Demand Forecasting and Inventory Analytics

Introduction and Context

Inventory decisions are made before the future is known. A firm must decide how much to purchase, produce, position, and protect while customer demand, replenishment time, prices, promotions, and supply conditions remain uncertain. Demand forecasting provides a structured estimate of future requirements; inventory analytics converts that estimate, together with uncertainty and managerial constraints, into operational decisions.

Forecasting is therefore not simply the production of a single number. It is a decision-support process involving data definition, model selection, judgment, uncertainty measurement, parameter translation, monitoring, and corrective action. In material requirements planning (MRP), the forecast may drive the master production schedule and independent demand for finished goods. In distribution and inventory control, it influences reorder points, safety stock, order quantities, capacity, transportation, and service commitments.

A technically sophisticated forecast can still create poor decisions if the demand measure is inappropriate, forecast errors are biased, lead-time variability is ignored, or parameters are not updated. Conversely, a relatively simple method may perform well when it is transparent, stable, and properly governed.

Learning Objectives

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

1. Distinguish time-series, causal, intermittent-demand, and judgmental forecasting.
2. Select forecasting methods in relation to demand characteristics and decision context.
3. Calculate and interpret forecast errors, bias, and accuracy measures.
4. Translate demand and lead-time uncertainty into inventory-control parameters.
5. Apply scenario analysis, simulation, optimization, and exception analytics.
6. Evaluate forecasting systems critically, including their ethical, technological, and governance implications.

1. Forecasting as an Operations Decision Process

A demand forecast is an estimate of future demand for a defined item, location, time bucket, and decision horizon. Four specifications are essential:

• Item definition: stock-keeping unit (SKU), product family, component, or service.
• Location: plant, warehouse, market, channel, or total network.
• Time bucket: day, week, month, quarter, or another relevant interval.
• Horizon: the period into the future for which the forecast is required.

Demand should be distinguished from sales. Sales may be lower than true demand when stockouts, allocation rules, lost orders, or constrained capacity prevent fulfillment. Using sales as demand without adjustment can systematically underestimate requirements. Conversely, bookings, shipments, returns, substitutions, and cancellations may need reconciliation before analysis.

Forecasts generally contain a predictable component and an unpredictable component. A useful conceptual representation is:

where D_t is observed demand in period t, F_t is the forecastable component, and ε_t is the error or innovation. This representation does not imply that errors are always independent, normally distributed, or constant over time; those assumptions must be tested rather than presumed.

Forecasts should be evaluated against the decision they support. A forecast for capacity planning may use monthly family-level demand, whereas a reorder-point decision may require SKU-location demand during replenishment lead time. Aggregation can reduce random variation but may conceal local shortages and does not automatically improve every decision.

2. Demand Patterns and Method Selection

The principal demand patterns are level, trend, seasonality, cycles, intermittent occurrence, and irregular shocks.

• Level: demand fluctuates around a relatively stable mean.
• Trend: the mean changes persistently upward or downward.
• Seasonality: a recurring pattern is associated with calendar position or another known cycle.
• Cycle: broad movements occur over longer, less regular economic or market periods.
• Intermittency: many periods have zero demand, separated by irregular nonzero demands.
• Structural change: the data-generating process changes because of a new product, policy, competitor, technology, or market.

No method is universally best. Method choice should consider forecast horizon, data frequency, demand granularity, intermittency, promotions, causal information, computational requirements, explainability, and the cost of forecast errors.

A baseline is indispensable. For seasonal demand, a seasonal naïve forecast often provides a difficult-to-beat reference: the forecast for a period equals demand observed in the corresponding prior season. Complex models should demonstrate incremental value relative to such baselines.

3. Time-Series Forecasting

Time-series methods use the historical behavior of the variable itself. They are particularly useful when causal data are unavailable, the system is stable, and short- to medium-term patterns persist.

3.1 Naïve and Moving-Average Methods

The naïve forecast sets the next forecast equal to the latest observation:

It is appropriate as a benchmark and can be effective for random-walk-like demand. A simple moving average of the most recent n observations is:

Here n is the window length. All observations receive equal weight. A larger window smooths noise but responds slowly to genuine change; a smaller window responds faster but is more variable.

A weighted moving average is:

The weights must be dimensionless and sum to one, so the forecast has the same units as demand.

