Advanced Analytics, AI and the Future of Inventory Control

Introduction

Inventory control is increasingly becoming a data-intensive, adaptive and digitally connected discipline. Traditional methods such as the economic order quantity, reorder-point systems and periodic review policies remain useful, but they often assume relatively stable demand, reliable lead times and clearly defined cost parameters. Contemporary supply chains face volatile markets, short product life cycles, geopolitical disruption, omnichannel fulfilment, sustainability pressures and cyber risks. These conditions require inventory decisions that can learn from data, evaluate uncertainty and respond rapidly without abandoning managerial accountability.

Advanced inventory control combines optimization, simulation, scenario analysis, digital twins, machine learning and artificial intelligence-assisted forecasting. The objective is not to replace inventory theory with technology. Rather, technology should improve the quality, speed and transparency of decisions grounded in sound operations principles.

Learning Objectives

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

1. Explain how optimization, simulation and scenario analysis support dynamic inventory decisions.
2. Describe the architecture and managerial value of an inventory digital twin.
3. Distinguish machine-learning forecasting from traditional statistical forecasting.
4. Evaluate AI systems according to accuracy, explainability, bias, robustness and cybersecurity.
5. Design appropriate human oversight for autonomous or semi-autonomous replenishment.
6. Assess resilient and circular inventory models in relation to cost, service, risk and sustainability.
7. Apply a structured framework to the implementation and governance of advanced inventory analytics.

1. From Static Policies to Dynamic Decision Systems

A conventional inventory policy converts forecasts and operating parameters into replenishment decisions. For example, a reorder point may be written as:

Here, r is the reorder point in units, μ_L is expected demand during lead time, σ_L is the standard deviation of lead-time demand, and z is the standard-normal service-factor corresponding to the desired cycle-service probability. The formula assumes that demand during lead time can be represented adequately by a distribution with a stable mean and standard deviation, and that the selected service factor is appropriate. If demand and lead-time variability are independent, with demand standard deviation σ_D per period, mean lead time L periods and lead-time standard deviation σ_Lt, a commonly used approximation is:

where μ_D is mean demand per period. The first term represents demand variability during an average lead time; the second represents variability caused by uncertain lead time. Units must be consistent: if demand is measured in units per week, lead time must be measured in weeks.

Dynamic inventory systems extend this logic by repeatedly updating forecasts, constraints, costs, supplier information and risk estimates. The decision is no longer simply “when should an order be placed?” It may include:

• which supplier should receive the order;
• how much capacity should be reserved;
• whether to expedite, substitute, postpone or ration stock;
• how inventory should be positioned across multiple echelons;
• whether a returned, repaired or remanufactured item can satisfy demand;
• how service, cost, cash, emissions and resilience should be balanced.

The central principle is closed-loop control: observe the system, estimate its current state, decide, act, measure the outcome and update the model.

2. Optimization for Dynamic Inventory Decisions

2.1 Deterministic and stochastic optimization

Optimization identifies a decision or policy that performs best according to a specified objective and constraints. A deterministic model treats relevant parameters as known values. A stochastic model represents uncertainty through probability distributions, scenarios or risk measures.

A simplified multi-period inventory balance is:

where I_t is ending inventory in period t, q_t is replenishment received, and d_t is demand. If shortages are allowed, I_t may be negative; alternatively, separate positive inventory and backorder variables can be used.

A basic finite-horizon cost objective is:

where T is the planning horizon, c_q is unit procurement cost, h is holding cost per unit-period, I_t^+ is positive ending inventory, p is shortage or backorder penalty, B_t is backorder quantity and c_e e_t represents an expediting cost. The model requires constraints for supplier capacity, order minimums, warehouse capacity, budget, lead time and service requirements.

Optimization is valuable when decisions interact. A local rule may minimize cost at one facility while increasing total network cost. Multi-echelon optimization can determine where safety stock should be held, taking account of upstream lead times, downstream service requirements and transportation choices.

2.2 Robust and risk-aware optimization

Expected-cost optimization can produce fragile decisions if the probability estimates are unreliable. Robust optimization instead seeks decisions that perform acceptably across a defined uncertainty set. A planning team may optimize for demand ranging from 80% to 130% of the baseline forecast, supplier lead times from 5 to 12 days, and specified capacity losses.

