Business Problem Framing and Decision Processes

Introduction

Business analytics creates value only when analysis is connected to a decision. A technically sophisticated model can be commercially useless if it addresses the wrong question, uses an inappropriate outcome measure, ignores constraints, or produces results that managers cannot act upon. Business problem framing is therefore the foundation of data-driven decision making.

Problem framing translates a broad managerial concern—such as declining sales, high employee turnover, delayed deliveries, or weak customer retention—into a structured decision problem. The process identifies what must be decided, why the decision matters, who is affected, what evidence is needed, which alternatives are feasible, and how success will be judged.

This chapter develops a practical framework for framing business problems. It explains objectives, stakeholders, constraints, key performance indicators (KPIs), hypotheses, decision criteria, issue trees, SMART objectives, and the CRISP-DM lifecycle. It also examines common errors, ethical considerations, and the professional skills required to connect analytical work with organizational action.

Learning Objectives

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

1. Distinguish a managerial symptom from an underlying business problem.
2. Translate a business concern into objectives, decisions, constraints, and measurable KPIs.
3. Identify stakeholders and evaluate their roles, interests, influence, and risks.
4. Formulate testable hypotheses and construct issue trees.
5. Apply SMART objectives and the CRISP-DM framework.
6. Compare alternatives using explicit decision criteria.
7. Recognize ethical, organizational, and data-governance issues in problem framing.

1. From Symptoms to Decision Problems

A symptom is an observable condition that signals a possible problem. For example, “monthly sales fell by 12%” is a symptom. It does not explain why sales fell or identify what action should be taken. A business problem is a condition that requires a decision and has consequences for organizational objectives. A well-framed problem might be: “Which combination of pricing, promotion, and customer-service actions should the retailer implement during the next quarter to restore profitable sales while preserving customer retention?”

A useful distinction is:

• Symptom: What is being observed?
• Cause or driver: What may be producing the observation?
• Decision: What choice must be made?
• Objective: What outcome should the choice achieve?
• Evidence: What information will improve the choice?

Analytics should not begin with “What data do we have?” It should begin with “What decision are we trying to improve?” Data availability can influence the feasible analysis, but it should not define the business problem by itself.

Decision problems and research questions

A decision problem concerns action. A research or analytical question concerns knowledge needed to support that action. For example:

The distinction matters because prediction alone does not establish that an intervention will work. A churn model may identify customers likely to leave, but a retention decision also requires evidence about treatment effectiveness, cost, capacity, and possible adverse effects.

2. The Architecture of a Well-Framed Business Problem

A structured decision problem usually contains the following elements.

Decision owner and decision horizon

The decision owner is the person or group authorized to act. The decision horizon specifies when the decision will be made and how long its effects will be evaluated. A chief marketing officer deciding on a six-month campaign faces different information and constraints from a branch manager deciding on next week's staffing schedule.

Objective

An objective states the desired result. Objectives may concern revenue, profit, service quality, risk reduction, fairness, customer experience, sustainability, or organizational capability. Objectives should be specific enough to guide measurement but broad enough to represent the real purpose of the decision.

A useful financial relationship is:

Revenue and cost must be measured over the same period and using consistent accounting definitions. Increasing revenue is not necessarily desirable if the associated cost increases by more than the revenue gain. Similarly, an objective such as “reduce delivery time” must be balanced against cost, reliability, safety, and customer priorities.

Alternatives

A decision requires at least two feasible courses of action, including the possibility of maintaining the status quo. Examples include changing a price, redesigning a process, reallocating staff, introducing a recommendation system, or taking no action. An analysis that identifies a pattern but does not connect it to alternatives has limited decision value.

Constraints

Constraints limit feasible alternatives. They may include budget, staffing, regulation, technology, contractual obligations, time, data availability, operational capacity, or organizational policy. A mathematically attractive solution that violates a regulatory or operational constraint is not a valid recommendation.

Outcomes and KPIs

A key performance indicator is a defined measure used to monitor progress toward an objective. A KPI should have a clear numerator, denominator where relevant, population, time period, data source, and owner.