3.2 Simple Exponential Smoothing

Simple exponential smoothing (SES) is suitable for approximately level demand without systematic trend or seasonality:

The smoothing constant α controls responsiveness. A high value gives greater weight to recent observations; a low value produces more smoothing. The initial forecast F₁ must be specified, for example as the first observation or an initial average. SES is not designed to represent persistent trend or seasonality without extensions.

3.3 Trend and Seasonality

Holt’s linear method maintains a level and trend. One common formulation is:

Here ℓ_t is the estimated level, b_t is the estimated trend, h is the forecast horizon, and α and β lie between zero and one. The additive trend formulation is most appropriate when trend changes by an approximately constant number of units per period. Damped-trend methods reduce long-horizon extrapolation by gradually attenuating the trend.

Seasonal methods add a recurring seasonal component. Additive seasonality is appropriate when seasonal fluctuations have roughly constant absolute size; multiplicative seasonality is more suitable when seasonal amplitude grows with the level. A multiplicative model must be used carefully when observations are zero or near zero.

3.4 Forecast Combination

Combining forecasts from different models can improve robustness when models contain distinct information or errors. A combined forecast is:

Weights may be equal, optimized using validation data, or assigned by a governed process. Combination does not guarantee improvement, particularly when component models share the same misspecification.

4. Causal Forecasting

Causal forecasting relates demand to explanatory variables such as price, advertising, promotions, weather, economic indicators, installed base, and distribution coverage. A basic linear model is:

D_t is demand, X_j,t are explanatory variables, β_j are coefficients, and ε_t is the unexplained component. The units of each coefficient depend on the units of demand and its associated predictor. The model assumes, at minimum, that the specified relationship is useful for prediction over the intended range and that future predictor values are available or can themselves be forecast.

A coefficient should not automatically be interpreted as a causal effect. Price may be reduced when demand is already weak, creating reverse causality. Promotions may be targeted to products expected to sell, creating selection bias. Omitted variables, multicollinearity, autocorrelated errors, changing relationships, and data leakage can invalidate apparently strong models.

For inventory planning, causal models should include operational variables carefully. A promotion calendar may be known in advance, whereas competitor price may not be. Forecasting demand during a promotion requires deciding whether the model estimates baseline demand, incremental promotional demand, or total demand. Cannibalization among related SKUs and post-promotion dips should be represented where material.

Modern machine-learning methods can model nonlinearities and interactions, but predictive accuracy in a test sample is not proof of stable causal structure. Time-ordered validation is required; random shuffling can leak future information into training data. Explainability, monitoring, and controlled overrides are especially important when forecasts affect customer allocation or employee performance measures.

5. Intermittent-Demand Forecasting

Intermittent demand occurs when demand is zero in many periods. It is common for service parts, slow-moving components, and specialized maintenance items. Two characteristics should be separated: the interval between nonzero demands and the size of a nonzero demand.

Croston-type methods estimate these components separately. If z_t is the estimated nonzero demand size and p_t is the estimated interval, a basic forecast for average demand per period is:

The method updates the size and interval only when a nonzero demand occurs. Variants, including SBA and TSB, address statistical bias or changing probability of demand. TSB explicitly updates the probability that demand occurs and can respond better when an item becomes obsolete.

Intermittent-demand forecasting has important implementation issues. A forecast of average demand per period does not by itself describe the probability of a large demand during lead time. Inventory policy may need a discrete-demand model, empirical lead-time demand distribution, or simulation. Aggregating intermittent demand across customers or periods may improve forecast stability, but it can obscure the service requirement for a particular location.

6. Judgmental Forecasting and Human–Model Collaboration

Judgment is necessary when historical data are sparse, markets are changing, products are new, or strategic information is not encoded in the data. Useful methods include structured expert panels, Delphi processes, sales-force estimates, market research, analogy, and scenario workshops.

Judgment is vulnerable to anchoring, optimism, availability bias, political incentives, and double counting. A sales team may inflate forecasts to secure inventory, while a purchasing group may suppress them to reduce working capital. Judgmental adjustments should therefore be recorded with a reason, owner, date, magnitude, and later outcome.

A disciplined process separates the statistical baseline from the judgmental adjustment. For example, a promotion manager may add an evidence-based uplift, while a supply planner evaluates capacity and cannibalization. Overrides should be measured for incremental accuracy, not rewarded merely because they agree with a senior executive. Human expertise is most valuable when it supplies information unavailable to the model; it is least valuable when it simply reacts to random noise.