Risk can also be incorporated through service constraints or downside measures. One example is conditional value at risk, which focuses on the average loss in the worst specified percentage of outcomes. Risk measures must be selected carefully: minimizing cost alone may encourage excessive stockouts, whereas minimizing worst-case exposure may produce unaffordable inventory.

2.3 Rolling-horizon control

A practical advanced system often uses a rolling horizon. It optimizes several future periods, implements only the first decision, receives new information, and then re-optimizes. This approach is related to model predictive control. It is useful when forecasts, orders and supply conditions change frequently. Its quality depends on forecast updating, computational speed, accurate constraints and appropriate treatment of end-of-horizon effects.

3. Simulation and Scenario Analysis

Optimization recommends decisions under a model; simulation tests how a policy behaves under repeated or alternative conditions. Discrete-event simulation can represent arrivals, orders, queues, production stages, transport delays and stock movements. Monte Carlo simulation generates many possible demand and lead-time paths from specified distributions or historical resampling.

For a candidate policy, a simulation experiment may estimate:

• fill rate and cycle-service level;
• average and peak inventory;
• stockout frequency and duration;
• working capital;
• expedite frequency;
• capacity utilization;
• emissions and waste;
• resilience under disruption.

The simulation output is an estimate, not a guarantee. Results should include confidence intervals or sensitivity analysis where appropriate. A model that produces a single precise-looking number without uncertainty information can mislead decision-makers.

Scenario analysis differs from ordinary forecasting. A forecast estimates what is most likely conditional on available information. A scenario describes a coherent possible future, such as a port closure, a sudden promotional surge, a new tariff or the introduction of a substitute product. Scenarios are especially useful for stress testing policies and identifying trigger points.

A good scenario portfolio includes at least:

1. a central or baseline case;
2. upside demand and constrained supply;
3. downside demand and excess inventory;
4. a severe but plausible disruption;
5. a structural-change case involving technology, regulation or customer behaviour.

Scenario analysis should not be used to assign false probabilities to events that are poorly understood. Its purpose may be strategic learning rather than probability-weighted optimization.

4. Digital Twins for Inventory Control

A digital twin is a continuously updated digital representation of a physical system, its state, relationships and relevant operating rules. In inventory management, the twin may represent products, locations, suppliers, orders, shipments, capacities, demand signals, returns and quality states.

A useful architecture has five layers:

1. Physical layer: warehouses, stores, production assets, vehicles, suppliers and products.
2. Data layer: transactions, sensors, enterprise-resource-planning records, forecasts, event streams and external signals.
3. Model layer: inventory balances, lead-time models, demand models, optimization and simulation engines.
4. Decision layer: replenishment, allocation, substitution, expediting and disposition recommendations.
5. Governance layer: permissions, audit trails, validation, cybersecurity, model monitoring and human approvals.

A digital twin is more than a dashboard. A dashboard displays data; a twin maintains a structured relationship between the physical system and analytical models and can test interventions before they are implemented.

For example, a manager may ask: “What happens if supplier A loses 30% of capacity for three weeks, demand rises by 15% and returned units can be refurbished within ten days?” The twin can estimate service, cost, capacity and inventory consequences under alternative policies.

The principal risks are data latency, inconsistent master data, incorrect physical assumptions and excessive confidence in simulated results. A digital twin should therefore display data freshness, model version, assumptions and uncertainty.

5. Machine Learning and AI-Assisted Forecasting

Traditional forecasting methods, including moving averages, exponential smoothing and autoregressive models, remain strong baselines. Machine learning can add value when demand depends on many explanatory variables, nonlinear effects, interactions or high-dimensional product and customer information.

Potential predictors include price, promotion, calendar effects, weather, search activity, product availability, competitor actions and macroeconomic indicators. However, a predictor must be available at the time the forecast is made. Using information that becomes known only later creates data leakage and invalid performance claims.

Forecasting systems should distinguish at least three quantities:

• point forecast: the expected or central demand estimate;
• prediction interval: a range intended to contain future demand with a stated frequency;
• decision forecast: a forecast adjusted for the cost and consequences of errors.

For a demand forecast \hat{D}_t and actual demand D_t, mean absolute error is:

where n is the number of observations. MAE is expressed in demand units and treats positive and negative errors symmetrically. Mean absolute percentage error can become unstable or undefined when actual demand is zero or very small. Forecast metrics should therefore be selected according to demand characteristics and business decisions.

