Skip to main content

Chapter 01 Becoming an AI-Augmented Marketing Analyst: Chapter 01

Chapter 01 Becoming an AI-Augmented Marketing Analyst
Chapter 01
  • Show the following:

    Annotations
    Resources
  • Adjust appearance:

    Font
    Font style
    Color Scheme
    Light
    Dark
    Annotation contrast
    Low
    High
    Margins
  • Search within:
    • My Notes + Comments
    • Notifications
    • Privacy
  • Project HomeAPPLIED BUSINESS ANALYTICS FOR MARKETING DECISION-MAKING
  • Projects
  • Learn more about Manifold

Notes

table of contents
  1. Chapter Information
    1. Chapter Learning Objectives
    2. 1.1 Marketing Decision Context: The Holiday Budget Decision
    3. 1.2 What Does a Marketing Analyst Do?
      1. 1.2.1 The Analyst as Translator
      2. 1.3 Business Analytics, Marketing Analytics, and Data-Informed Decisions
        1. 1.3.1 Why Data-Informed, Not Data-Driven?
      3. 1.4 The Four Types of Analytics
        1. 1.4.1 Example: Email Campaign Performance
      4. 1.5 From Business Question to Analytical Question
      5. 1.6 The Marketing Analytics Workflow
      6. 1.7 The AI-Assisted Workflow Used in This Guide
        1. 1.7.1 Example: A Weak Prompt and a Stronger Prompt
      7. 1.8 Key Marketing Data Concepts
      8. 1.9 AI in Marketing Analytics: Benefits and Risks
      9. 1.10 Marketing Interpretation and Managerial Insight
        1. 1.10.1 Revisiting the StyleCraft Case
      10. 1.11 Business Analytics in Practice
        1. 1.11.1 When the Coding Gets Easier, the Framing Gets Harder
        2. 1.11.2 The Deliverable That Could Not Be Verified
        3. 1.11.3 What Hiring Managers Can Test For
        4. 1.11.4 In Your First Analyst Job
      11. 1.12 Ethics, Privacy, and Responsible Analytics
      12. 1.13 Chapter Summary
      13. 1.14 Exercises for Practice
        1. 1.14.1 Core Chapter PracticeExercise 1.1 Concept Check
        2. Exercise 1.2 Classify the Analytics Type
        3. Exercise 1.3 Match the Metric to the Question
        4. Exercise 1.4 AI-Assisted Practice
        5. Exercise 1.5 Managerial Memo
      14. 1.15 Glossary of Terms
      15. 1.16 References

Chapter 01

Becoming an AI-Augmented Marketing Analyst

From Business Questions to Evidence-Based Marketing Decisions

Dr. Jose Mendoza, Academic Director and Clinical Associate Professor

Version 1.0 · July 2026

Except where otherwise noted, this chapter is licensed under CC BY 4.0.

Chapter Information

ABSTRACT

This chapter presents marketing analytics as disciplined decision support and defines the role of the AI-augmented marketing analyst. It distinguishes business analytics, marketing analytics, and data-informed decision-making; explains descriptive, diagnostic, predictive, and prescriptive analytics; and shows how business questions become analytical questions. A seven-step workflow adapted from CRISP-DM is paired with four habits for accountable AI use: specify, predict-then-verify, explain, and document. The chapter also distinguishes artificial intelligence from generative AI assistants and reviews risks associated with generated text and code. The final sections address managerial interpretation, industry practice, responsible analytics, and applied exercises.

KEYWORDS

marketing analytics; business analytics; artificial intelligence; generative artificial intelligence; data-informed decision-making; descriptive analytics; diagnostic analytics; predictive analytics; prescriptive analytics; verification

VERSION AND DATE

Version 1.0 · July 2026 · Language: English (United States)

SUGGESTED CITATION

Mendoza, J. (2026). Becoming an AI-augmented marketing analyst. In Applied business analytics for marketing decision-making: Business analytics and data visualization (Chapter 1, Version 1.0) [Open educational resource]. CC BY 4.0.

LICENSE AND RIGHTS

Copyright © 2026 Jose Mendoza. Except where otherwise noted, this work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). You may share and adapt this material for any purpose, provided appropriate credit is given. Third-party trademarks, screenshots, figures, and other materials remain subject to their respective rights and licenses.

Google Colab is a product of Google LLC. “Python” and the Python logos are trademarks or registered trademarks of the Python Software Foundation. pandas is a sponsored project of NumFOCUS, a 501(c)(3) nonprofit charity in the United States. Product names are used for identification only and do not imply endorsement. StyleCraft Collective is a fictional company created for instruction.

COMPANION REPOSITORY

Datasets, notebooks, and figure sources for this chapter: https://github.com/jrmst102/businessanalytics


GENERATIVE AI USE

Generative artificial intelligence and other AI-assisted tools were used in the research, writing, revision, and production of this chapter, including literature discovery, source organization, outlining, preliminary drafts, prose revision, support for code and analytical examples, and document formatting. These tools were used under the author’s direction and are not credited as authors, researchers, or sources. The author determined the chapter’s scope, learning objectives, methods, interpretations, and recommendations, and reviewed and approved all AI-assisted material: factual claims and citations were checked against the underlying sources rather than accepted from AI-generated summaries, and code and analytical outputs were tested or otherwise reviewed for accuracy. Responsibility for the accuracy, originality, and final form of this chapter rests entirely with the author. A fuller statement appears in the front matter of the complete guide.

Chapter Learning Objectives

  • By the end of this chapter, students should be able to:
  • Describe the role of the marketing analyst in data-rich and AI-assisted business environments.
  • Distinguish among descriptive, diagnostic, predictive, and prescriptive analytics in marketing contexts.
  • Explain how business questions, analytical questions, data, evidence, and recommendations are connected.
  • Distinguish artificial intelligence broadly from generative AI assistants and identify where each may support marketing analytics and where human judgment remains necessary.
  • Translate analytic output into a cautious, evidence-based marketing recommendation.

