Functional and Industry Applications

Introduction

Generative artificial intelligence (GenAI) creates new text, images, audio, video, software code, synthetic data, and structured recommendations from learned patterns. In business, its importance is not simply that it can produce content quickly. Its strategic value lies in its ability to augment decisions, redesign workflows, personalize interactions, and make specialized knowledge more accessible.

However, business application is not equivalent to automation. A useful application must connect a clearly defined business problem with an appropriate model, reliable data, human accountability, and measurable performance. A fluent response may still be factually wrong, legally risky, biased, or commercially ineffective. Consequently, GenAI should be evaluated as part of a business system rather than as an isolated software feature.

This chapter examines applications in marketing, finance, operations, supply chains, human resources, customer service, and professional services. It emphasizes use cases, benefits, limitations, governance, and performance measurement.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain how GenAI creates value across major business functions.
2. Distinguish augmentation, automation, and decision support.
3. Select suitable performance measures for GenAI applications.
4. Analyze risks involving accuracy, privacy, bias, security, and accountability.
5. Evaluate an integrated organizational case involving GenAI.
6. Recommend implementation practices that connect technology with business outcomes.

1. A Framework for Business Application

A practical GenAI application can be analyzed through six linked elements:

The distinction between three modes of use is particularly important:

• Augmentation: GenAI assists a person, while the person remains responsible for the decision.
• Automation: GenAI performs a defined task with limited human intervention.
• Decision support: GenAI summarizes evidence or proposes alternatives, but does not determine the final action.

The more consequential the decision, the stronger the required human oversight. Drafting a product description may be low risk; recommending an employment decision or approving a loan is high risk.

A business case should include both benefits and costs. A simple net-benefit model is:

Gross benefit may include additional revenue, labor-time savings, quality improvements, or avoided losses. Expected loss from risk can be estimated as:

These expressions are decision models, not precise predictions. They assume that benefit and cost estimates are measured over the same time period and expressed in comparable monetary units.

2. Marketing and Sales

Applications

Marketing is one of the most visible areas of GenAI adoption. Common applications include:

• generating campaign concepts, copy, images, and video variations;
• adapting content to languages, channels, and customer segments;
• summarizing market research and social-media feedback;
• producing conversational product recommendations;
• supporting sales representatives with account summaries and proposal drafts;
• simulating customer personas for early message testing.

GenAI can shorten the content-production cycle. A marketing analyst may provide a campaign objective, brand guidelines, audience description, and legal constraints; the system then produces alternatives for human review. Retrieval-augmented generation (RAG) can connect the model to approved product information, reducing reliance on unsupported general knowledge.

Benefits and limitations

Benefits include faster experimentation, lower production cost, improved personalization, and greater accessibility for small marketing teams. Yet personalization can become intrusive if customers do not understand how their data is being used. Generated content may also contain stereotypes, unverified claims, copyright problems, or an inconsistent brand voice.

Marketing teams should distinguish creative quality from business effectiveness. A visually attractive advertisement is not necessarily persuasive. Useful measures include conversion rate, incremental revenue, cost per acquisition, engagement quality, unsubscribe rate, complaint rate, and brand-safety incidents.

For a campaign, conversion rate is:

An experiment should compare a treatment group exposed to GenAI-supported content with a control group receiving the existing content. Random assignment and a sufficiently large sample improve the credibility of the comparison. The correct question is not whether generated content performs well in isolation, but whether it performs better than a realistic alternative.

3. Finance and Accounting

Applications

Financial institutions and corporate finance teams use GenAI for document review, financial-report commentary, investor-question preparation, policy search, transaction monitoring support, and scenario analysis. It can extract information from invoices, contracts, filings, and expense documents; summarize management reports; and generate first drafts of explanations.

In investment research, a model may organize public information and identify themes for analysts to investigate. In commercial banking, it may assist relationship managers by summarizing customer histories. In accounting, it may classify transactions or explain unusual movements in an account.