Probabilistic forecasting is particularly relevant to inventory control because safety stock depends on uncertainty, not only on the mean forecast. A model may have an accurate average forecast but poorly calibrated prediction intervals. Calibration asks whether, for example, intervals labelled 90% contain approximately 90% of future observations over comparable cases.

Machine learning should be evaluated using time-ordered validation rather than random shuffling when temporal dependence exists. Backtesting must reproduce the information available at each historical decision date. Performance should be compared with simple baselines and measured at the level at which decisions are made: item-location, category, network or customer segment.

6. Explainability, Bias and Responsible AI

An accurate model is not automatically a suitable model. Inventory decisions affect customers, suppliers, workers and communities. Explainability is the ability to communicate how a model generated a forecast or recommendation. It may involve global explanations, such as important variables across the system, and local explanations, such as why a particular item received a high replenishment recommendation.

Explainability is valuable for error detection, user trust, auditability and exception management. It should not be confused with a simplistic story that is easy to understand but technically false. Explanations must reflect the model’s actual behaviour and uncertainty.

Bias can arise from historical data, missing observations, measurement practices, proxy variables or unequal service priorities. For example, a model trained on historical sales may interpret past stockouts as low demand, creating a feedback loop that further reduces replenishment. A supplier-risk model may penalize smaller suppliers because historical data contain fewer observations about them, not because they are objectively less reliable.

Governance should include:

• documented data provenance and definitions;
• tests for leakage, drift and subgroup performance;
• review of proxy variables and unintended discrimination;
• clear escalation thresholds;
• model versioning and change control;
• a documented process for correcting errors;
• records of human overrides and their outcomes.

Fairness is context-dependent. Equal service levels may not always be equitable if products are essential, lifesaving or geographically critical. Such decisions require explicit policy rather than hidden model behaviour.

7. Cybersecurity and Data Integrity

Connected inventory systems increase the attack surface. Risks include manipulated stock balances, ransomware, compromised supplier credentials, falsified sensor readings, malicious changes to reorder parameters and denial of access to planning systems.

Controls should follow defence-in-depth principles:

• least-privilege access and multifactor authentication;
• encryption in transit and at rest;
• network segmentation between operational and corporate systems;
• signed or verified data feeds where feasible;
• immutable audit logs;
• backups and tested recovery procedures;
• supplier cybersecurity requirements;
• anomaly detection for impossible or suspicious transactions;
• manual fallbacks for critical replenishment processes.

Cybersecurity is also an inventory issue. If a system cannot be trusted, organizations may need quarantine stock, manual counts, alternate suppliers or temporary safety buffers. The cost of resilience should be compared with the operational and service consequences of system compromise.

8. Human Oversight and Autonomous Replenishment

Autonomous replenishment ranges from automated execution of simple rules to systems that forecast, optimize and place orders with limited human intervention. Autonomy should be graduated rather than treated as a binary choice.

A practical control design may include:

• advisory mode: the system recommends; a planner approves;
• guarded automation: the system executes within quantity, supplier and value limits;
• exception-based autonomy: routine decisions are automatic; unusual cases are escalated;
• full automation for bounded processes: suitable only where consequences are limited and recovery is rapid.

Human oversight must be meaningful. A planner who can only click “approve” without adequate information or time is not exercising real control. The system should show the recommendation, key drivers, confidence or uncertainty, constraint violations, alternatives and consequences of inaction.

Override rates should not be interpreted automatically as evidence that the model is poor. Frequent overrides may indicate model weakness, poor user interface, changed business conditions or valuable expert knowledge. Overrides should be classified, reviewed and fed into model improvement without creating uncontrolled manual bias.

A robust policy defines a “stop button,” fallback rules, approval thresholds and responsibility for losses. Critical items may require dual approval, while low-value and low-risk items can use higher automation.

9. Resilient and Circular Inventory Models

Resilience is the capability to prepare for, respond to and recover from disruption while maintaining important functions. It may require redundancy, flexibility, visibility, collaboration and rapid recovery. The lowest-cost supply chain is not necessarily the lowest-cost system when disruption consequences are included.

Inventory resilience options include:

• strategic safety stock for critical components;
• multiple qualified suppliers;
• geographically distributed capacity;
• postponement and modular product design;
• substitution rules;
• reserved capacity and flexible contracts;
• dual-use or common components;
• recovery and repair networks.