CONCEPT

What This Chapter Is Really About

This chapter is not mainly about learning code. It is about learning the professional discipline of an analyst. Code, charts, models, dashboards, and AI tools are useful only when they help a decision-maker understand a problem and act with better evidence.

The guiding question is therefore: How does a marketing professional become a better decision-maker in a world where data and AI tools are widely available?

1.1 Marketing Decision Context: The Holiday Budget Decision

Imagine that you have just joined the marketing analytics team at StyleCraft Collective, a fictional mid-sized apparel retailer that sells through its website, mobile app, email program, and selected pop-up events. The company has a loyal customer base, but sales growth has slowed. The chief marketing officer wants the team to improve campaign performance before the holiday season, and she has asked for a recommendation in three weeks, before media commitments are finalized. Whatever the team recommends will determine where the holiday budget goes and which customers hear from the brand first.

At the weekly marketing performance review, each team brings a different view of the business. The email manager reports that open rates improved after a subject-line test. The paid media specialist argues that search ads generated the highest number of conversions. The social media manager points to engagement on short-form videos. The CRM manager worries that repeat purchase rates have declined among customers who bought only once. The finance team asks whether the marketing budget should be shifted away from low-margin promotions.

Everyone has data. Not everyone has the same question. Some participants want to describe what happened. Others want to understand why performance changed. Others want to predict what is likely to happen next. Still others want to decide what action the company should take. The review stalls because reports, opinions, metrics, and recommendations are being mixed before the decision has been clarified.

Notice what is missing. No one has stated the decision. Without a stated decision, every metric looks relevant and none is decisive. The improvement in open rates is real, but it does not tell the chief marketing officer whether to move money out of paid search. The finance team’s margin concern is legitimate, but it does not identify which promotions destroy margin. Analytics cannot resolve the disagreement until someone names the choice that has to be made and the date by which it must be made.

There is also a constraint that will shape everything the team can do. Three weeks is enough time to describe what happened, compare a few segments, and assemble a defensible recommendation. It is not enough time to run a clean experiment, rebuild the customer data model, or wait for a quarter of new results. Part of the analyst’s job is to say plainly what can and cannot be answered inside the time available, and then to propose the strongest evidence that fits. A recommendation that arrives after the media commitments are signed has no value; however rigorous it may be.

This situation is common. Marketing teams often have dashboards, campaign reports, CRM exports, web analytics, social media metrics, and customer feedback. However, the availability of data does not automatically produce better decisions. The value of analytics depends on how well the team frames the problem, prepares the data, selects the method, verifies the output, and communicates the recommendation.

Over the next three weeks, the team will likely use AI tools to compare channels and draft code for promotion analysis. Those uses are reasonable. The issue is whether the results are checked and whether the person signing the recommendation can explain how each number was produced. Accountability for the evidence remains with the analyst throughout the guide.

StyleCraft returns throughout the book as a running case. The company does not need analytics because it lacks data. It needs analytics because it must decide what to do.

CONCEPT

Decision Context

In this guide, the decision context refers to the managerial situation in which an analysis will be used. It includes the decision to be made, the stakeholders, the available evidence, the constraints, the timing of the decision, and the consequences of action. This is an operational course concept rather than a standalone academic construct.

1.2 What Does a Marketing Analyst Do?

A marketing analyst helps organizations make better decisions about customers, markets, campaigns, products, pricing, channels, and communication. This work may involve data preparation, descriptive summaries, visualization, predictive modeling, experiments, dashboards, and written recommendations. However, the analyst is not simply a person who produces charts or runs code. The analyst is responsible for disciplined reasoning.

Marketing analytics is especially challenging because marketing decisions are often made under uncertainty. Customers change their preferences. Competitors respond. Channels shift. Advertising platforms change their algorithms. Macroeconomic conditions affect spending. Metrics may be incomplete, biased, or difficult to compare. Even when a dataset is accurate, it may not fully represent the decision problem. As a result, marketing analysts must combine quantitative evidence with business context.

Furthermore, the modern analyst often works with AI tools. These tools can draft code, explain an error message, summarize customer comments, suggest visualizations, or create a preliminary managerial summary. These uses can save time. They do not remove the need for judgment. As Davenport, Guha, Grewal, and Bressgott (2020) argue, AI is likely to be most valuable in marketing when it augments, rather than replaces, human managers.

An AI-augmented analyst does not outsource thinking to software. The analyst defines the question, selects the evidence, checks the work, interprets the result, and takes responsibility for the recommendation.

1.2.1 The Analyst as Translator

One useful way to understand the marketing analyst is as a translator among three languages. The first is the language of business: revenue, customers, margins, budgets, campaigns, channels, and growth. The second is the language of data: rows, variables, distributions, missing values, relationships, models, and uncertainty. The third is the language of decision-making: alternatives, trade-offs, risks, recommendations, and next steps.

Many analyses fail because these languages are not connected. A technically correct model may not answer the manager’s question. A polished dashboard may not clarify what action should be taken. A written recommendation may go beyond what the evidence supports. The analyst’s task is to keep these languages aligned.

CONCEPT

Analytic Translation

Analytic translation is the process of converting a business problem into an analytical task, interpreting the analytical result in business language, and turning the evidence into a recommendation that a decision-maker can use.

In other words, the analyst does not merely produce output; the analyst connects output to action.

1.3 Business Analytics, Marketing Analytics, and Data-Informed Decisions

The following definitions connect the analyst’s role to marketing decision support.

DEFINITION

Business Analytics

Business analytics is the use of data, quantitative methods, models, visualization, and technology to understand business problems and support decisions. In this guide, the phrase emphasizes decision support rather than technique for its own sake.

Source: Adapted from Camm et al. (2023), Davenport (2006), and Provost and Fawcett (2013).

DEFINITION

Marketing Analytics

Marketing analytics is the application of analytic methods, data, models, and technologies to marketing decisions. These decisions may involve customer understanding, campaign evaluation, segmentation, targeting, personalization, pricing, retention, forecasting, and marketing communication.

Source: Adapted from Wedel and Kannan (2016) and France and Ghose (2019).