7. Forecast Error, Bias, and Accuracy

Let the forecast error be defined as:

Positive error means actual demand exceeded the forecast; negative error means the forecast was higher than actual demand. Organizations must state their sign convention because some systems use F_t − D_t.

Common measures include:

ME measures average bias in demand units. MAE measures typical absolute error in demand units. RMSE penalizes large errors more strongly. MAPE is undefined when actual demand is zero and unstable when actual demand is very small; it should not be the default metric for intermittent demand.

Weighted absolute percentage error can be written as:

WAPE is defined when total actual demand is positive, but it gives greater influence to high-volume observations. A scale-free alternative is the mean absolute scaled error (MASE), which compares model error with the error of a naïve benchmark. Metrics should be selected according to economic consequences, not convenience.

Bias can be monitored using cumulative error:

A tracking signal is sometimes defined as:

where MAD_t is a specified mean absolute deviation measure. Control limits such as ±4 or ±5 are conventions, not universal laws. They should be calibrated to the process, data volume, and false-alarm cost.

Worked Example: Comparing Forecasts

Suppose actual monthly demand is 100, 120, and 80 units. Forecast A is 90, 130, and 90; Forecast B is 105, 115, and 85.

For Forecast A, errors are 10, −10, and −10. Thus, ME = (10−10−10)/3 = −3.33 units, MAE = 10 units, and RMSE = √(300/3) = 10 units.

For Forecast B, errors are −5, 5, and −5. Thus, ME = −1.67 units, MAE = 5 units, and RMSE = √(75/3) = 5 units. Forecast B is more accurate and has less negative bias. This conclusion is based on three observations only; validation over a longer representative period is necessary before changing policy.

8. From Forecast Error to Inventory Parameters

Forecasting and inventory control are coupled. A forecast provides expected demand; its error distribution informs protection against uncertainty. For a continuous-review system, a basic reorder point is:

where μ_L is expected demand during replenishment lead time and SS is safety stock. If demand per period has mean μ_D, fixed lead time is L periods, and periods are independent, then:

If the standard deviation of demand per period is σ_D, then lead-time demand standard deviation is:

This relationship assumes independent, identically distributed period demand and fixed lead time. It is inappropriate when demand is autocorrelated, lead time varies materially, or the demand distribution is strongly skewed.

Under a normal approximation and a cycle-service target represented by standard-normal quantile z, safety stock is:

For random demand and random lead time, with independence between demand and lead time, a commonly used approximation is:

where μ_Lt and σ_Lt are the mean and standard deviation of lead time, and σ_L on the left denotes the standard deviation of lead-time demand. To avoid ambiguity in implementation, many organizations use separate notation such as σ_DL. The approximation assumes stable moments and independence; correlated demand and lead time require a covariance term or empirical simulation.

The service target must be defined. Cycle service is the probability of no stockout during a replenishment cycle. Fill rate is the proportion of demand satisfied immediately from stock. They are not equivalent. A policy designed for a fill-rate target should use expected shortage or a simulated lead-time demand distribution rather than automatically substituting a normal quantile.

Forecast error should be incorporated into demand uncertainty. If the operational forecast is unbiased and forecast errors have standard deviation σ_e, then using only historical demand variation may double count or omit uncertainty. The correct treatment depends on whether σ_e represents uncertainty in future demand, uncertainty in the estimated mean, or residual error after accounting for known variation. Parameter definitions must be documented.

Worked Example: Reorder Point

Mean weekly demand is 200 units, weekly standard deviation is 40 units, and replenishment lead time is four fixed weeks. For a 95% cycle-service approximation, use z ≈ 1.645. Expected lead-time demand is μ_L = 4×200 = 800 units. Lead-time standard deviation is σ_L = 40√4 = 80 units. Safety stock is SS = 1.645×80 = 131.6, rounded to 132 units. Therefore:

ROP = 800 + 132 = 932 units.

This result assumes independent weekly demand, a stable distribution, fixed lead time, and a normal approximation. If demand is highly skewed or promotional, empirical quantiles or simulation may be more defensible.

9. Scenario Analysis, Simulation, and Optimization

Scenario Analysis

Scenario analysis evaluates coherent combinations of assumptions rather than changing one variable in isolation. A planning set may include base, high-demand, low-demand, supply-disruption, and promotion-failure scenarios. Each scenario should specify demand, lead time, capacity, costs, and decision rules. Scenarios are not probabilities unless probabilities are explicitly assigned.