Circular inventory models extend the flow beyond sale. Products may be returned, inspected, repaired, remanufactured, harvested for parts or recycled. Inventory is then classified not only by quantity but also by condition, location, ownership and recoverability.

A circular system must account for uncertain return rates, inspection yield, recovery time, quality grades, cannibalization and customer acceptance. Treating all returned units as immediately available inventory can create service failures and quality risk.

A simplified net-demand relationship is:

where N_t is effective new-unit requirement, D_t is demand, R_t is returned quantity and Y_t is the proportion of returns that pass recovery and become usable within the required period. The relationship is useful only when returned units are functionally substitutable and timing, quality and location are properly represented.

Sustainability metrics should include more than warehouse quantity. Relevant measures may include carbon emissions, material consumption, waste, repair rate, product life extension and the environmental effect of expedited transport or obsolete inventory. Trade-offs should be made explicit rather than assuming that every circular action is automatically beneficial.

10. An Integrated Implementation Framework

Organizations should implement advanced analytics as an operating-model change, not merely as a software purchase.

Step 1: Define the decision and value objective

Specify the decision frequency, planning horizon, service promise, cost categories and risk appetite. Decide whether success means lower working capital, improved availability, reduced waste, faster recovery or a balanced scorecard.

Step 2: Establish data and process foundations

Clean item, location, supplier, unit-of-measure, lead-time and transaction data. Reconcile physical counts with system balances. Document data ownership and the timing of each signal.

Step 3: Build a transparent baseline

Begin with an established forecasting method and inventory policy. A complex model should demonstrate improvement over a credible baseline, not over an unrealistic comparison.

Step 4: Test through backtesting and simulation

Use historical decision-time information, realistic constraints and stress scenarios. Evaluate both average performance and tail outcomes.

Step 5: Pilot with bounded autonomy

Select a segment with measurable demand, manageable risk and adequate data. Begin in advisory mode or guarded automation. Establish a rollback plan.

Step 6: Monitor model and business performance

Monitor forecast error, interval calibration, stockouts, excess inventory, overrides, drift, latency, cyber alerts and subgroup effects. Model accuracy alone is insufficient.

Step 7: Institutionalize governance

Create a cross-functional ownership structure involving supply chain, information technology, finance, operations, risk, cybersecurity and relevant business users. Define who may change parameters, approve deployment and suspend the system.

Illustrative Case: Northstar Medical Devices

Background

Northstar Medical Devices supplies sterile kits to hospitals from two distribution centres. Demand is intermittent and highly variable. A kit stockout may delay treatment, but excessive inventory creates expiry and obsolescence costs. Returns from hospitals may be inspected and, when packaging integrity is confirmed, returned to available stock.

Problem

The existing monthly spreadsheet forecast produced frequent emergency shipments. Planners held large safety stocks for some low-demand kits while critical kits occasionally experienced shortages. Supplier lead times also became less reliable during regional transport disruptions.

Analysis

Northstar created an item-location data set containing demand, stockouts, order dates, receipt dates, expiry dates, returns, supplier performance and emergency shipments. The team discovered that observed sales understated true demand during stockout periods. It therefore treated stockout demand as censored observations rather than ordinary zero demand.

A probabilistic forecasting model was compared with intermittent-demand baselines. Forecasts were evaluated through rolling-origin backtesting. A digital twin represented two warehouses, supplier capacities, expiry states, inspection capacity and emergency transport. Scenario tests included a 20% demand increase, a ten-day supplier delay and a temporary inspection-capacity reduction.

Approach and solution

The organization adopted a rolling-horizon replenishment model with service constraints for critical kits, expiry penalties for short-dated inventory and explicit capacity constraints. Recommendations were placed in guarded automation: routine orders below a value threshold were released automatically, while critical, unusual or short-dated decisions required planner approval.

The system displayed forecast intervals, stockout-adjusted demand, supplier assumptions and the reason for each exception. A cybersecurity review introduced multifactor authentication, segmented access and a manual ordering procedure. Returned products were counted as available only after inspection and quality release.

Outcome

In the pilot category, Northstar reduced emergency shipments and improved availability without simply increasing total stock. The most important benefit was not a single forecast algorithm; it was the integration of better data, scenario testing, quality status, human review and explicit service priorities. The company also identified cases where the model should not act automatically, especially when expiry risk and patient-critical availability conflicted.

