Digital Warehousing, Automation and Artificial Intelligence

Introduction and context

Warehousing has evolved from a storage function into a digitally coordinated operating system for inventory, fulfilment, transportation and customer service. Modern facilities may receive supplier data through an enterprise resource planning (ERP) system, identify cartons with barcodes or radio-frequency identification (RFID), direct tasks through a warehouse management system (WMS), move loads with autonomous mobile robots (AMRs), and use artificial intelligence (AI) to forecast demand, allocate labour and detect exceptions.

This transformation is not simply a matter of purchasing technology. Digital warehousing is a socio-technical system: information architecture, physical equipment, processes, people, controls and governance must work together. A technically advanced warehouse can still produce poor results if master data are inaccurate, inventory transactions are delayed, interfaces fail, or employees do not trust automated decisions.

The central managerial question is therefore not “Which technology is most advanced?” but “Which combination of capabilities improves service, inventory accuracy, throughput, safety and total cost for this operating context?”

Learning objectives

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

1. Explain the roles and boundaries of WMS, ERP, warehouse control systems and warehouse execution systems.
2. Compare barcodes, RFID, IoT sensing, cloud platforms and digital twins.
3. Evaluate automation technologies including robotics, AS/RS and autonomous vehicles.
4. Describe AI applications in forecasting, slotting, computer vision, labour planning and exception management.
5. Design an integration and implementation approach using appropriate data, process and governance controls.
6. Assess technology investments using operational, financial, human and risk criteria.

1. The digital warehouse as an operating architecture

A warehouse digital architecture normally contains several layers.

The ERP is generally the system of record for enterprise transactions such as purchase orders, sales orders, financial postings and supplier master data. The WMS manages warehouse-specific execution: inventory status by location, receiving, put-away, replenishment, picking, packing, cycle counting and shipping confirmation. A WCS communicates with equipment such as conveyors, sorters and automated storage and retrieval systems (AS/RS). A WES, where deployed, coordinates labour and automation across the warehouse, often making short-horizon decisions about work prioritisation and resource balancing.

These boundaries vary by vendor and design. A small operation may use an ERP warehouse module instead of a separate WMS; a highly automated operation may use a WES as an orchestration layer. Functional labels must therefore be tested against actual capabilities rather than assumed from product names.

Transaction integrity and the system of record

A digital warehouse depends on a single, trusted representation of inventory. Key data objects include item identity, unit of measure, dimensions, weight, lot or serial number, ownership, location, status, handling constraints and order priority. A transaction should identify at least what moved, how much moved, from where, to where, when, by whom or by what equipment, and under what status.

Inventory accuracy can be expressed as:

The numerator counts records whose physical quantity and relevant attributes agree with the system under the organisation’s defined tolerance. The measure is a percentage and applies to a specified sample, location population or stock-keeping-unit (SKU) population. It should not be confused with order accuracy or inventory availability. A warehouse may have accurate records but still lack stock because inventory is allocated to another order, blocked by quality status or physically inaccessible.

Integration principles

Effective integration requires:

• Clear ownership: each data object has a responsible business owner.
• Stable identifiers: SKU, location, order, license-plate and equipment identifiers are consistent across systems.
• Event discipline: receiving, movement, picking and shipping events are recorded once and propagated reliably.
• Acknowledgement and reconciliation: interfaces confirm receipt and expose rejected or duplicated messages.
• Latency awareness: a real-time robotic decision cannot depend on a batch update made several hours earlier.
• Exception visibility: failed interfaces must create actionable alerts rather than silently losing transactions.
• Cybersecurity by design: authentication, least privilege, segmentation, patching and recovery are part of the architecture.

An integration can be technically operational but operationally unsafe. For example, if a host system confirms an order before a robot has completed the pick, inventory and customer status may diverge. Integration testing must therefore include timing, duplicate messages, partial failures, network loss, emergency stops and recovery after restart.

2. Identification technologies: barcodes, RFID and IoT

Barcodes

Barcodes encode data in one- or two-dimensional visual patterns. A scanner reads the symbol when it has sufficient line of sight, contrast, focus and orientation. They are inexpensive, mature and suitable for receiving, picking, packing and shipping. Two-dimensional codes can contain more information than traditional linear codes, although organisations commonly use the code as a key to retrieve richer data from a database.