Controls and limitations

Financial use requires particularly strong controls. Hallucinated facts, incorrect calculations, stale information, data leakage, and overconfident recommendations can cause material loss. A language model should not be treated as an authoritative ledger or valuation engine unless its outputs are independently verified.

Important controls include source citation, reconciliation to system-of-record data, segregation of duties, access controls, audit logs, model validation, and approval thresholds. Numerical calculations should preferably be performed by a validated rules engine, spreadsheet, or analytical program, with GenAI used to explain results rather than invent them.

Useful measures include extraction accuracy, reconciliation error rate, review time, false-positive rate, exception detection rate, regulatory findings, and cost per processed document. For classification tasks, precision and recall are useful:

Precision measures how many flagged cases are actually relevant; recall measures how many relevant cases are detected. The appropriate balance depends on context. Fraud screening may prioritize recall, while an expensive manual investigation may require higher precision.

4. Operations and Manufacturing

Operations applications include work-instruction generation, maintenance support, quality inspection assistance, incident analysis, scheduling suggestions, and process-documentation search. A technician might ask a system to retrieve the correct procedure for a machine condition. An operations manager might use GenAI to summarize recurring causes of delay from maintenance logs and shift reports.

GenAI can also act as a natural-language interface to analytical systems. For example, a manager may ask, “Which production lines experienced the greatest downtime last month?” The system should translate the question into a validated query, show the data source, and display the result for review.

The main operational risks are incorrect instructions, unsafe recommendations, weak integration with real-time systems, and automation bias. In a safety-critical environment, generated instructions must be bounded by approved procedures and reviewed by qualified personnel.

Performance measures include cycle time, first-pass yield, downtime, mean time to repair, scrap rate, schedule adherence, worker adoption, and safety incidents. A useful operational assessment compares the whole process, including time spent reviewing AI outputs, rather than measuring only generation speed.

5. Supply Chains and Procurement

Supply-chain professionals use GenAI to summarize supplier communications, draft requests for quotation, compare contract clauses, identify disruption signals, classify spend, and support demand-planning discussions. It can combine information from purchase orders, supplier reports, logistics updates, weather notices, and public news.

A major opportunity is exception management. Instead of requiring managers to inspect every transaction, GenAI can rank exceptions and explain why a shipment, supplier, or order deserves attention. This is most effective when connected to structured analytics and clear escalation rules.

Demand planning should not rely on generated prose alone. Forecast accuracy can be expressed as mean absolute percentage error (MAPE):

Here, A_t is actual demand, F_t is forecast demand, and n is the number of observations. MAPE is interpretable as a percentage but becomes unstable when actual demand is zero or close to zero. In such cases, alternatives such as mean absolute error or weighted measures may be more appropriate.

GenAI may improve planner productivity without improving forecast accuracy. Therefore, organizations should measure both planning effort and supply outcomes, including stockouts, excess inventory, supplier lead time, expedited freight, and service level.

6. Human Resources

HR applications include drafting job descriptions, answering policy questions, supporting onboarding, summarizing employee feedback, recommending learning resources, and assisting workforce planning. A secure internal assistant can help employees locate policies without requiring an HR specialist to answer every routine question.

Recruitment is a high-risk domain. Generating interview questions or summarizing résumés may be acceptable with controls, but automatically ranking candidates can reproduce historical discrimination or disadvantage applicants whose language, education, disability, or career history differs from the training data.

Fairness should be assessed across relevant groups using legally and contextually appropriate measures. One commonly discussed measure is selection-rate parity:

The selection rate for a group is the proportion of applicants selected from that group. This ratio alone does not establish fairness, because legitimate job-related differences, sample size, measurement error, and applicable law must also be considered. Fairness assessment should involve legal specialists, HR experts, affected stakeholders, and statistical analysis.

Performance measures include time to hire, cost per hire, employee-question resolution time, training completion, retention, employee satisfaction, complaint rates, and disparate-impact indicators. Human review must be meaningful rather than a ceremonial approval of an automated recommendation.

7. Customer Service

Customer-service systems can draft responses, classify intent, retrieve knowledge-base articles, translate conversations, summarize calls, and route complex cases. A well-designed assistant helps agents find accurate information while allowing them to adapt tone and exercise judgment.