The distinction matters because this guide is not a general statistics or programming text. It approaches analytics from the perspective of marketing decision-making. The central question is not simply whether an analysis can be performed, but whether the analysis helps a marketer understand a business problem and choose a more defensible course of action.

CONCEPT

Data, Information, Insight, and Recommendation

Data are recorded observations, values, or symbols. Information is data organized so that it becomes meaningful for a purpose. An insight is an interpretation that explains why a pattern matters. A recommendation is a proposed action supported by evidence, judgment, and context.

For example, a file containing 50,000 email recipients and clicks is data. A table showing that the click-through rate was 4.2 percent is information. The finding that recent purchasers clicked at a higher rate is an insight. The proposal to prioritize recent purchasers in the next email sequence is a recommendation.

Source: Adapted from Ackoff (1989), Rowley (2007), Farris et al. (2015), and Provost and Fawcett (2013).

1.3.1 Why Data-Informed, Not Data-Driven?

This guide uses data-informed because metrics and models do not determine actions on their own. Marketing decisions also require judgment about brand positioning, customer trust, ethics, competitive response, resource limits, and uncertainty.

For example, if a model predicts that a steep discount will increase conversions, a manager still needs to consider margin, customer expectations, brand equity, and whether frequent discounting trains customers to wait for promotions. Analytics can improve a decision without transferring responsibility for it.

CONCEPT

Data-Informed Decision-Making

Data-informed decision-making uses data and analysis as essential inputs into managerial judgment while recognizing that context, ethics, uncertainty, and strategic priorities also matter. This wording is used as an operational concept for this guide.

Source: Adapted from Provost and Fawcett (2013) and Davenport (2006).`

1.4 The Four Types of Analytics

Introductory business analytics is often organized into descriptive, predictive, and prescriptive categories. This chapter adds diagnostic analytics because marketers often need to investigate why performance changed before recommending an action.

The four categories are related, but they are not interchangeable. A dashboard that reports last month’s sales is not a forecast. A forecast is not a recommendation. A recommendation should use evidence, but it must also consider the decision context and the available alternatives. Table 1.1 summarizes the four types and their managerial value.

Table 1.1
Four Types of Analytics in Marketing

Type of analytics

Main question

Common outputs

Marketing example

Managerial value

Descriptive analytics

What happened?

Reports, dashboards, summary statistics, charts

Campaign conversion rate by channel last month

Clarifies performance

Diagnostic analytics

Why might it have happened?

Comparisons, drill-downs, cohort analysis, segment analysis

Lower conversion among first-time visitors than returning customers

Identifies possible drivers

Predictive analytics

What is likely to happen?

Forecasts, scores, probabilities, predictive models

Probability that a customer will churn in the next 30 days

Anticipates outcomes

Prescriptive analytics

What should we do?

Decision rules, optimization, recommendations, action plans

Allocate budget to campaigns with the highest expected incremental contribution

Supports action

DEFINITION

Descriptive Analytics

Descriptive analytics summarizes past or current data to show what happened. In marketing, it may include sales reports, campaign dashboards, customer profiles, channel comparisons, and summary statistics.

Source: Adapted from Camm et al. (2023) and Farris et al. (2015).

CONCEPT

Diagnostic Analytics

Diagnostic analytics investigates why performance may have changed by comparing groups, segments, time periods, or possible drivers. In marketing, it may involve cohort analysis, segment comparison, or channel drill-downs.

Caution: diagnostic analytics can identify plausible drivers and narrow the investigation, but observational comparisons do not by themselves establish causality.

Source: Course concept developed for this guide. The comparative methods it relies on are discussed in Camm et al. (2023) and France and Ghose (2019).

That caution deserves emphasis, because diagnostic language invites causal claims. A drill-down showing lower conversion among first-time visitors has identified an association, not a cause. Those visitors may differ in intent, in traffic source, in device, or in what they were shopping for. Chapter 7 examines how regression describes relationships without demonstrating cause, and Chapter 11 shows what kind of evidence a causal claim requires. Until then, treat every diagnostic finding as a candidate explanation to be tested rather than a conclusion to be reported.

DEFINITION

Predictive Analytics

Predictive analytics uses patterns in existing data to estimate what is likely to happen for new or future observations. In marketing, it may include churn prediction, sales forecasting, lead scoring, response modeling, or conversion probability estimation.

Predicting an outcome well and explaining what causes it are different goals, and a model that does one may do the other poorly.

Source: Adapted from Shmueli (2010) and Provost and Fawcett (2013).

DEFINITION

Prescriptive Analytics

Prescriptive analytics uses evidence, models, constraints, and judgment to recommend actions. In marketing, it may support budget allocation, targeting, offer selection, pricing, promotion planning, or resource prioritization.

Source: Adapted from Camm et al. (2023) and Davenport (2006).

1.4.1 Example: Email Campaign Performance

Suppose StyleCraft sent three campaigns: a new-arrivals campaign, a loyalty discount campaign, and a clearance campaign. The descriptive question is: Which campaign had the highest open rate, click-through rate, conversion rate, and revenue? The diagnostic question is perhaps the most interesting: Why did the loyalty campaign perform better among returning customers but worse among new customers? The predictive question is: Which customers are most likely to respond to the next email? The prescriptive question is: Which customers should receive which offer next week, given budget limits and margin goals?

The same dataset may support all four types of analysis, but each type asks a different question. One of the analyst’s first responsibilities is to identify which question is being asked.

1.5 From Business Question to Analytical Question

Analytics begins with the business question, not the dataset. A business question is the managerial issue that the organization needs to address; an analytical question is the measurable version of that issue. The business question states what must be decided; the analytical question states what must be measured.

Consider the question, “How can we improve customer loyalty?” This is a reasonable business concern, but it is too broad for immediate analysis. An analyst might translate it into more specific analytical questions: Which customer segments have the highest repeat purchase rates? Which first-purchase categories are associated with a second purchase? How long does it typically take a new customer to make a second purchase? Which campaign contacts are associated with retention? Each analytical question points to different data and methods. Table 1.2 illustrates this translation for several common marketing questions.