Learning points

1. Stockout observations can distort demand models.
2. A digital twin is useful only when its state and constraints are credible.
3. Circular flows require condition and timing information.
4. Automation should be bounded by risk and reversibility.
5. Governance, cybersecurity and process redesign are part of inventory performance.

Challenges and Limitations

Advanced analytics has important limitations. Historical data may be incomplete, biased or generated by earlier policies. Structural breaks can invalidate learned relationships. Rare disruptions provide too few examples for purely data-driven learning. Optimization can produce mathematically optimal but operationally unacceptable solutions if constraints are missing. Simulation can create false confidence when its assumptions are poorly validated.

There are also organizational limitations. Planners may distrust opaque systems, while executives may expect rapid financial returns. Different functions may define service, cost and risk differently. Data ownership may be fragmented. Automation may shift rather than eliminate work, creating new responsibilities for monitoring and exception management.

Regulatory requirements vary by jurisdiction and sector. Medical, food, aerospace and other critical products may require traceability, validation, documented change control and qualified release processes. Privacy obligations apply when customer, employee or behavioural data are used. Legal review is appropriate when automated decisions materially affect suppliers, customers or access to essential products.

Best-Practice Checklist

• Begin with a clearly defined inventory decision and measurable business objective.
• Compare advanced models with transparent baselines.
• Use time-aware validation and prevent data leakage.
• Forecast uncertainty, not only expected demand.
• Represent stockouts, lead-time variation, capacity and quality states.
• Test policies through scenarios and simulation before deployment.
• Make data freshness, assumptions and model versions visible.
• Use graduated autonomy with meaningful human oversight.
• Monitor bias, drift, cybersecurity and operational outcomes.
• Maintain fallback procedures and tested recovery capability.
• Treat returns, repair and remanufacturing as condition-dependent flows.
• Review models after disruptions, policy changes and structural market shifts.

Chapter Summary

Advanced inventory control is a closed-loop decision discipline that combines operational theory with data, models and governance. Optimization supports coordinated decisions under constraints; simulation and scenario analysis reveal variability and disruption consequences; digital twins connect physical operations with analytical experimentation. Machine learning can improve forecasts when data quality, temporal validation and uncertainty estimation are handled correctly.

AI-assisted inventory decisions must be evaluated beyond accuracy. Explainability, bias assessment, cybersecurity, auditability and human accountability determine whether a system is trustworthy and operationally safe. Autonomous replenishment should be introduced gradually, with approval thresholds, exception handling and fallback procedures. Resilient and circular models broaden inventory control from simple quantity management to the management of capacity, condition, recovery, risk and sustainability. The future of inventory control is therefore not unrestricted automation, but governed intelligence that improves decisions while preserving human responsibility.

Key Terms

Digital twin: A continuously updated digital representation of a physical system and its operating relationships.

Dynamic inventory control: Repeatedly updated inventory decision-making using changing information and conditions.

Rolling horizon: A planning method that re-optimizes a future horizon after implementing the next decision.

Stochastic optimization: Optimization that explicitly represents uncertain parameters or outcomes.

Robust optimization: Optimization designed to perform acceptably across a defined range of uncertainty.

Monte Carlo simulation: Repeated random or resampled experimentation used to estimate system outcomes.

Scenario analysis: Evaluation of decisions under coherent alternative future conditions.

Machine learning: Computational methods that learn patterns from data for prediction or decision support.

Data leakage: Use of information in model development that would not be available at decision time.

Probabilistic forecast: A forecast describing a distribution or interval rather than only one value.

Calibration: The correspondence between stated predictive probabilities or intervals and observed outcomes.

Explainability: The capacity to communicate how a model produces outputs or recommendations.

Model drift: Deterioration or change in model relationships as data or operating conditions change.

Guarded automation: Automated execution subject to defined limits, exceptions and controls.

Human oversight: Meaningful human responsibility for reviewing, approving, monitoring or stopping automated decisions.

Resilience: The ability to prepare for, absorb, recover from and adapt to disruption.

Circular inventory: Inventory managed through reuse, repair, remanufacture, recovery or recycling flows.

Data provenance: Information about the origin, transformation and ownership of data.

Service level: A measure of the extent to which inventory satisfies demand or availability requirements.

Review Questions

Short Answer Questions

1. What is the difference between a forecast and a scenario?
2. Why is rolling-origin backtesting preferable to random data splitting for many inventory forecasts?
3. Define a digital twin in the context of inventory control.
4. What is meant by guarded automation?
5. Identify two cybersecurity controls relevant to connected inventory systems.

