations of using average demand in capacity planning.
4. What is the utilization effect described by queuing theory?
5. Identify three forms of safety capacity.

Long Answer Questions

1. Explain how availability, performance, yield, absenteeism and rework jointly influence effective capacity.
2. Compare scenario analysis, sensitivity analysis and simulation as capacity-planning methods.
3. Discuss the assumptions and managerial usefulness of the M/M/1 queue model.
4. Develop a risk-based framework for responding to supply disruption.
5. Critically evaluate the use of AI and digital twins in capacity requirement planning.

Analytical / Application Questions

1. A cell has 2,400 scheduled hours, availability of 0.86, performance of 0.93, yield of 0.96 and a standard rate of 2 units per hour. Calculate effective good-output capacity and explain the assumptions.
2. Forecast workload is normally distributed with mean 10,000 hours and standard deviation 700 hours. Estimate capacity for a 90% service probability using z = 1.282, and calculate safety capacity above the mean.
3. A plant is below average capacity but experiences long queues. Propose a diagnostic using variability, bottleneck, quality and dispatching data, and recommend two interventions other than purchasing a new machine.

Multiple Choice Questions with Answers

Question 1

Which expression best represents effective good-output capacity when losses are modeled as separable factors?

A. C_e = H_s × A × P × Y × r

B. C_e = H_s + A + P + Y + r

C. C_e = H_s × (A + P + Y) ÷ r

D. C_e = H_s ÷ A × P × Y × r

Correct Answer: A. C_e = H_s × A × P × Y × r

Question 2

A process has 1,000 scheduled hours, availability 0.90, performance 0.80, yield 0.95 and a rate of 5 units per hour. What is its effective good-output capacity?

A. 3,420 units

B. 3,600 units

C. 3,420?

D. 3,420 units after rounding

Correct Answer: D. 3,420 units after rounding

Question 3

If demand is normally distributed with mean μ and standard deviation σ, which expression estimates capacity at service probability α?

A. μ − z_α σ

B. μ + z_α σ

C. z_α ÷ (μ + σ)

D. μ × z_α ÷ σ

Correct Answer: B. μ + z_α σ

Question 4

What is the principal implication of utilization approaching 100% in a variable production system?

A. Waiting time generally rises sharply

B. Quality losses necessarily become zero

C. Demand variation disappears

D. Required capacity becomes independent of processing time

Correct Answer: A. Waiting time generally rises sharply

Question 5

In the availability relationship A = MTBF ÷ (MTBF + MTTR), MTTR denotes:

A. Mean time between requirements

B. Mean total resource rate

C. Mean time to repair

D. Maximum time to recover demand

Correct Answer: C. Mean time to repair

Question 6

Which issue is most likely to be missed by using only an overall absenteeism percentage?

A. The effect of absence on scarce skills

B. The existence of scheduled hours

C. The definition of product demand

D. The standard unit of currency

Correct Answer: A. The effect of absence on scarce skills

Question 7

What is the main purpose of sensitivity analysis in capacity planning?

A. To prove that one forecast is correct

B. To identify how changes in assumptions affect decisions

C. To eliminate all uncertainty from the model

D. To replace operational data with intuition

Correct Answer: B. To identify how changes in assumptions affect decisions

Question 8

A simulation replication in a capacity model should normally record which type of result?

A. Only the most favorable output

B. Output, queues, utilization and shortages

C. Only the random-number seed

D. Only the purchased-equipment cost

Correct Answer: B. Output, queues, utilization and shortages

Question 9

If saleable output required is 1,000 units and yield is 0.80, how many input units are required, ignoring other constraints?

A. 800

B. 1,000

C. 1,250

D. 1,800

Correct Answer: C. 1,250

Question 10

Which is an example of safety capacity?

A. Removing preventive maintenance

B. Eliminating all cross-training

C. Retaining flexible overtime or qualified backup labor

D. Planning exactly at average demand with no reserve

Correct Answer: C. Retaining flexible overtime or qualified backup labor