Full automation is appropriate only for bounded, low-risk interactions. Billing disputes, safety complaints, vulnerable customers, legal threats, and requests involving sensitive personal information should be escalated according to explicit rules.

Important measures include first-contact resolution, average handling time, wait time, customer satisfaction, customer-effort score, escalation rate, repeat-contact rate, resolution accuracy, and hallucination rate. Reducing average handling time at the expense of repeat contacts is not genuine improvement.

A service-quality dashboard should combine efficiency, effectiveness, and experience. For example, an AI assistant may reduce handling time by 20 percent but increase repeat contacts by 10 percent. Management must evaluate the net effect on customer outcomes and total cost, not celebrate a single favorable metric.

8. Professional Services

Professional services include consulting, law, accounting, architecture, engineering, medicine, education, and research. GenAI can support literature review, document comparison, meeting preparation, proposal writing, coding, contract analysis, and knowledge management.

The central issue is professional responsibility. A lawyer remains responsible for a filing containing an invented case citation; an auditor remains responsible for an unsupported conclusion; and an engineer remains responsible for an unsafe design. GenAI can expand a professional’s capacity, but it does not transfer accountability to the software vendor.

Professional firms should define approved uses, prohibited uses, confidentiality rules, citation requirements, client-disclosure practices, and review standards. They should also maintain version histories so that important outputs can be traced to sources, prompts, tools, and human approvals.

9. Integrated Illustrative Case: Northstar Home Appliances

Background

Northstar Home Appliances is a fictional regional manufacturer and retailer. It sells through its website, stores, and independent dealers. The company has a large product-support knowledge base but receives repeated customer questions, experiences seasonal stockouts, and spends significant time producing marketing content.

Problem

Customer-service agents search several disconnected systems. Marketing staff create separate product descriptions for different channels. Procurement teams receive supplier updates in inconsistent formats. Senior management proposes a general-purpose chatbot, but the analytics team argues that an uncontrolled chatbot could generate unsupported product claims and expose customer information.

Analysis

The team maps the workflow and identifies three initial use cases:

1. An internal service assistant grounded in approved manuals and warranty policies.
2. A content assistant that drafts channel-specific descriptions from approved product data.
3. A supplier-update summarizer that highlights late shipments and required actions.

The team rejects automatic warranty decisions and unsupervised purchasing recommendations because these activities have financial and legal consequences.

Approach

Northstar creates a controlled architecture. Product manuals, policies, and specifications are stored in a permissioned knowledge repository. The assistant retrieves relevant passages before generating an answer and displays citations. Customer data is masked where possible. Responses involving refunds, safety, or legal claims are routed to a trained employee.

A pilot compares 40 service agents using the assistant with 40 agents using the existing search process. Assignment is randomized where operationally feasible. Measures include search time, first-contact resolution, response accuracy, escalation rate, and customer satisfaction. Marketing content is tested using a controlled online experiment rather than being judged only by staff preference.

Outcome

In this illustrative case, the pilot shows shorter information-search time and higher agent satisfaction. Accuracy improves for common product questions but declines when manuals are outdated or customer descriptions are ambiguous. The company therefore introduces document-ownership rules, monthly knowledge reviews, and a visible “source unavailable” response rather than allowing unsupported answers.

The content assistant increases production capacity, but only a minority of generated variants outperform the existing control in customer conversion. Northstar keeps human editorial review and uses the system for ideation and adaptation rather than unrestricted publication.

Learning points

First, the best starting point is a bounded workflow with clear data and ownership. Second, retrieval and citations reduce but do not eliminate hallucination. Third, productivity measures must be paired with outcome measures. Fourth, refusing or escalating a request can be a sign of a well-governed system, not a system failure. Finally, implementation is organizational: it requires process redesign, training, accountability, and measurement.

10. Cross-Functional Performance Measurement

A balanced evaluation framework should cover five dimensions:

A useful evaluation process begins with a baseline. Without a baseline, an organization cannot determine whether GenAI produced improvement. Evaluation should also include representative edge cases, not only average transactions. Test sets should contain ambiguous, adversarial, sensitive, and out-of-date inputs.