Limitations include the need for a readable label, potential damage or contamination, and the possibility that operators scan the wrong item or location. Good barcode control therefore includes label standards, verification, duplicate prevention, scan confirmation and human-readable backup information.

RFID

RFID uses radio communication between a reader and a tag. Unlike a barcode, an RFID tag may be read without direct line of sight, and multiple tags may be identified in one interrogation zone. Passive tags obtain energy from the reader field; active tags contain their own power source and generally support longer range or additional sensing, but at higher cost and maintenance complexity.

RFID performance depends on frequency, reader configuration, tag orientation, materials, electromagnetic interference, shielding and the density of objects. Liquids and metals can attenuate or reflect radio signals. RFID is not automatically more accurate than barcoding: poorly designed portals can generate missed reads or duplicate reads. A business case should compare the value of faster capture and reduced manual handling with tag, reader, integration and exception-management costs.

IoT sensing

Internet of Things (IoT) devices collect environmental or operational data such as temperature, humidity, vibration, door status, battery level, location and equipment utilisation. A typical IoT chain is sensor → gateway → network → platform → event rule or analytics application.

IoT creates value when a measurement leads to a controlled response. A temperature sensor is useful for pharmaceuticals only when thresholds, calibration, alarm ownership, escalation and product disposition are defined. Sensor data should have timestamps, device identity, calibration status and quality flags. Excessive unstructured data can increase cost without improving decisions.

3. Automation technologies and operating design

Conveyors, sortation and mechanisation

Conveyors and sorters provide high throughput on stable, repetitive flows. They reduce walking and can improve ergonomic conditions, but they are relatively inflexible. Changes in product dimensions, layout or order profile may require expensive modification. Mechanisation is often appropriate where volume is high and process variation is moderate.

Automated storage and retrieval systems

AS/RS equipment stores and retrieves loads through computer-directed mechanisms. Common forms include unit-load cranes, mini-load systems, shuttle systems, vertical lift modules and automated bin systems. Their economic and operational logic depends on storage density, throughput, load characteristics, ceiling height, labour cost, service requirements and the cost of downtime.

A useful first-order capacity relationship is:

Flow rate may be measured in cases per hour, totes per hour or pallets per hour. Target utilisation is a proportion between 0 and 1 and should include practical allowance for variability, maintenance and recovery. For example, a system expected to process 480 totes per hour at a target utilisation of 0.80 requires nominal capacity of 600 totes per hour. This is not a complete engineering design: peak duration, queueing, travel times, induction limits, replenishment and failure modes must also be modelled.

Storage density can reduce building footprint, but dense systems may increase dependency on equipment availability and software control. A resilience design should specify manual bypasses, spare parts, preventive maintenance, recovery procedures and degraded operating modes.

Autonomous mobile robots and automated guided vehicles

Automated guided vehicles (AGVs) generally follow predefined routes or guidance infrastructure. AMRs use sensors and software to navigate more flexibly around changing environments. Both can support goods-to-person picking, pallet movement, replenishment and line-side delivery.

The appropriate comparison is not “robot versus worker” but redesigned process versus current process. Relevant measures include travel reduction, task cycle time, charging time, fleet congestion, safety separation, pick ergonomics, system availability and integration effort. Fleet-management software must prevent deadlock, prioritise urgent work and coordinate traffic at intersections and shared resources.

Autonomous vehicles require risk assessment, marked zones where appropriate, speed controls, obstacle detection, emergency stops, pedestrian training and procedures for abnormal conditions. Compliance with applicable machinery, occupational safety and electrical standards must be verified in the jurisdiction of operation.

Robotics and collaborative systems

Robotic arms may perform palletising, depalletising, case handling or piece picking. Collaborative robots are designed to share space with people under defined conditions, but “collaborative” does not mean risk-free. Payload, reach, tooling, pinch points, speed, force and task variability must be assessed. Vision-guided picking is effective for some products but remains difficult when items are deformable, reflective, occluded or densely packed.

4. Digital twins and cloud platforms

A digital twin is a continuously updated digital representation of a physical asset, process or facility, connected to relevant data and used for monitoring, analysis, simulation or control. A static three-dimensional layout is a model, not necessarily a digital twin. The distinction lies in data connection, state representation and intended use.