Monte Carlo Simulation

Simulation is useful when demand, lead time, order constraints, substitution, service rules, and network interactions create distributions that analytical formulas cannot represent. A basic process is:

1. Specify empirical or parametric distributions for demand and lead time.
2. Generate random realizations over the planning horizon.
3. Apply the actual replenishment policy.
4. Record service, inventory, shortage, and cost outcomes.
5. Repeat sufficiently many times and examine confidence intervals.

Simulation should use out-of-sample data and preserve temporal dependence where it matters. Random-number generation does not compensate for poor distribution assumptions. Results should be reported as ranges or distributions, not as a falsely precise single value.

Optimization

Optimization selects decisions subject to objectives and constraints. A simplified multi-item allocation model may minimize shortage and holding cost:

subject to inventory-balance, capacity, budget, minimum-order, and service constraints. Here I_i is inventory, S_i is shortage, h_i is holding-cost weight, and p_i is shortage-cost weight for item i. The objective is meaningful only if costs and constraints reflect managerial priorities. Optimizers can produce mathematically feasible but operationally unacceptable results when costs are misstated or forecasts are treated as certain.

10. Exception Analytics and Forecast Governance

Analytics should focus managerial attention on material exceptions rather than require equal review of every SKU. Useful exception indicators include:

• persistent positive or negative bias;
• forecast error exceeding a value or percentage threshold;
• abrupt demand-level or variance changes;
• unusual promotional uplift or post-promotion decline;
• stockout-censored demand;
• forecast overrides with repeated poor outcomes;
• supplier lead-time deterioration;
• high inventory value combined with low service;
• model drift or missing data.

A mature governance process defines ownership, approval thresholds, audit trails, forecast hierarchies, version control, and retraining frequency. Forecasts should be reconciled across levels—for example, SKU forecasts should be coherent with family and total-company forecasts when decisions require that consistency. Reconciliation can be judgmental, proportional, or optimization-based, but the chosen method should be transparent.

Data quality controls should detect duplicated transactions, unit changes, calendar shifts, returns, product substitutions, and stockout periods. A forecast system without data lineage cannot provide reliable accountability.

11. Illustrative Case: Service-Parts Network

Background

A fictional industrial-equipment manufacturer stocks 4,000 service parts at regional warehouses. Many parts have intermittent demand. The company used a 12-month moving average and a fixed two-week safety-stock rule. Planners frequently overrode forecasts after urgent customer requests.

Problem

The firm experienced both expensive excess stock and emergency shipments. A preliminary analysis showed that sales data understated demand for parts that were unavailable. In addition, the moving average produced zero or very low forecasts for items with long zero-demand intervals, while it reacted slowly to a newly installed equipment population.

Analysis

The company segmented parts by demand volume, intermittency, criticality, and lead time. For high-volume regular items, seasonal naïve and exponential-smoothing baselines were compared with causal models using installed-base and maintenance-cycle information. For intermittent items, Croston-type and TSB methods were evaluated. Demand was reconstructed from back orders and documented substitutions where evidence supported the adjustment.

Forecast accuracy was measured using MAE and scaled measures rather than MAPE, because many periods had zero demand. Bias was reviewed by item class. The inventory team estimated empirical lead-time demand distributions and simulated the existing and proposed policies. Critical parts were evaluated using a higher service objective, but the target was linked to equipment downtime cost rather than applied uniformly.

Approach and Solution

The firm adopted a segmented policy. Stable, high-volume parts used exponential smoothing with automatic parameter selection constrained by time-ordered validation. Intermittent parts used a TSB-type method and empirical simulation. New parts used an analogy-based forecast with explicit low, base, and high scenarios. Planner overrides required a reason code and were reviewed after the demand period.

The system also introduced exception dashboards showing cumulative bias, stockout-censored periods, lead-time changes, and inventory value at risk. Safety stock was recalculated from the selected service objective and the relevant lead-time demand distribution, rather than from a universal percentage of forecast.

Outcome

In this illustrative case, the company reduced emergency shipments and excess inventory over two planning cycles while maintaining service for critical parts. The improvement did not arise from one superior algorithm. It resulted from better demand definitions, segmentation, error measurement, policy linkage, and governance. Results would need continued monitoring because equipment populations, repair practices, and supplier conditions can change.