Examples include:

• Conversion rate = completed purchases ÷ eligible visits.
• On-time delivery rate = deliveries meeting the promised time ÷ completed deliveries.
• Employee turnover rate = defined departures during a period ÷ the chosen average employee population during that period.
• Customer retention rate = customers active at the end of a period who were also active at the beginning, subject to a clearly defined customer population.

These measures are not interchangeable. A conversion rate may rise because low-value customers purchase more often, while profit falls because discounts are excessive. A balanced KPI set may therefore include a primary outcome, financial guardrails, operational measures, and risk or fairness measures.

Decision criteria

Decision criteria specify how alternatives will be judged. Typical criteria include expected financial value, customer impact, implementation time, risk, strategic fit, fairness, scalability, and reversibility. Criteria should be agreed before results are presented where possible, reducing the risk of selecting criteria merely because they favor a preferred option.

When criteria conflict, organizations may use a weighted scoring model. If alternative j receives a standardized score s_ij on criterion i, and criterion weights are w_i, a simple weighted score is:

Here, weights should sum to 1, scores must be placed on a comparable scale, and the direction of preference must be consistent—for example, a higher score should always mean better performance. This model is useful for transparent comparison, but it does not eliminate judgment and may conceal uncertainty or threshold requirements. A legally mandatory condition should normally be treated as a constraint rather than merely given a low weight.

3. SMART Objectives

SMART is a practical test for improving objectives:

• Specific: identifies the outcome and population.
• Measurable: defines how progress will be assessed.
• Achievable: is feasible given resources and capabilities.
• Relevant: supports a meaningful organizational purpose.
• Time-bound: identifies a deadline or evaluation period.

“Improve customer service” is too vague. “Increase the proportion of priority support tickets resolved within 24 hours from 78% to 90% by the end of the third quarter, without increasing complaint recurrence” is more useful. It identifies the population, target, time frame, and an important guardrail.

SMART does not mean that every objective must have a single numerical target. Some objectives involve compliance, safety, inclusion, or learning and require qualitative as well as quantitative evidence. Moreover, an achievable target should be based on capacity and evidence rather than optimism.

4. Stakeholder Analysis

Stakeholders are individuals or groups affected by a decision, able to influence it, or responsible for implementing it. They may include executives, customers, employees, suppliers, regulators, shareholders, communities, data subjects, and technology teams.

Stakeholder analysis asks:

1. Who owns the decision?
2. Who supplies or controls relevant data?
3. Who will implement the recommendation?
4. Who bears the costs or risks?
5. Who benefits from the outcome?
6. Whose interests may be underrepresented?

A power–interest matrix can guide engagement. High-power, high-interest stakeholders require close management; high-power, low-interest stakeholders require concise assurance; low-power, high-interest stakeholders need meaningful consultation; and low-power, low-interest groups should still be monitored where their rights or welfare may be affected.

Stakeholder disagreement is normal. A finance team may emphasize cost, operations may emphasize feasibility, marketing may emphasize growth, and customers may emphasize experience and privacy. The analyst’s role is not to remove legitimate disagreement but to make assumptions, trade-offs, and evidence visible.

5. Hypotheses and the Issue Tree

A hypothesis is a provisional, testable explanation or expectation. A good business hypothesis links a possible driver to an outcome and indicates what evidence could support or challenge it. For example: “The increase in cart abandonment is primarily associated with unexpected delivery charges shown late in checkout.” This is more useful than “Customers dislike the website,” because it suggests measurable evidence and possible interventions.

Hypotheses must not be confused with facts. They should be tested against appropriate data, alternative explanations, and, where possible, experiments or quasi-experimental designs. Correlation alone may identify an association but not prove that changing the suspected driver will change the outcome.

Issue trees

An issue tree decomposes a broad question into smaller, logically related questions. A tree for declining online profit might include:

Figure: An issue tree decomposing a broad profit problem into analysable branches.

A good issue tree is mutually exclusive and collectively exhaustive as far as practical: branches should not substantially overlap, and important explanations should not be omitted. Perfect completeness is rarely possible, so the tree should be revised as evidence and stakeholder knowledge develop. The tree is a thinking tool, not proof of causality.