Warehouse twins may represent:

• rack and location geometry;
• inventory positions and order status;
• equipment states and travel paths;
• labour availability and task queues;
• energy, temperature and maintenance conditions; and
• alternative operating scenarios.

A twin can test layout changes, wave-release policies, robot fleet sizes or replenishment rules before implementation. Its credibility depends on model validation. If travel times, equipment failures or human interruptions are omitted, simulation results may systematically overstate performance.

Cloud platforms provide scalable computing, centralised updates, remote access and integration services. They can support multi-site visibility and machine-learning workloads. Risks include connectivity dependence, data residency obligations, vendor lock-in, unclear responsibility for security, subscription escalation and service outages. A cloud decision should specify recovery-point and recovery-time objectives, exit arrangements, interface portability, encryption, identity management and operational ownership.

5. Artificial intelligence applications

AI in warehousing includes statistical learning, optimisation, computer vision, natural-language interfaces and rule-based reasoning. The best application is not always the most complex model. A transparent rule may outperform an opaque model when data are sparse, decisions are low frequency or accountability is critical.

AI-assisted forecasting

Demand forecasting supports purchasing, replenishment, labour planning and slotting. A simple moving average is:

Here, F_t+1 is the forecast for the next period, D_t is observed demand in the most recent period, and n is the number of periods averaged. Demand units must be consistent, and the method assumes that the recent average is informative. It reacts slowly to structural changes and does not explicitly model promotions, seasonality or intermittent demand.

Forecast accuracy should be evaluated using a measure appropriate to the decision. Mean absolute error is:

where m is the number of evaluated periods. MAE is expressed in demand units and treats positive and negative errors symmetrically. Percentage measures can become unstable when actual demand is zero or very small. Forecast evaluation should use time-ordered holdout data and compare against a meaningful baseline, not merely against an earlier version of the model.

AI forecasting can incorporate promotions, weather, price, calendar effects, customer segments and external signals. However, causal interpretation requires care: correlation in historical data does not prove that a variable will cause future demand. Human review remains important for new products, exceptional events and known changes in commercial policy.

AI-supported slotting

Slotting assigns SKUs to locations to reduce travel, congestion, replenishment effort or ergonomic risk. A practical objective may combine weighted travel and handling costs:

where q_i is the expected number of visits or picks for SKU i, c_i is the cost per unit distance or handling action, and d_i is the relevant distance. In practice, the model must include location capacity, compatibility, weight, cube, temperature, hazard, expiry, access frequency, replenishment constraints and co-location rules.

A slotting model trained only on historical picks may reinforce a poor layout or fail when demand changes. Re-slotting also creates disruption: stock must be moved, locations relabelled and system records updated. The value of a mathematically attractive layout must therefore be compared with implementation effort and stability.

Computer vision

Vision systems can inspect labels, count items, verify package dimensions, detect damage and confirm safety conditions. A classification system produces outcomes such as “acceptable” or “exception”; detection systems identify the location of an object; segmentation identifies its pixels or boundary.

Important evaluation terms include:

• Precision: the proportion of flagged cases that are truly positive.
• Recall: the proportion of true positive cases that are detected.
• False positive: an acceptable item incorrectly flagged.
• False negative: a defect or unsafe condition missed by the system.

The preferred balance depends on consequences. A safety inspection may prioritise recall, accepting more false alarms; a low-cost packaging check may prioritise precision to avoid excessive manual review. Models must be tested across lighting, camera angles, packaging variations, seasons and product introductions. Images may contain workers or customer information and therefore require privacy controls.

Labour planning

AI can forecast workload by process and time interval, then recommend staffing, cross-training and shift allocation. A basic workload estimate is:

If task volume is in tasks and standard time is in minutes per task, the result is hours. The estimate assumes the standard time is representative and excludes or separately accounts for breaks, meetings, travel, training, absence and allowance factors. A more realistic plan includes shrinkage and a service constraint; it should not convert a statistical forecast into an automatic employment decision without managerial review.

Exception management

The greatest practical value of AI may be identifying exceptions before they become service failures. Examples include predicted stockouts, unusual pick durations, repeated scan failures, temperature excursions, congestion, equipment degradation and suspected inventory misplacement.

An effective exception system ranks events by risk and actionability. A useful alert should state the event, confidence or evidence, affected order or asset, recommended action, owner, deadline and escalation path. Alert volume must be controlled: if every anomaly produces an urgent notification, users learn to ignore the system.