Table 1.2
Translating Business Questions into Analytical Questions

Business question

Possible analytical question

Potential data needed

Possible decision supported

Should we increase paid search spending?

Which paid search campaigns generate the highest conversion rate and contribution margin?

Campaign costs, clicks, conversions, order value, margin

Budget allocation

Why are repeat purchases declining?

Which customer cohorts show the largest decline in second-purchase rate?

Customer IDs, purchase dates, order sequence, product categories

Retention campaign design

Which customers should receive a discount?

Which customers have high predicted response but low likelihood of buying without an offer?

Purchase history, engagement, prior discounts, margins

Offer targeting

Is the loyalty program working?

Did members show higher repeat purchase or average order value after enrollment than comparable nonmembers?

Enrollment date, purchases, customer attributes, time periods

Program continuation or redesign

What should the dashboard emphasize?

Which metrics best indicate whether campaign performance is improving against business goals?

KPIs, targets, historical performance, stakeholder needs

Executive reporting design

CONCEPT

Business Questions and Analytical Questions

A business question identifies the decision or managerial issue. An analytical question identifies what must be measured, compared, modeled, or visualized to support that decision. Strong analytics requires both.

Source: Adapted from Provost and Fawcett (2013) and Chapman et al. (2000).

1.6 The Marketing Analytics Workflow

CRISP-DM, a widely cited data mining process, begins with business understanding, then moves through data understanding, data preparation, modeling, evaluation, and deployment (Chapman et al., 2000). This guide adapts that logic for marketing students and AI-assisted practice.

The adapted workflow has seven steps: define the marketing decision, translate the decision into analytical questions, identify and prepare the data, explore patterns, apply an appropriate method, verify and interpret the output, and communicate a recommendation. Table 1.3 presents the seven steps and the analyst’s work at each one. Table 1.4 then shows, for the same seven steps, where AI assistance typically helps and what must be verified before the step is trusted.

Table 1.3
A Marketing Analytics Workflow: Steps and Analyst Work

Step

Question to ask

Typical analyst activity

1. Define the decision

What decision needs support?

Clarify decision, audience, timing, constraints

2. Translate the question

What can be measured or modeled?

Define unit of analysis, variables, outcome

3. Prepare the data

Is the data usable?

Clean, join, filter, reshape, document

4. Explore patterns

What does the data suggest?

Summaries, charts, segments, comparisons

5. Apply a method

What method fits the question?

Regression, classification, forecasting, clustering, A/B test, dashboard

6. Interpret evidence

What does the output mean?

Translate output into business language

7. Recommend action

What should be done next?

Write a clear, qualified recommendation

Table 1.4
A Marketing Analytics Workflow: AI Support and Verification Checks

Step

Where AI may help

Verification check

1. Define the decision

Draft a problem statement

Confirm the decision and deadline with the stakeholder

2. Translate the question

Suggest candidate analytical questions

Check that each question maps to a decision

3. Prepare the data

Generate starter cleaning code

Check rows, columns, missing values, duplicates

4. Explore patterns

Suggest charts or summary code

Check scales, labels, totals

5. Apply a method

Draft model or chart code

Compare the result with a baseline or a stated expectation

6. Interpret evidence

Draft a plain-language explanation

Check whether the claims follow from the evidence

7. Recommend action

Draft a summary or memo

Identify risks, limits, and the next measurement

CONCEPT

Verification

Verification is the practice of checking whether analytic output is accurate, reasonable, and aligned with the analytical question before using it in a recommendation. Verification is especially important when code, charts, or summaries are produced with AI support. In this guide, verification is treated as a professional habit rather than as a single statistical test.

Source: Adapted from Chapman et al. (2000) and Provost and Fawcett (2013).

1.7 The AI-Assisted Workflow Used in This Guide

Within the seven-step workflow, AI use follows four habits: specify, predict-then-verify, explain, and document. Together, they make AI-assisted work accountable.

First, specify the task. Before asking for help, define the business question, what one row represents, the relevant variables, the intended output, the constraints, and the verification criteria. A vague prompt such as “analyze this data” is unlikely to produce a useful result. A stronger prompt tells the tool what the data represent and what decision the analysis should support.

Second, predict-then-verify. Before running code or accepting an interpretation, write down what you expect the result to be: a number, range, shape, or direction. Then compare the actual output with that expectation. If they do not match, stop and investigate. Either the output is wrong, your expectation is wrong, or the data contain something that requires explanation.

Third, explain the work. The analyst should be able to explain what each major step does and what the output means. Fourth, document the use of AI. Documentation should record the tool used, the prompt, the output, the revisions made, the errors or limitations found, and the verification performed.

CONCEPT

Specify, Predict-Then-Verify, Explain, Document

Specify: define the task before asking for help.

Predict-then-verify, write your expectation, run the step, compare the output with the expectation, and investigate any mismatch.

Explain: be able to explain what each step does and what the result means.

Document: record how AI was used and how the work was checked.

Source: Author-developed framework for this guide, informed by the analytic process logic in Chapman et al. (2000) and Provost and Fawcett (2013) and by the AI-augmentation argument in Davenport et al. (2020).

1.7.1 Example: A Weak Prompt and a Stronger Prompt

The difference between the two prompts below is the specify step. Neither prompt requires special technique; the second simply says what the data are and what the analysis is for.

A weak prompt. “Analyze this campaign data and tell me what to do.”

A stronger prompt. “I am analyzing a marketing campaign dataset with one row per campaign. The columns are campaign_name, channel, impressions, clicks, conversions, revenue, and ad_spend, where ad_spend is advertising spend only. Please write pandas code to calculate click-through rate, post-click conversion rate (conversions divided by clicks), cost per conversion, and return on ad spend by campaign. Do not interpret the results yet. Include checks for missing values and for zero denominators.”

The second prompt is better because it specifies what one row represents, the variables, the calculations, the denominator for each rate, the software context, and the limits on what the assistant should do. It also asks for verification before interpretation.

1.8 Key Marketing Data Concepts

Before the first lab, students need a small set of data concepts. Chapter 3 develops them formally; here they are introduced only as working ideas.