Human evaluation is necessary for qualities such as tone, usefulness, and professional judgment, but it should use explicit rubrics. Automated measures are efficient but may fail to detect subtle factual or ethical problems. Production monitoring should track drift, changing customer behavior, knowledge-base updates, and model changes.

11. Governance, Ethics, and Regulation

GenAI governance is the allocation of authority, responsibility, and controls across the AI life cycle. Core principles include:

• Validity and reliability: outputs should be fit for the intended use.
• Transparency: users should understand when AI is involved and where information came from.
• Privacy: personal and confidential data should be collected and processed lawfully and minimally.
• Fairness: systems should be tested for unjustified disadvantage.
• Security: prompts, data, models, and interfaces require protection from unauthorized access and manipulation.
• Accountability: a named person or function must own outcomes.
• Contestability: affected people should have a way to question or appeal important decisions.

Governance should be risk-based. Low-risk drafting may require disclosure and review; high-risk employment, credit, health, or safety applications require stronger validation, documentation, monitoring, and human authority. Organizations must also consider intellectual property, copyright, consumer-protection law, employment law, financial regulation, and sector-specific requirements. Legal obligations vary by jurisdiction, so this chapter does not replace professional legal advice.

12. Implementation Best Practices

A practical implementation sequence is:

1. Identify a measurable business problem rather than beginning with a model.
2. Classify the use case by potential harm and reversibility.
3. Establish data ownership, access permissions, and retention rules.
4. Select the simplest reliable technical approach, such as retrieval, rules, or a specialized model.
5. Design human review, escalation, and override procedures.
6. Pilot against a documented baseline.
7. Evaluate quality, cost, business outcomes, and risks together.
8. Train users to verify outputs and report failures.
9. Monitor performance after deployment.
10. Retire or redesign the application when benefits no longer justify risks.

A common failure is to measure only model capability. A technically impressive model may produce little value if employees do not trust it, data is inaccessible, workflows are unchanged, or review costs exceed savings. Business analytics professionals therefore need skills in process mapping, experimentation, data governance, communication, and ethical reasoning in addition to model literacy.

Chapter Summary

GenAI has practical applications across marketing, finance, operations, supply chains, HR, customer service, and professional services. Its value comes from improving a business workflow, not merely generating plausible content. Augmentation, automation, and decision support involve different levels of human responsibility and risk.

Marketing applications emphasize content variation and personalization; finance requires reconciliation, auditability, and numerical control; operations and supply chains benefit from knowledge retrieval and exception management; HR requires careful fairness and privacy assessment; customer service depends on accurate knowledge and escalation; and professional services require preservation of professional accountability.

A robust evaluation combines quality, efficiency, business outcomes, human experience, and risk measures. Baselines, controlled pilots, representative test cases, human review, source grounding, monitoring, and clear governance are essential. Responsible implementation treats GenAI as a component of a socio-technical system involving people, processes, data, technology, and regulation.

Key Terms

Generative artificial intelligence: Technology that produces new content or outputs from learned patterns.

Augmentation: Use of AI to assist a human who remains responsible for the work.

Automation: Performance of a defined task with limited human intervention.

Decision support: AI-generated evidence, summaries, or recommendations used by a decision-maker.

Retrieval-augmented generation: A method that retrieves relevant sources before generating a response.

Hallucination: An unsupported or fabricated AI output presented as if it were valid.

Groundedness: The extent to which an output is supported by reliable source material.

Baseline: The performance of an existing process used for comparison.

Conversion rate: The percentage of eligible exposures that produce a defined action.

Precision: The proportion of positive classifications that are actually positive.

Recall: The proportion of actual positive cases that are detected.

MAPE: Mean absolute percentage error, a forecast-error measure expressed as a percentage.

Human-in-the-loop: A design in which a person reviews or approves an AI output.

Model drift: Decline in performance caused by changing data, behavior, or conditions.