Long Answer Questions

1. Explain how optimization, simulation and scenario analysis complement one another in inventory decision-making.
2. Discuss the benefits and limitations of machine-learning forecasting compared with traditional forecasting methods.
3. Explain why explainability and bias assessment are necessary even when an AI model has high forecast accuracy.
4. Develop a governance framework for autonomous replenishment in a multi-echelon supply chain.
5. Discuss how resilience and circularity change the definition and measurement of inventory performance.

Analytical / Application Questions

1. A company experiences frequent stockouts and has trained a demand model using sales history that records zero sales during stockouts. Diagnose the likely modelling problem and propose a validation and correction approach.
2. Design a scenario-analysis experiment for a supply chain exposed to uncertain demand, supplier delay and warehouse capacity loss. Specify performance measures and explain how management should use the results.
3. A returned product may be reused only after inspection, with uncertain recovery yield and a ten-day processing time. Explain how the inventory policy should represent this flow and identify the risks of counting all returns as immediately available stock.

Multiple Choice Questions with Answers

Question 1

Which feature most clearly distinguishes a digital twin from a conventional inventory dashboard?

A. It displays historical inventory totals only.

B. It represents system relationships and can test interventions against an updated system state.

C. It eliminates the need for physical inventory records.

D. It guarantees that forecasts are correct.

Correct Answer: B. It represents system relationships and can test interventions against an updated system state.

Question 2

In the reorder-point expression r = μ_L + zσ_L, what does z represent?

A. Unit procurement cost

B. Mean lead time

C. A service-factor associated with the desired probability of avoiding a stockout during lead time

D. The number of warehouses

Correct Answer: C. A service-factor associated with the desired probability of avoiding a stockout during lead time

Question 3

Why can random shuffling be inappropriate when validating a time-dependent inventory forecast?

A. It always reduces the sample size to zero.

B. It may allow information from the future to influence training for an earlier decision.

C. It prevents the use of baseline models.

D. It guarantees underfitting.

Correct Answer: B. It may allow information from the future to influence training for an earlier decision.

Question 4

What is a principal purpose of scenario analysis?

A. To claim that every possible future has a precise probability

B. To replace all operational data with managerial judgement

C. To test policies under coherent alternative conditions, including disruptions

D. To ensure that inventory never reaches zero

Correct Answer: C. To test policies under coherent alternative conditions, including disruptions

Question 5

Which practice is an example of guarded automation?

A. Allowing an algorithm to change all supplier contracts without approval

B. Automatically releasing routine orders within defined value and quantity limits

C. Disabling audit logs to improve processing speed

D. Rejecting every model recommendation

Correct Answer: B. Automatically releasing routine orders within defined value and quantity limits

Question 6

What problem occurs when historical stockouts are treated as ordinary low demand in model training?

A. The model may underestimate true demand and reinforce future stockouts.

B. The model necessarily overestimates every lead time.

C. The model becomes a digital twin automatically.

D. The model cannot calculate any forecast metric.

Correct Answer: A. The model may underestimate true demand and reinforce future stockouts.

Question 7

Which statement about MAE is correct?

A. It is measured in the same units as demand.

B. It is undefined whenever demand is zero.

C. It squares every forecast error.

D. It measures only positive errors.

Correct Answer: A. It is measured in the same units as demand.

Question 8

Why should returned products not automatically be counted as available inventory?

A. Returns always have zero value.

B. They may require inspection, processing time and quality release before substitution is possible.

C. Returned products cannot be recorded digitally.

D. Circular systems prohibit inventory records.

Correct Answer: B. They may require inspection, processing time and quality release before substitution is possible.

Question 9

Which control most directly limits unauthorized access to a replenishment system?

A. Multifactor authentication and least-privilege permissions

B. Increasing the reorder quantity

C. Using a longer forecast horizon

D. Removing supplier data

Correct Answer: A. Multifactor authentication and least-privilege permissions

Question 10

What is the main purpose of a fallback procedure in an autonomous inventory system?

A. To make the model more complex

B. To provide a safe operating method when the automated system is unavailable or untrustworthy

C. To eliminate human accountability

D. To guarantee minimum procurement cost

Correct Answer: B. To provide a safe operating method when the automated system is unavailable or untrustworthy