6. CRISP-DM and the Analytical Decision Process

The Cross-Industry Standard Process for Data Mining (CRISP-DM) provides a widely used lifecycle:

1. Business understanding: define objectives, context, decision, constraints, and success criteria.
2. Data understanding: identify sources, inspect structure, assess quality, and explore patterns.
3. Data preparation: clean, combine, transform, and document data.
4. Modeling: select and fit suitable analytical methods.
5. Evaluation: test whether results are valid, useful, ethical, and aligned with the business objective.
6. Deployment: communicate findings, implement the decision, monitor performance, and learn.

CRISP-DM is iterative rather than strictly linear. Data limitations may require the business question to be refined; evaluation may reveal that the selected KPI is misleading; deployment may expose changes in customer behavior. Returning to an earlier phase is a sign of disciplined learning, not failure.

A critical distinction is between model performance and decision performance. A classification model may have high accuracy but provide little value if the positive class is rare, intervention costs are high, or the organization cannot act on predictions. Evaluation should therefore include technical measures, financial consequences, operational feasibility, stakeholder acceptance, and unintended effects.

7. Worked Example: Framing a Delivery-Service Decision

Consider an illustrative regional retailer whose customer complaints about late deliveries have increased. Management initially asks the analytics team to “build a model to predict late deliveries.”

Step 1: Clarify the decision

The actual decision is whether to change carrier allocation, warehouse cut-off times, delivery promises, or staffing during the next eight weeks. The decision owner is the operations director. Relevant stakeholders include customers, warehouse staff, carriers, customer service, finance, and legal or compliance personnel.

Step 2: Define the objective

A suitable objective might be: “Increase on-time delivery from its current level to at least 95% during the next eight weeks while keeping incremental delivery cost below £0.40 per order and avoiding a reduction in order capacity.” The exact baseline and definitions must be confirmed before measurement.

If 8,000 deliveries are completed in a period and 7,440 meet the promised time, then:

The calculation assumes that each delivery is counted once, the denominator includes all eligible deliveries, and the promise-time definition is consistent across periods.

Step 3: Construct hypotheses

Potential hypotheses include:

• Late orders are concentrated in specific carrier–region combinations.
• Warehouse cut-off times create avoidable next-day delays.
• Weather and seasonal demand increase lateness.
• The promise shown to customers is too optimistic for particular routes.
• Some delays are caused by incomplete addresses rather than carrier performance.

These hypotheses imply different actions. Reassigning carriers may help the first hypothesis but not the fourth. Changing customer promises may improve the KPI without improving physical delivery time, so customer satisfaction and cancellation should also be monitored.

Step 4: Establish criteria and alternatives

Alternatives might include retaining the current process, reallocating routes, changing cut-off times, adding temporary staff, or revising delivery promises. Criteria include on-time rate, incremental cost per order, capacity, customer complaints, employee workload, and fairness across regions.

Step 5: Evaluate and deploy

The analyst should examine historical data, validate definitions, and consider a controlled pilot. For example, selected regions could receive a carrier-allocation change while comparable regions retain the current process, provided operational and ethical conditions permit. Results should be interpreted cautiously because regions may differ in demand, weather, or customer mix. After implementation, KPIs should be monitored and the decision revisited.

The central lesson is that a prediction model is only one possible analytical component. The business need is an operational decision supported by evidence, not a model as an end in itself.

8. Common Framing Errors

Starting with available data

Organizations sometimes define problems around a convenient dataset. This can produce precise answers to irrelevant questions and exclude important outcomes not captured in the data.

Confusing correlation with causation

A variable associated with poor performance may be a marker rather than a cause. Decisions based on it can be ineffective or discriminatory. Causal claims require stronger designs, such as randomized experiments, natural experiments, or carefully justified observational methods.

Using a single KPI

Optimizing one measure can produce gaming or harmful trade-offs. For example, reducing average call duration may increase repeat calls and reduce customer satisfaction. Primary KPIs should be paired with guardrails and diagnostic measures.

Ignoring the counterfactual

To assess an intervention, decision makers need to ask what would have happened without it. Before-and-after comparisons can be misleading when seasonality, economic conditions, or customer composition change. A comparison group or other credible evaluation design may be needed.