Learning Points

1. Intermittent demand requires methods and metrics suited to zero observations.
2. Stockout-adjusted demand is often more useful than unadjusted sales.
3. Forecast performance must be judged together with inventory outcomes.
4. Criticality should influence service policy, but arbitrary safety-stock inflation is not a substitute for analysis.
5. Overrides create value only when they add information and are evaluated objectively.

12. Limitations, Ethics, and Contemporary Developments

Forecasts can reinforce historical inequities when past allocation reflected biased access, discriminatory service policies, or incomplete data. Automated prioritization may disadvantage smaller customers or regions if the objective function values revenue without considering contractual or social obligations. Governance should therefore include fairness review where allocation affects essential goods, safety-related parts, or vulnerable populations.

Privacy matters when forecasts use customer-level behavior, location, or operational personnel data. Data minimization, access controls, retention rules, and documented lawful use are necessary. Models should be auditable, particularly when they influence employment evaluation, customer service, or regulated products.

Contemporary systems increasingly use probabilistic forecasts, hierarchical reconciliation, gradient-boosting models, deep learning, and generative AI assistants. Probabilistic forecasting is valuable because inventory decisions depend on distributions, quantiles, and tail risk—not only expected demand. Deep learning may be useful with large, rich, changing data sets, but it is not automatically superior for sparse industrial demand. Generative AI can assist with exception explanations and scenario narratives, but its outputs require validation and must not silently alter approved forecasts.

Best practice is to begin with a strong baseline, use time-aware validation, maintain a champion–challenger process, monitor drift, document assumptions, and connect forecast metrics to service and financial outcomes. The appropriate question is not “Which algorithm is most advanced?” but “Which governed forecasting and inventory process makes the best decisions under the actual uncertainty and constraints?”

Chapter Summary

Demand forecasting is a structured decision process that estimates future requirements for a specified item, location, time bucket, and horizon. Time-series methods exploit historical patterns; causal methods incorporate explanatory variables; intermittent-demand methods separate occurrence and size; and judgmental methods contribute information unavailable in historical data. Method selection should reflect demand characteristics, horizon, data quality, explainability, and cost consequences.

Forecast errors must be defined consistently and evaluated using measures such as ME, MAE, RMSE, WAPE, and scaled metrics. Bias is distinct from random error and should be monitored through cumulative error and appropriate control rules. MAPE is unsuitable when actual demand is zero or very small.

Inventory parameters depend on expected demand and uncertainty during lead time. Reorder points and safety stock require explicit assumptions concerning demand distribution, lead-time variability, dependence, and service definition. Scenario analysis, simulation, and optimization extend analysis when uncertainty, constraints, and network interactions make simple formulas inadequate. Exception analytics and governance convert forecasts into accountable managerial action.

The central managerial lesson is that forecasting accuracy cannot be separated from inventory policy. Data definitions, segmentation, uncertainty representation, human judgment, technology, and governance jointly determine operational performance.

Key Terms

Demand forecast: An estimate of future demand for a defined item, location, period, and horizon.

Forecast horizon: The future interval covered by a forecast.

Time-series forecasting: Forecasting based primarily on historical values of the variable itself.

Causal forecasting: Forecasting that uses explanatory variables such as price, promotion, or weather.

Intermittent demand: Demand containing many zero periods and irregular nonzero observations.

Seasonality: A recurring pattern associated with calendar or cycle position.

Naïve forecast: A benchmark that commonly uses the latest observation as the next forecast.

Exponential smoothing: A recursive method that weights recent observations more heavily.

Forecast error: The difference between actual demand and forecast, here defined as e_t = D_t−F_t.

Bias: Persistent average forecast error in one direction.

MAE: Mean absolute error, measuring average error magnitude in demand units.

RMSE: Root mean squared error, which penalizes large errors more strongly.

WAPE: Weighted absolute percentage error based on total absolute error divided by total actual demand.

Lead-time demand: Demand occurring during replenishment lead time.

Safety stock: Inventory held to protect against demand or supply uncertainty.

Reorder point: Inventory position at which replenishment is initiated.

Cycle service: Probability of avoiding a stockout during a replenishment cycle.

Fill rate: Proportion of demand supplied immediately from available stock.

Scenario analysis: Evaluation of decisions under coherent alternative assumptions.