Begin with one question: what does one row represent? In the lab that follows, each row is one marketing campaign. In a customer file, each row is one customer. In a transaction file, each row is one purchase. The answer matters because it determines what a total, an average, or a rate can legitimately describe. Unless those quantities have been aggregated into explicit columns, a campaign-level table does not reveal individual customers, and a customer-level table does not reveal individual transactions. A surprising share of analytic errors are simply one kind of row being interpreted as if it were another.

Marketing data can come from CRM systems, point-of-sale files, websites, email platforms, social media platforms, advertising accounts, surveys, loyalty programs, customer service interactions, experiments, and third-party data providers. Each source has strengths and limitations.

Because marketing data are imperfect, analysts should not treat datasets as neutral mirrors of reality. A dataset is a representation of business activity shaped by what the organization measured, what systems captured, what customers consented to share, and what processing occurred before the analyst received the file. Whether a particular dataset is fit for a particular purpose is a question this guide returns to repeatedly.

Chapter 3 develops these concepts formally. It defines the unit of analysis and the grain of a table, separates stored data types from levels of measurement, specifies the four elements of a ratio metric, and sets out the dimensions along which data quality is assessed. For now, one habit is enough: before analyzing any file, state what one row represents, and write it down where the reader of your analysis can see it.

1.9 AI in Marketing Analytics: Benefits and Risks

Davenport et al. (2020) and Huang and Rust (2021) describe broad applications of AI across marketing research, strategy, and action. This course relies mainly on generative AI assistants, which have a different risk profile from predictive and algorithmic systems.

CONCEPT

Artificial Intelligence and Generative Artificial Intelligence

Artificial intelligence is the broad category: computational systems that perform tasks associated with human intelligence, including pattern recognition, prediction, classification, and language processing.

Generative artificial intelligence refers to systems that produce new content such as text, code, images, audio, or video, in response to a prompt.

In this course, most AI-assisted exercises use generative-AI assistants such as chatbots and notebook copilots. Later chapters also examine predictive and algorithmic systems, including the scoring and targeting models that marketing organizations run without generating any text at all.

Source: Adapted from Davenport et al. (2020), Huang and Rust (2021), and National Institute of Standards and Technology (2024).

For students, the most immediate use of AI may be as a workflow aid. A tool can explain why Python code failed, suggest a panda’s command, summarize the meaning of an output, or propose chart alternatives. These tools can speed up learning and routine work, but students still need to understand and verify the output.

The risks are practical, and most of the ones that will affect your coursework belong specifically to generative AI rather than to artificial intelligence in general. A generative assistant may invent sources, misread columns, assume the wrong unit of analysis, ignore missing values, or generate code that runs cleanly while computing the wrong metric. The federal risk framework for generative AI names this failure directly as confabulation, the generation and confident presentation of erroneous content in response to a prompt (National Institute of Standards and Technology, 2024). That framework also identifies data privacy, intellectual property, information integrity, and human–AI configuration as risk areas, all of which apply when confidential material is placed into an external system or when generated output is passed along unchecked. Responsible use therefore requires the analyst to supervise the assistant. Table 1.5 summarizes where AI support is useful across the workflow and what the analyst remains responsible for.

Table 1.5
AI Support and Risks Across the Marketing Analytics Workflow

Workflow step

Useful AI support

Risk to watch for

Analyst responsibility

Framing the problem

Suggest possible analytical questions

Questions may be generic or misaligned

Select questions based on the decision

Preparing data

Generate cleaning code or explain errors

Code may remove important records or mishandle missing values

Check row counts, logic, and assumptions

Exploring data

Suggest summaries or chart types

Charts may be decorative rather than useful

Match visuals to the question

Modeling

Draft starter code

Method may be inappropriate or poorly evaluated

Choose the method and compare baselines

Interpretation

Draft plain-language explanation

Interpretation may overclaim causality or certainty

Qualify conclusions

Recommendation

Draft a memo or summary

Recommendation may ignore constraints or ethics

Make the final judgment

CONCEPT

AI in Marketing Analytics

In this guide, AI in marketing analytics refers to the use of computational systems to automate tasks, process data, identify patterns, support predictions, personalize actions, or assist communication in marketing research, strategy, and execution.

Source: Adapted from Davenport et al. (2020) and Huang and Rust (2021).

AI IN PRACTICE

Review Before You Rely

A useful professional standard is simple: do not send work to a manager that you cannot explain. AI can help draft code, summarize output, or improve a first version of a memo. However, the analyst must still check the work, revise it, explain it, and stand behind the recommendation.

Audit routine. After an assistant returns code or an interpretation, ask four questions in order. Does the output answer the question I specified? Does one row still represent what I said it represents? Can I reproduce at least one number by hand? Can I state in one sentence what this result does not show?

1.10 Marketing Interpretation and Managerial Insight

Campaign metrics are not recommendations. Interpretation begins by asking what each metric suggests, which metric matters for the decision, and what trade-offs or limitations remain. Does a campaign with high revenue also have high spend? Does a high post-click conversion rate translate into high contribution?

Suppose the Loyalty Offer has the highest post-click conversion rate and the highest ROAS. A weak interpretation would say, “The Loyalty Offer is best.” A stronger interpretation would say, “Among these four campaigns, the Loyalty Offer produced the strongest combination of conversion efficiency and return on ad spend. However, the data do not show whether customers would have purchased without the offer, whether the discount reduced margin, or whether the campaign can scale to a larger audience.”

The stronger interpretation is better because it connects the result to a decision while acknowledging uncertainty. It does not confuse performance measurement with causal proof. It does not assume that the best campaign in a small dataset should automatically receive the full budget. It identifies what the evidence supports and what remains unknown.

1.10.1 Revisiting the StyleCraft Case

The StyleCraft team should begin by clarifying the holiday-season decision. If the decision is budget allocation, the team needs campaign spend, revenue, margin, incremental lift, audience size, and operational constraints. If the decision is customer targeting, the team needs customer-level response and purchase history. If the decision is message strategy, the team may need creative performance, segment behavior, and customer feedback.