Prompt injection: An attempt to manipulate an AI system through malicious or conflicting instructions.

Selection-rate ratio: A comparison of selection rates across groups.

Exception management: Prioritizing unusual or risky cases for human attention.

Auditability: The ability to reconstruct how an output or decision was produced.

Review Questions

Short Answer Questions

1. Distinguish augmentation from automation.
2. Why should GenAI applications be evaluated against a baseline?
3. State one benefit and one limitation of GenAI in marketing.
4. Why is retrieval-augmented generation useful in customer service?
5. Name three dimensions that should appear in a GenAI performance dashboard.

Long Answer Questions

1. Explain how GenAI can be applied in finance and discuss the controls needed to reduce error and accountability risks.
2. Compare GenAI applications in supply chains and human resources, paying particular attention to measurement and fairness.
3. Explain why professional responsibility remains with the practitioner when GenAI is used.
4. Discuss the differences between quality, efficiency, and business-outcome measures.
5. Describe a risk-based governance framework for deploying GenAI in a large organization.

Analytical / Application Questions

1. A retailer wants to automate product recommendations using customer histories. Design a pilot, identify a control group, specify at least four measures, and explain privacy and fairness safeguards.
2. A service chatbot reduces average handling time but increases repeat contacts. Analyze possible causes and recommend a revised evaluation approach.
3. A manufacturer wants to use GenAI to recommend maintenance actions for safety-critical equipment. Decide what should be automated, what should require human approval, and how the system should be monitored.

Multiple Choice Questions with Answers

Question 1

Which statement best describes augmentation?

A. AI performs all decisions without human review

B. AI assists a person who remains responsible for the work

C. AI is used only to store historical data

D. AI replaces the organization’s entire workflow

Correct Answer: B. AI assists a person who remains responsible for the work

Question 2

What is the main purpose of retrieval-augmented generation?

A. To eliminate the need for source data

B. To make every response creative rather than factual

C. To retrieve relevant information before generating a response

D. To prevent users from reviewing outputs

Correct Answer: C. To retrieve relevant information before generating a response

Question 3

A campaign has 2,000 eligible exposures and 100 conversions. What is the conversion rate?

A. 2%

B. 5%

C. 20%

D. 50%

Correct Answer: B. 5%

Question 4

Which finance control is most appropriate for an AI-generated financial explanation?

A. Publish it without review if the language is fluent

B. Reconcile it with system-of-record data

C. Remove all source citations

D. Allow the model to change ledger entries automatically

Correct Answer: B. Reconcile it with system-of-record data

Question 5

Recall measures:

A. The proportion of flagged cases that are correct

B. The proportion of actual relevant cases that are detected

C. The monetary value of all detected cases

D. The average time required to review a case

Correct Answer: B. The proportion of actual relevant cases that are detected

Question 6

Why can MAPE be problematic when actual demand is zero or close to zero?

A. It cannot be expressed as a percentage

B. It ignores forecast values completely

C. Its percentage error can become unstable

D. It always produces a negative result

Correct Answer: C. Its percentage error can become unstable

Question 7

Which HR use case generally requires the strongest fairness controls?

A. Drafting an internal meeting agenda

B. Generating a generic training-room booking message

C. Automatically ranking job applicants

D. Formatting a public holiday calendar

Correct Answer: C. Automatically ranking job applicants

Question 8

Which measure best captures whether a customer’s issue was solved without another contact?

A. First-contact resolution

B. Model parameter count

C. Prompt length

D. Content generation speed

Correct Answer: A. First-contact resolution

Question 9

What is a central governance principle for high-consequence AI decisions?

A. No human should be able to override the system

B. Accountability should be assigned to a named person or function

C. Outputs should never be logged

D. Testing should use only average cases

Correct Answer: B. Accountability should be assigned to a named person or function

Question 10

In the Northstar case, why were automatic warranty decisions rejected?

A. They had no possible business value

B. They involved financial and legal consequences

C. The company did not use customer service

D. Generated text cannot be displayed on a website

Correct Answer: B. They involved financial and legal consequences