Failing to define terms

Words such as “customer,” “active,” “late,” “profit,” and “retained” often have multiple meanings. Definitions should be documented in a KPI dictionary, including calculation logic, source systems, refresh frequency, and data owner.

Treating the model as the decision

A score or forecast does not determine action automatically. Thresholds, capacity, costs, exceptions, human review, and appeal processes must be specified.

9. Ethics, Governance, and Responsible Framing

Problem framing determines whose interests count and what outcomes are optimized. Ethical analysis should therefore begin before data collection or model development.

Important questions include:

• Is the proposed use consistent with the purpose for which data were collected?
• Are individuals informed where required, and is the use legally justified?
• Could the decision disadvantage protected or vulnerable groups?
• Are proxy variables reproducing discrimination?
• Can affected people understand, challenge, or appeal the decision?
• Are data access, retention, security, and quality controlled?

Privacy, data protection, sector regulation, and organizational policies vary by jurisdiction and industry. Compliance is a minimum requirement, not a substitute for ethical judgment. A legally permissible objective may still be unfair or harmful. Fairness measures also require careful definition because different fairness criteria can conflict; the selected criterion should reflect the decision context and affected stakeholders.

Documentation supports governance. A problem statement, assumption log, data dictionary, stakeholder register, decision record, model card where relevant, and post-deployment monitoring plan make analytical work auditable and reproducible.

10. Contemporary Tools and Professional Practice

Modern organizations use cloud platforms, dashboards, process-mining systems, automated forecasting, and generative AI assistants to accelerate analysis. These technologies can help summarize evidence, identify patterns, and draft alternative problem statements. However, generated content may contain unsupported assumptions, expose confidential information, or reproduce bias. Human review remains essential, particularly for high-impact decisions.

Employable analysts combine technical competence with inquiry and communication. They can conduct a structured interview, challenge ambiguous requests respectfully, write a one-page decision brief, define a KPI, explain uncertainty, and present an actionable recommendation. A strong decision brief normally includes:

1. Decision required and decision owner.
2. Business context and evidence of the problem.
3. Objective, scope, and time horizon.
4. Stakeholders and constraints.
5. Alternatives and decision criteria.
6. Data, assumptions, and limitations.
7. Recommended analysis or pilot.
8. Success measures, risks, and review date.

Chapter Summary

Business analytics begins with problem framing rather than data manipulation. A managerial symptom must be translated into a decision with a clear owner, objective, alternatives, constraints, stakeholders, KPIs, and decision criteria. SMART objectives improve specificity and accountability. Hypotheses express testable explanations, while issue trees decompose broad questions into structured analytical branches. CRISP-DM connects business understanding, data work, modeling, evaluation, and deployment in an iterative lifecycle. Effective analysts distinguish prediction from action, correlation from causation, and technical model performance from decision value. Ethical, legal, and governance considerations must be incorporated from the beginning because framing determines whose interests and risks are represented.

Key Terms

Business problem: A condition requiring a decision to achieve a desired organizational outcome.

Symptom: An observable indication of a possible underlying problem.

Decision owner: The person or group accountable and authorized to make a decision.

Objective: A desired outcome that guides action and evaluation.

Alternative: A feasible course of action, including the status quo.

Constraint: A limitation that restricts feasible alternatives.

KPI: A defined measure used to monitor progress toward an objective.

Decision criterion: A standard used to compare alternatives.

SMART objective: An objective that is specific, measurable, achievable, relevant, and time-bound.

Stakeholder: A person or group affected by, influencing, or responsible for a decision.

Hypothesis: A provisional, testable explanation or expectation.

Issue tree: A hierarchical decomposition of a broad question into smaller questions.

CRISP-DM: An iterative lifecycle for business understanding, data understanding, preparation, modeling, evaluation, and deployment.

Counterfactual: The outcome that would have occurred under an alternative condition or without an intervention.

Guardrail metric: A measure used to detect unacceptable side effects of optimizing a primary KPI.

Causal inference: The process of estimating the effect of an action or exposure on an outcome.

Data dictionary: Documentation defining data fields, meanings, formats, and sources.

Weighted scoring model: A method that combines criterion scores using explicit weights.