Descriptive analytics could summarize past campaign performance. Diagnostic analytics could compare performance across channels, segments, customer cohorts, or offer types. Predictive analytics could estimate which customers are most likely to respond to future campaigns. Prescriptive analytics could recommend how to allocate budget, or which offers to send to which customers. AI tools could help with code, summaries, and drafting. The team must still verify the results, evaluate the assumptions, and make the final recommendation.

1.11 Business Analytics in Practice

This section examines how the chapter’s habits appear in real organizations: how AI changes analysts’ work, what happens when a deliverable is not verified, and what employers test in analytics interviews.

1.11.1 When the Coding Gets Easier, the Framing Gets Harder

Consider a consumer-packaged-goods insights team of eight analysts supporting brand and shopper marketing. Two years ago, a routine request, weekly promotion performance by retailer and pack size, took an analyst most of a day: pull the data, reshape it, build the summary, format the deck. Today an assistant drafts the query and the chart code in minutes. The time saved on coding shifted to other decisions: which retailers belong in the comparison, whether the promotion window should be defined by ship date or scan date, and whether the lift being reported is lift against last year, against the non-promoted weeks, or against nothing at all.

That redistribution of effort is consistent with what field experiments have found. In a preregistered study of 758 consultants at Boston Consulting Group, participants with access to GPT-4 completed 12.2 percent more tasks, completed them 25.1 percent more quickly, and produced work of measurably higher quality. On a task deliberately designed to fall outside the model’s competence, however, participants using AI were 19 percentage points less likely to produce correct solutions than participants working without it (Dell’Acqua et al., 2026). The authors describe a jagged frontier: capability that is uneven in ways not visible from the outside. Because incorrect output can be fluent and polished, teams cannot judge reliability from presentation quality.

Teams that adapt well tend to do two things. They shift senior time toward problem framing, because framing decisions are the ones the assistant cannot make. And they make quality assurance an explicit, named step rather than something that happens if there is time. On the team described here, no recurring report leaves the group until a second analyst has confirmed the denominator, the date window, and the filter set. A review of that kind costs minutes and catches the errors a single analyst working quickly does not. As production becomes cheaper, specifying the question and checking the answer become the scarce skills.

1.11.2 The Deliverable That Could Not Be Verified

The failure mode is easiest to see in professional services, where the deliverable is the product. In October 2025, Deloitte Australia agreed to issue a partial refund on a report it had produced for the Australian government under a contract valued at approximately A$440,000, after the published version was found to contain references to academic papers that do not exist and a fabricated quotation attributed to a federal court judgment (Associated Press, 2025). The firm maintained that the substance of the review held. The citations, however, had been invented, and no one had checked them before publication.

Marketing analytics fails the same way, usually with less publicity. An agency deliverable arrives carrying a confident summary metric, say “incremental reach efficiency,” that no one on the client side can reproduce, because the assistant that drafted the deck synthesized a plausible-sounding measure rather than computing a defined one. The client cannot verify it. The agency analyst cannot fully explain it. The metric nonetheless enters the next planning cycle, and by the time someone asks where the number came from, three decisions have been built on top of it.

The federal risk framework for generative AI gives this behavior a name: confabulation, the confident presentation of erroneous content (National Institute of Standards and Technology, 2024). The naming matters, because saying that the tool made a mistake locates the problem in the software. In the cases above, the software did what generative systems do. What was missing was a person who checked. An unverifiable number is a liability owned by the person who signs the deliverable.

1.11.3 What Hiring Managers Can Test For

Hiring has moved in the same direction. Employers identify analytical thinking as the most required core skill, cited by seven in ten organizations, while AI and big data rank as the fastest-growing skill through 2030 (World Economic Forum, 2025). Employers therefore need analysts who can use AI tools and judge their output.

The verification habit is now common enough to be measured. In a survey of more than 33,000 respondents, 84 percent reported using or planning to use AI tools, yet only about 3 percent expressed high trust in the accuracy of what those tools produced. Sixty-six percent named “AI solutions that are almost right, but not quite” as their leading frustration, and roughly 45 percent reported that debugging AI-generated code took more time than expected (Stack Overflow, 2025). Analysts are not developers, but the pattern transfers directly: heavy use, low trust, and real time spent on the gap between the two.

That gap is what an interview can usefully probe. Instead of asking a candidate to write code from scratch, an interviewer might hand over a short notebook or a one-page analysis and ask what is wrong with it: a reversed denominator, a silently dropped set of rows, a rate above 1, a chart whose axis begins at 40 rather than 0. A candidate who can describe how they check work, and who can name an occasion when an assistant’s output looked right and was not, is demonstrating exactly the judgment the survey evidence above suggests is scarce. Verification is therefore part of the analyst’s work, not a final defensive step.

1.11.4 In Your First Analyst Job

The course mirrors the accountability structure you will work inside: labs require a prediction before execution, AI-assisted exercises require documentation of the tool, the prompt, the revision, and the check, and recommendation exercises require one limitation and one next step. Work in this course is assessed on whether it can be defended, not on whether the code ran. That is the same standard that will apply when the deliverable carries your name and the person reading it has a budget to spend.

1.12 Ethics, Privacy, and Responsible Analytics

Marketing analytics often uses customer data, creating legal, professional, and strategic responsibilities. Analysts should consider whether the data are appropriate to use, whether customers would reasonably expect the use, whether sensitive information is involved, whether the analysis could create unfair treatment, and whether the recommendation protects customer trust.

AI also creates specific privacy and governance risks. Uploading confidential customer data into an external AI system may violate company policy, privacy rules, or customer expectations, and data privacy is one of the risk areas that generative-AI governance frameworks specifically flag (National Institute of Standards and Technology, 2024). In this course, confidential or personally identifying data are never to be placed into an external AI tool. Model recommendations may also reproduce patterns of exclusion if historical data reflect unequal treatment. A responsible analyst asks not only, “Does the model work?” but also, “Should we use this result in this way?”

CONCEPT

Responsible Analytics

Responsible analytics is the practice of using data, methods, AI tools, and recommendations in ways that are accurate, transparent, privacy-aware, fair, and accountable to the decision context.