6. Evaluation and implementation framework

A technology investment should be assessed through a multi-dimensional business case. Benefits may include labour-hour reduction, higher throughput, fewer errors, lower damage, improved space utilisation, reduced injury exposure, better traceability and improved service. Costs include equipment, software licences, integration, civil works, training, maintenance, cybersecurity, changeover disruption and energy.

A simple payback period is:

The formula assumes a constant annual net benefit and ignores the time value of money, taxes, residual value and benefits occurring at different times. For strategic investments, discounted cash-flow measures such as net present value are more appropriate. Non-financial constraints—safety, regulatory traceability, resilience and employee acceptance—should not be reduced to a single payback number.

A robust implementation proceeds through stages:

1. Diagnose the current state: map physical and information flows, identify failure modes and establish baseline measures.
2. Define the target operating model: specify service levels, process ownership, system boundaries and human roles.
3. Cleanse and govern data: validate dimensions, weights, units, locations, barcodes, status codes and master-data ownership.
4. Select technology by use case: compare alternatives against demand variability, volume, product profile, risk and scalability.
5. Pilot safely: test representative products, peak conditions, exceptions and recovery procedures.
6. Integrate and validate: conduct unit, interface, performance, cybersecurity, user-acceptance and operational-readiness testing.
7. Train and transition: use simulation, super-users, standard work and a controlled cutover plan.
8. Stabilise and improve: monitor leading and lagging indicators, tune parameters and review model drift.

Useful measures include order cycle time, dock-to-stock time, pick rate, lines per labour hour, inventory accuracy, order accuracy, on-time shipment, equipment availability, mean time to repair, energy per unit and exception closure time. Measures should be segmented by shift, zone, SKU family and process; aggregate averages can conceal local failure.

7. Governance, ethics and resilience

Digital warehouses process operational, employee and sometimes customer data. Governance should address:

• data minimisation and retention;
• access control and audit trails;
• privacy notices and lawful processing;
• model documentation and validation;
• human review for consequential decisions;
• bias and disparate impact in productivity or scheduling models;
• cybersecurity and ransomware response;
• safety validation for automated equipment;
• supplier responsibility and software-update controls; and
• continuity during network, power, platform or equipment failure.

Worker monitoring deserves special attention. Productivity analytics can improve planning but may create excessive surveillance, unsafe speed pressure or unfair comparisons between tasks. Standards must account for product mix, congestion, equipment faults, training and reasonable human variation. Workers should know what is measured, why it is measured and how disputed data are corrected.

AI governance should establish model owners, approval thresholds, retraining triggers, performance dashboards and an override process. A model’s confidence score is not a guarantee of correctness. Decisions affecting safety, employment, quality release or regulated products require appropriate human accountability.

8. Illustrative case: omni-channel distribution centre transformation

Background

The following is a fictional illustrative case. Meridian Home Products operates a regional distribution centre supplying stores and direct-to-consumer customers. It carries 18,000 SKUs, experiences strong seasonal peaks and uses paper-based replenishment with a basic ERP inventory module.

Problem

Inventory accuracy is 94%, order accuracy is 97.8%, and pickers spend a large share of each shift walking. Peak-season overtime is high, while fast-moving SKUs frequently compete for the same aisle locations. Management proposes RFID, AMRs, an AI slotting model and a cloud WMS.

Analysis

A process study finds that the largest losses are not caused by insufficient storage capacity. They arise from late receiving transactions, inconsistent units of measure, replenishment performed after pick faces are empty, and weak exception ownership. A full automation project would therefore amplify unreliable data. The team defines baseline measures, cleanses item dimensions and assigns ownership for receiving, location and inventory-status data.

A value-stream analysis separates stable store-case flows from variable e-commerce piece picking. RFID is selected for reusable store totes and high-value returnable assets, while barcodes remain on individual consumer products because their labels already exist and RFID tagging would not yet produce sufficient value. A cloud WMS is integrated with the ERP, and a WES coordinates AMRs in the piece-picking zone.

Approach and solution

The implementation occurs in three waves. First, the company introduces barcode-directed receiving, location scanning, cycle-count rules and interface reconciliation. Second, it deploys AMRs for goods-to-person movement in a pilot zone and uses a rules-based slotting engine constrained by product compatibility and replenishment capacity. Third, it adds machine-learning forecasts for labour planning and computer vision for carton-label verification.