Decision brief: A concise document specifying a decision, evidence, alternatives, and evaluation plan.

Review Questions

Short Answer Questions

1. What is the difference between a symptom and a business problem?
2. Why should constraints be identified before recommending an alternative?
3. What characteristics make a KPI well defined?
4. What is the purpose of an issue tree?
5. Why is CRISP-DM considered iterative?

Long Answer Questions

1. Explain how a managerial concern can be translated into a structured decision problem. Use an example.
2. Describe SMART objectives and discuss two limitations of relying on numerical targets alone.
3. Explain the role of stakeholders in business problem framing and decision implementation.
4. Compare model performance with decision performance in an analytics project.
5. Discuss how ethical and governance considerations should influence problem framing.

Analytical / Application Questions

1. A retailer reports that revenue has fallen for three consecutive months. Construct an issue tree with at least three major branches and propose two testable hypotheses.
2. A call centre wants to reduce average handling time. Design a primary KPI and at least two guardrail metrics, explaining the risks of optimizing the primary KPI alone.
3. An organization is considering an employee-attrition intervention. Identify the decision owner, objective, alternatives, constraints, stakeholders, and evidence required before deployment.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a business problem from a symptom?

A. A business problem is always expressed as a numerical KPI.

B. A business problem identifies a decision and desired outcome, whereas a symptom is an observed condition.

C. A business problem is the same as a dataset.

D. A business problem describes only the technical model to be built.

Correct Answer: B. A business problem identifies a decision and desired outcome, whereas a symptom is an observed condition.

Question 2

Which objective is most clearly SMART?

A. Improve customer service soon.

B. Make customers happier using more data.

C. Increase tickets resolved within 24 hours from 78% to 90% by the end of the third quarter without increasing recurrence.

D. Investigate service quality when resources permit.

Correct Answer: C. Increase tickets resolved within 24 hours from 78% to 90% by the end of the third quarter without increasing recurrence.

Question 3

In a weighted scoring model, what must be true for scores to be combined meaningfully?

A. All criteria must have identical weights.

B. Scores must be comparable and criterion direction must be consistent.

C. Only financial criteria may be used.

D. Constraints should always be represented as low scores.

Correct Answer: B. Scores must be comparable and criterion direction must be consistent.

Question 4

Which CRISP-DM phase focuses on cleaning, combining, and transforming data?

A. Business understanding

B. Data understanding

C. Data preparation

D. Deployment

Correct Answer: C. Data preparation

Question 5

What is the main purpose of an issue tree?

A. To guarantee causal proof

B. To decompose a broad question into smaller analytical questions

C. To replace stakeholder consultation

D. To select a machine-learning algorithm automatically

Correct Answer: B. To decompose a broad question into smaller analytical questions

Question 6

A retailer has 7,440 on-time deliveries out of 8,000 eligible deliveries. What is the on-time delivery rate?

A. 7.44%

B. 88%

C. 93%

D. 95%

Correct Answer: C. 93%

Question 7

Why should a primary KPI often be paired with guardrail metrics?

A. To make the dashboard longer

B. To detect harmful trade-offs or unintended effects

C. To eliminate the need for an objective

D. To ensure every metric increases simultaneously

Correct Answer: B. To detect harmful trade-offs or unintended effects

Question 8

Which question most directly addresses the counterfactual?

A. How many columns are in the dataset?

B. What would have happened without the intervention?

C. Who owns the database?

D. Which chart has the highest resolution?

Correct Answer: B. What would have happened without the intervention?

Question 9

Which statement about a churn prediction model is correct?

A. High predictive accuracy proves that a retention intervention will work.

B. Prediction can identify likely churners, but intervention effectiveness and cost still require evaluation.

C. A churn model removes the need for decision criteria.

D. A churn model automatically establishes fairness.

Correct Answer: B. Prediction can identify likely churners, but intervention effectiveness and cost still require evaluation.

Question 10

Which item is normally best treated as a constraint rather than merely a weighted preference?

A. A legally mandatory requirement

B. A manager's preferred chart style

C. A possible marketing slogan

D. A minor formatting convention

Correct Answer: A. A legally mandatory requirement