Source: Author-developed definition for this guide, synthesizing the marketing-privacy, data-ethics, and algorithmic-accountability literatures represented by Martin and Murphy (2017), Barocas and Selbst (2016), Mittelstadt et al. (2016), and Floridi and Taddeo (2016).

1.13 Chapter Summary

This chapter introduced the role of the AI-augmented marketing analyst. The main point is that analytics is a disciplined approach to decision support. It begins with a marketing decision, translates that decision into analytical questions, uses data and methods to generate evidence, verifies the output, and communicates a recommendation.

The chapter distinguished data, information, insight, and recommendation; introduced the four types of analytics and the caution that diagnostic comparisons do not establish cause; separated artificial intelligence broadly from the generative assistants’ students will use most; and presented the four habits of specify, predict-then-verify, explain, and document. It also stepped outside StyleCraft to look at how those habits appear in an insights team, in a professional-services failure, and in analytics hiring.

The rest of this guide builds from this foundation. Later chapters examine marketing data, measurement, data preparation, exploratory analysis, segmentation, regression, predictive modeling, forecasting, experiments, visualization, dashboards, and storytelling. The tools will become more sophisticated, but the principle remains the same: analytics becomes valuable when it improves decisions. The next chapter examines marketing analytics as decision support in greater depth, showing how a decision is structured, how a vague request becomes a written specification, and how the deliverable is matched to the choice being made. Chapter 3 then turns to data, measurement, and marketing variables, where the concepts this chapter treated informally are defined in full.

1.14 Exercises for Practice

These exercises practice four habits: classify the analytic question, state what one row represents, verify the output, and connect the result to a marketing decision.

1.14.1 Core Chapter PracticeExercise 1.1 Concept Check

Explain the difference among data, information, insight, and recommendation.

  • What is the difference between a business question and an analytical question?
  • Why might data-informed be a better phrase than data-driven for many marketing decisions?
  • A dataset has one row for each customer purchase. What does one row represent, and what can that table not tell you?
  • Why is verification especially important when using AI-assisted code or interpretation?
  • What is the difference between artificial intelligence and generative artificial intelligence, and why does the distinction matter for the risks discussed in Section 1.9?

Exercise 1.2 Classify the Analytics Type

For each scenario, identify whether the example is primarily descriptive, diagnostic, predictive, or prescriptive. Then explain your reasoning in one sentence.

  • A dashboard shows monthly revenue by marketing channel for the past twelve months.
  • A model estimates which customers have the highest probability of canceling a subscription next month.
  • An analysis compares conversion rates by device type and finds that mobile users abandon the checkout page more often.
  • A system recommends sending different retention offers to different customer segments based on expected profit.
  • A report shows that average order value increased after a new free-shipping threshold was introduced.
  • A forecast estimates expected weekly demand for the next quarter.
  • An analyst recommends reducing spend on a campaign because its cost per acquisition exceeds the target margin.
  • A chart shows that repeat purchase rates are lower among customers acquired through one promotion.

Exercise 1.3 Match the Metric to the Question

For each business question, explain why the suggested metric is useful, state the denominator the metric requires, and identify one thing the metric does not tell you.

  • Are people engaging with our email? Suggested metric: click-through rate.
  • Are clicks turning into purchases? Suggested metric: post-click conversion rate.
  • Are we spending efficiently? Suggested metric: cost per conversion.
  • Are we generating revenue relative to advertising spend? Suggested metric: ROAS.
  • Are we retaining customers? Suggested metric: repeat purchase rate.
  • Are customers becoming more valuable over time? Suggested metric: customer lifetime value.

Exercise 1.4 AI-Assisted Practice

Use an AI tool to support one part of Exercise 1.4, but do not allow the tool to replace your own interpretation.

  • Complete the compact documentation record shown in Table 1.6 in your notebook
  • State one thing the assistant produced that you revised or rejected and explain how you detected it.


Table 1.6
AI-Use Documentation: Compact Record

Question

Student response

Which AI tool did you use, and for which step?

What prompt did you submit?

What did you accept, revise, or reject?

How did you verify the result?

Table 1.6 is a compact record for use inside a notebook.

Exercise 1.5 Managerial Memo

Write a short memo of 200–300 words to the StyleCraft chief marketing officer. The memo should answer this question: What should the team examine before reallocating the holiday campaign budget? Include one descriptive question, one diagnostic question, one predictive question, one prescriptive question, and one caution about AI-assisted analysis. For the diagnostic question, state explicitly what it could and could not establish. Recommended memo structure: decision context, evidence needed, caution or limitation, and next step.

1.15 Glossary of Terms

This glossary includes only the terms this chapter introduces. Terms that other chapters own are used with a pointer rather than redefined here; the unit of analysis, data quality, and the formal definition of a metric belong to Chapter 3.

AI in marketing analytics. The use of computational systems to automate tasks, process data, identify patterns, support predictions, personalize actions, or assist communication in marketing research, strategy, and execution (Davenport et al., 2020; Huang & Rust, 2021).

Analytic translation. The process of converting a business problem into an analytical task, interpreting the analytical result in business language, and turning the evidence into a usable recommendation. Course concept for this guide, informed by Provost and Fawcett (2013).

Analytical question. A measurable question that can be addressed with data, comparison, modeling, visualization, or experimentation (Chapman et al., 2000; Provost & Fawcett, 2013).

Business analytics. The use of data, quantitative methods, models, visualization, and technology to understand business problems and support decisions (Camm et al., 2023; Davenport, 2006).

Business question. The managerial issue or decision that an organization needs to address (Provost & Fawcett, 2013; Chapman et al., 2000).

Confabulation. The generation and confident presentation of erroneous content by a generative AI system in response to a prompt (National Institute of Standards and Technology, 2024).

Data-informed decision-making. The use of data and analysis as essential inputs into managerial judgment while recognizing context, ethics, uncertainty, and strategy (Davenport, 2006; Provost & Fawcett, 2013).

Decision context. The managerial situation in which an analysis will be used, including the decision, the stakeholders, the evidence, the constraints, the timing, and the consequences of action. Course concept for this guide.