The AI model recommends slots, but a supervisor approves changes during the initial period. A control group of comparable aisles is retained to distinguish technology effects from seasonal changes. Exception dashboards show failed scans, predicted stockouts, abnormal pick times and equipment downtime.

Outcome

After stabilisation, the illustrative operation reports inventory accuracy of 99.2%, order accuracy of 99.5%, and a 31% reduction in picker travel in the pilot zone. Peak overtime falls, but the AMR area initially experiences congestion because release rules send too many tasks to a shared induction point. Revising task release and adding a second induction position improves flow. The case demonstrates that benefits arise from coordinated process redesign, data quality and controlled scaling—not from robotics alone.

Learning points

1. The correct sequence is often data and process discipline before advanced AI.
2. Different identification technologies may coexist; one technology need not replace all others.
3. Pilot design should include a comparison group and peak-like conditions.
4. Automation creates new bottlenecks, especially at interfaces and shared resources.
5. Supervisory judgement, worker participation and exception governance remain essential.

Contemporary developments and future direction

Current development is moving toward composable warehouse platforms, edge analytics, digital work instructions, foundation models for operational assistance, energy-aware scheduling and more capable robotic perception. These developments may reduce integration effort and improve adaptability, but they also increase dependence on data quality, connectivity and vendor ecosystems.

Future research questions include how to validate AI under distribution shift, how to coordinate human and robot learning, how to measure resilience rather than only efficiency, and how to design sustainable automation with repairable equipment and lower energy use. The strategic capability is not merely technology adoption; it is the organisation’s ability to sense operations, learn safely and redesign work continuously.

Chapter Summary

Digital warehousing integrates enterprise planning, WMS execution, equipment control, identification technologies, automation, cloud infrastructure, digital twins and AI. ERP systems coordinate enterprise transactions, while WMS and WES/WCS manage increasingly real-time warehouse activity. Barcodes remain economical and reliable when line-of-sight scanning is acceptable; RFID enables non-line-of-sight identification but requires careful attention to materials and read performance; IoT adds condition and equipment visibility.

AS/RS, conveyors, robots, AGVs and AMRs can increase throughput and reduce travel, but they introduce capital intensity, integration dependency and new failure modes. Digital twins support scenario analysis only when they are connected to credible operational data and validated against reality. AI can improve forecasting, slotting, computer vision, labour planning and exception management, but model quality, interpretability, drift, privacy and human accountability must be governed.

Successful implementation begins with process diagnosis, master-data discipline and a clearly defined target operating model. Evaluation should combine financial measures with service, safety, resilience, workforce and regulatory considerations. The most effective digital warehouse is not the one with the greatest number of technologies; it is the one in which technology, people, processes and governance produce dependable decisions and physical execution.

Key Terms

WMS: Warehouse management system that directs and records warehouse processes.

ERP: Enterprise resource planning system coordinating enterprise-wide transactions and resources.

WES: Warehouse execution system that coordinates labour and automation over short operating horizons.

WCS: Warehouse control system that communicates with and controls automated material-handling equipment.

Barcode: Optical representation of encoded information read by a scanner.

RFID: Radio-frequency identification using tags and readers to identify objects wirelessly.

IoT: Network of sensors and connected devices that capture and transmit operational data.

AS/RS: Automated storage and retrieval system for computer-directed storage and retrieval.

AGV: Automated guided vehicle operating with predefined guidance or routes.

AMR: Autonomous mobile robot navigating dynamically using sensors and software.

Digital twin: Data-connected digital representation of a physical asset, process or facility.

Slotting: Assignment of SKUs to storage or picking locations.

Computer vision: Analysis of images or video to detect, classify or measure objects and conditions.

Master data: Core reference information such as SKU, location, unit and handling attributes.

Exception management: Identification, prioritisation and resolution of deviations from expected operation.

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

Precision: Proportion of model-positive cases that are truly positive.

Recall: Proportion of true positive cases detected by a model.

Inventory accuracy: Agreement between recorded and physically verified inventory information.

Warehouse automation: Use of mechanised, robotic or software-controlled systems to perform warehouse activities.