Exception analytics: Analysis that directs attention to material deviations, risks, or unusual patterns.

Review Questions

Short Answer Questions

1. Why can sales be an unreliable measure of demand during stockouts?
2. What is the difference between bias and random forecast error?
3. Why is MAPE problematic for intermittent demand?
4. State two assumptions behind σ_L = σ_D√L.
5. What is the purpose of a forecast baseline?

Long Answer Questions

1. Compare moving averages, exponential smoothing, causal regression, and Croston-type methods. Explain the demand conditions under which each may be appropriate.
2. Explain how forecast-error measurement should influence inventory-policy design.
3. Discuss the distinction between cycle service and fill rate and explain why confusing them can lead to poor safety-stock decisions.
4. Evaluate the role of human judgment in a statistical forecasting process, including benefits, risks, and governance controls.
5. Explain how simulation and optimization complement analytical inventory formulas.

Analytical / Application Questions

1. Weekly demand has mean 150 units and standard deviation 30 units. Lead time is fixed at nine weeks. Using z = 1.645, calculate expected lead-time demand, safety stock, and reorder point under the independent-demand normal approximation. State the assumptions.
2. A planner observes actual demand of 200, 240, 160, and 220 units and forecasts of 180, 250, 170, and 210 units. Calculate ME and MAE using e_t = D_t−F_t, and interpret the bias.
3. Design a forecasting and inventory-analytics approach for a spare-parts portfolio containing regular high-volume parts, intermittent critical parts, and newly introduced parts. Specify methods, metrics, service considerations, and exception rules.

Multiple Choice Questions with Answers

Question 1

Which method is most directly intended for demand with many zero observations and irregular nonzero demands?

A. Croston-type forecasting

B. Simple moving average with no segmentation

C. Ordinary least squares using only an intercept

D. Seasonal naïve forecasting only

Correct Answer: A. Croston-type forecasting

Question 2

If forecast error is defined as e_t = D_t−F_t, what does a positive error indicate?

A. The forecast exceeded actual demand

B. Actual demand exceeded the forecast

C. The forecast was exactly unbiased

D. Lead time was longer than expected

Correct Answer: B. Actual demand exceeded the forecast

Question 3

Which accuracy measure is generally undefined when actual demand is zero?

A. MAE

B. RMSE

C. MAPE

D. ME

Correct Answer: C. MAPE

Question 4

Mean weekly demand is 100 units, weekly standard deviation is 20 units, and fixed lead time is four weeks. Under independent demand, what is lead-time demand standard deviation?

A. 20 units

B. 40 units

C. 80 units

D. 400 units

Correct Answer: B. 40 units

Question 5

In simple exponential smoothing, increasing α generally makes the forecast:

A. Less responsive to recent observations

B. More responsive to recent observations

C. Automatically seasonal

D. Independent of the previous forecast

Correct Answer: B. More responsive to recent observations

Question 6

Which statement best describes causal forecasting?

A. It uses only the latest observed demand

B. It relates demand to explanatory variables such as price or promotion

C. It is applicable only when demand is intermittent

D. It eliminates the need to forecast explanatory variables

Correct Answer: B. It relates demand to explanatory variables such as price or promotion

Question 7

A reorder point is most fundamentally composed of:

A. Annual demand minus cycle stock

B. Expected lead-time demand plus protection against uncertainty

C. Purchase price plus transportation cost

D. Maximum inventory divided by order quantity

Correct Answer: B. Expected lead-time demand plus protection against uncertainty

Question 8

Why can random shuffling be inappropriate when validating a forecasting model?

A. It always increases computational cost

B. It may allow future information to leak into the training data

C. It prevents calculation of MAE

D. It removes all seasonal patterns

Correct Answer: B. It may allow future information to leak into the training data

Question 9

Which action is a sound governance practice for planner overrides?

A. Allow undocumented overrides by senior staff

B. Ignore overrides when evaluating forecast accuracy

C. Record the reason, owner, date, and later outcome

D. Replace every statistical forecast with judgment

Correct Answer: C. Record the reason, owner, date, and later outcome

Question 10

What is a primary advantage of simulation in inventory analysis?

A. It guarantees the assumed distributions are correct

B. It can represent complex uncertainty and replenishment rules together

C. It removes the need for input data

D. It always produces a single exact answer

Correct Answer: B. It can represent complex uncertainty and replenishment rules together