Descriptive analytics. Analytics that summarize past or current data to show what happened (Camm et al., 2023; Farris et al., 2015).

Diagnostic analytics. Analytics that investigate why performance may have changed by comparing groups, segments, time periods, or possible drivers, without thereby establishing causality. Course concept for this guide; the comparative methods it relies on are discussed in Camm et al. (2023) and France and Ghose (2019).

Generative artificial intelligence. Systems that produce new content, including text, code, images, audio, or video, in response to a prompt (National Institute of Standards and Technology, 2024).

Input, transformation, and output. The three parts of any analytic step: the data or object used, the calculation or change applied, and the result produced. Course concept for this guide.

Marketing analytics. The application of analytic methods, data, models, and technologies to marketing decisions (Wedel & Kannan, 2016; France & Ghose, 2019).

Metric construction. Every metric is constructed through a definition. That definition specifies what is counted or calculated, the relevant time, and the records included. For a ratio metric, it also specifies the numerator and denominator.

Predictive analytics. Analytics that use patterns in existing data to estimate what is likely to happen for new or future observations (Shmueli, 2010; Provost & Fawcett, 2013).

Prescriptive analytics. Analytics that use evidence, models, constraints, and judgment to recommend actions (Camm et al., 2023; Davenport, 2006).

Responsible analytics. The use of data, methods, AI tools, and recommendations in ways that are accurate, transparent, privacy-aware, fair, and accountable to the decision context (Martin & Murphy, 2017; Barocas & Selbst, 2016; Mittelstadt et al., 2016; Floridi & Taddeo, 2016).

Specify, predict-then-verify, explain, document. The four AI-use habits applied throughout this guide: define the task before asking for help; state an expectation and compare it with the output; be able to explain each step; and record how AI was used and how the work was checked. Author-developed framework for this guide.

Verification. The practice of checking whether analytic output is accurate, reasonable, and aligned with the analytical question before using it in a recommendation (Chapman et al., 2000; Provost & Fawcett, 2013).

1.16 References

Ackoff, R. L. (1989). From data to wisdom. Journal of Applied Systems Analysis, 16, 3–9.

Associated Press. (2025, October 7). Deloitte to partially refund Australian government for report with apparent AI-generated errors. CTV News. https://www.ctvnews.ca/business/article/deloitte-to-partially-refund-australian-government-for-report-with-apparent-ai-generated-errors/

Barocas, S., & Selbst, A. D. (2016). Big data’s disparate impact. California Law Review, 104(3), 671–732.

Camm, J. D., Cochran, J. J., Fry, M. J., & Ohlmann, J. W. (2023). Business analytics (5th ed.). Cengage.

Chapman, P., Clinton, J., Kerber, R., Khabaza, T., Reinartz, T., Shearer, C., & Wirth, R. (2000). CRISP-DM 1.0: Step-by-step data mining guide. CRISP-DM Consortium.

Davenport, T. H. (2006). Competing on analytics. Harvard Business Review, 84(1), 98–107.

Davenport, T. H., Guha, A., Grewal, D., & Bressgott, T. (2020). How artificial intelligence will change the future of marketing. Journal of the Academy of Marketing Science, 48, 24–42. https://doi.org/10.1007/s11747-019-00696-0

Dell’Acqua, F., McFowland, E., III, Mollick, E., Lifshitz, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science, 37(2), 403–423. https://doi.org/10.1287/orsc.2025.21838

Farris, P. W., Bendle, N. T., Pfeifer, P. E., & Reibstein, D. J. (2015). Marketing metrics: The manager’s guide to measuring marketing performance (3rd ed.). FT Press.

Floridi, L., & Taddeo, M. (2016). What is data ethics? Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374(2083), Article 20160360. https://doi.org/10.1098/rsta.2016.0360

France, S. L., & Ghose, S. (2019). Marketing analytics: Methods, practice, implementation, and links to other fields. Expert Systems with Applications, 119, 456–475. https://doi.org/10.1016/j.eswa.2018.11.002

Huang, M.-H., & Rust, R. T. (2021). A strategic framework for artificial intelligence in marketing. Journal of the Academy of Marketing Science, 49, 30–50. https://doi.org/10.1007/s11747-020-00749-9

Martin, K. D., & Murphy, P. E. (2017). The role of data privacy in marketing. Journal of the Academy of Marketing Science, 45, 135–155. https://doi.org/10.1007/s11747-016-0495-4

Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of algorithms: Mapping the debate. Big Data & Society, 3(2), 1–21. https://doi.org/10.1177/2053951716679679

National Institute of Standards and Technology. (2024). Artificial intelligence risk management framework: Generative artificial intelligence profile (NIST AI 600-1). U.S. Department of Commerce. https://doi.org/10.6028/NIST.AI.600-1

Provost, F., & Fawcett, T. (2013). Data science and its relationship to big data and data-driven decision making. Big Data, 1(1), 51–59. https://doi.org/10.1089/big.2013.1508

Rowley, J. (2007). The wisdom hierarchy: Representations of the DIKW hierarchy. Journal of Information Science, 33(2), 163–180. https://doi.org/10.1177/0165551506070706

Rust, R. T., Ambler, T., Carpenter, G. S., Kumar, V., & Srivastava, R. K. (2004). Measuring marketing productivity: Current knowledge and future directions. Journal of Marketing, 68(4), 76–89. https://doi.org/10.1509/jmkg.68.4.76.42721

Shmueli, G. (2010). To explain or to predict? Statistical Science, 25(3), 289–310. https://doi.org/10.1214/10-STS330

Stack Overflow. (2025). 2025 developer survey: AI. https://survey.stackoverflow.co/2025/ai

Wedel, M., & Kannan, P. K. (2016). Marketing analytics for data-rich environments. Journal of Marketing, 80(6), 97–121. https://doi.org/10.1509/jm.15.0413

World Economic Forum. (2025). The future of jobs report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/

Annotate

Powered by Manifold Scholarship. Learn more at
Opens in new tab or windowmanifoldapp.org