Review Questions

Short Answer Questions

1. Distinguish the typical responsibilities of an ERP, WMS and WCS.
2. State two advantages and two limitations of RFID compared with barcodes.
3. What makes a digital model a digital twin?
4. Define precision and recall in the context of computer vision.
5. Why is master-data quality important before deploying AI or automation?

Long Answer Questions

1. Explain how a WMS, WES and WCS can be integrated in a highly automated warehouse. Discuss transaction integrity and failure recovery.
2. Compare conveyors, AS/RS, AGVs and AMRs in terms of flexibility, throughput, capital commitment and operational risk.
3. Evaluate the use of AI for demand forecasting and slotting, including data requirements, model limitations and human governance.
4. Develop an implementation framework for a cloud-based digital warehouse transformation.
5. Discuss the ethical, cybersecurity, safety and workforce implications of digitally monitored warehouse operations.

Analytical / Application Questions

1. A warehouse forecasts 480 tote movements per hour and sets a target utilisation of 0.80. Calculate the nominal throughput capacity required, explain the assumptions, and identify two reasons why an engineering design may require additional capacity.
2. A distribution centre is choosing between barcode scanning and RFID for reusable transport totes. Construct a decision analysis considering read conditions, cost, data quality, exception handling and expected operational benefit.
3. Design a pilot for AI-based labour planning in a seasonal warehouse. Specify the baseline, data inputs, outcome measures, control mechanisms, human review process and criteria for scaling.

Multiple Choice Questions with Answers

Question 1

Which system most typically manages location-level inventory, receiving, picking and shipping execution?

A. Warehouse management system (WMS)

B. General ledger

C. Customer relationship management system

D. Computer-aided design system

Correct Answer: A. Warehouse management system (WMS)

Question 2

Which identification technology generally requires line of sight between the scanner and the label?

A. Passive RFID

B. Barcode

C. IoT temperature sensor

D. Bluetooth gateway

Correct Answer: B. Barcode

Question 3

A warehouse must process 480 totes per hour at a target utilisation of 0.80. What nominal capacity is indicated by the first-order relationship?

A. 384 totes per hour

B. 480 totes per hour

C. 600 totes per hour

D. 800 totes per hour

Correct Answer: C. 600 totes per hour

Question 4

Which statement best distinguishes a digital twin from a static layout model?

A. A digital twin is always a three-dimensional drawing

B. A digital twin is connected to relevant physical or operational data

C. A digital twin can only represent robotic equipment

D. A digital twin eliminates the need for validation

Correct Answer: B. A digital twin is connected to relevant physical or operational data

Question 5

In computer vision, recall is the proportion of:

A. Flagged cases that are truly positive

B. True positive cases that are detected

C. All cases that receive a confidence score

D. False positives among all negative cases

Correct Answer: B. True positive cases that are detected

Question 6

Which is a major reason to retain human review when AI forecasts demand for a new product?

A. Historical data may be limited or unrepresentative

B. AI models cannot process numerical data

C. New products always have zero demand

D. Human review makes data governance unnecessary

Correct Answer: A. Historical data may be limited or unrepresentative

Question 7

What is the primary role of a WCS in an automated warehouse?

A. Prepare corporate financial statements

B. Control and communicate with material-handling equipment

C. Set employment law policy

D. Create customer advertising campaigns

Correct Answer: B. Control and communicate with material-handling equipment

Question 8

Which situation is an example of a false negative in a vision inspection system?

A. An acceptable carton is incorrectly rejected

B. A defective carton is incorrectly accepted

C. A camera records an image twice

D. A scanner reads a valid barcode

Correct Answer: B. A defective carton is incorrectly accepted

Question 9

Which implementation sequence is generally most appropriate?

A. Purchase robots, then define processes and data ownership

B. Deploy AI first, then establish performance measures

C. Diagnose the current state, govern data, pilot safely and scale based on evidence

D. Replace all existing systems before testing interfaces

Correct Answer: C. Diagnose the current state, govern data, pilot safely and scale based on evidence

Question 10

Why should warehouse productivity analytics be governed carefully?

A. They can create surveillance, unfair comparisons or unsafe work pressure

B. They cannot be used to improve planning

C. They are unrelated to employee experience

D. They automatically provide legally valid employment decisions

Correct Answer: A. They can create surveillance, unfair comparisons or unsafe work pressure