Probability and Business Uncertainty

Introduction

Business decisions are made under uncertainty. A retailer must estimate whether demand will exceed available stock; a bank must assess whether an applicant will default; a manufacturer must judge whether a supplier will deliver on time; and a digital platform must decide whether a transaction is fraudulent. In each case, the future outcome is not known with certainty, but available information can be used to describe and manage uncertainty.

Probability provides the mathematical language for doing this. It supports forecasting, risk assessment, quality control, insurance pricing, credit scoring, inventory planning, marketing experimentation and decision analysis. Probability does not eliminate uncertainty. Instead, it quantifies uncertainty and provides a disciplined basis for updating beliefs when new evidence becomes available.

This chapter develops the foundations required for business analytics: probability rules, conditional probability, Bayes’ theorem, counting methods, random variables and expected value. The emphasis is on interpretation and responsible use rather than mechanical calculation alone.

Learning objectives

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

1. Describe experiments, outcomes, events and sample spaces.
2. Apply addition, multiplication and complement rules.
3. Distinguish mutually exclusive events from independent events.
4. Calculate and interpret conditional probabilities.
5. Use Bayes’ theorem to update probabilities after observing evidence.
6. Apply permutations and combinations to business counting problems.
7. Distinguish discrete and continuous random variables.
8. Calculate expected value and use it in business decisions.
9. Recognise assumptions, limitations and ethical risks in probabilistic analysis.

1. Foundations of probability

A random experiment is a process with an uncertain outcome. Examples include selecting a customer, observing a delivery, testing a product or recording whether a campaign generates a purchase. The set of all possible outcomes is the sample space, denoted by S. An event is a subset of the sample space.

For example, suppose a retailer classifies an online order as either delivered on time (T) or late (L). The sample space is S = {T, L}. The event “late delivery” is {L}.

A probability is a number from 0 to 1 that represents the likelihood of an event. For an event A:

Here P(A) means the probability of event A. A probability of 0 indicates impossibility, while 1 indicates certainty. The probability of the entire sample space is:

If the possible outcomes are mutually exclusive and collectively exhaustive, their probabilities sum to 1.

Interpretations of probability

Probability may be interpreted in several related ways:

• Classical probability assigns equal likelihood to outcomes, as in a fair die.
• Empirical probability uses observed relative frequencies. If 42 of 100 deliveries are late, the observed late-delivery rate is 0.42.
• Subjective or Bayesian probability represents a degree of belief based on available information and is updated when evidence changes.

Business analytics commonly combines these interpretations. Historical data may provide an empirical starting point, while expert judgment and new evidence may lead to Bayesian updating.

Complement rule

The complement of A, written Aᶜ, consists of all outcomes not in A. Since either A occurs or A does not occur:

This is useful when the complement is easier to measure. If the probability that an invoice is paid late is 0.08, then the probability that it is not paid late is 1 − 0.08 = 0.92.

2. Combining events

Addition rule

For any two events A and B:

The symbol ∪ means “A or B, or both,” and ∩ means “A and B.” The intersection must be subtracted because it is counted twice when P(A) and P(B) are added.

If A and B are mutually exclusive, they cannot occur together, so P(A ∩ B) = 0. The rule becomes:

For example, if a support request can be classified as either a billing issue or a technical issue, but not both, and the probabilities are 0.30 and 0.45, the probability of either category is 0.75.

If categories can overlap, the general addition rule must be used. Suppose 40% of customers use a mobile application, 55% use the website, and 25% use both. Then:

P(application or website) = 0.40 + 0.55 − 0.25 = 0.70.

Multiplication rule

The probability that both A and B occur is:

P(B|A) is the probability of B given that A has occurred. Equivalently:

These formulas apply generally, provided the conditional probabilities are defined.

If A and B are independent, occurrence of A does not change the probability of B. Therefore:

and the multiplication rule simplifies to:

Independence is an assumption requiring justification. Two events are not independent merely because they concern different variables or occur at different times. For instance, two purchases by the same customer may be related through customer preferences.

Mutually exclusive versus independent events

These concepts are often confused. Mutually exclusive events cannot occur together. Independent events do not affect each other’s probabilities. If two events have positive probability and are mutually exclusive, they cannot be independent because P(A ∩ B) = 0, whereas independence would require P(A ∩ B) = P(A)P(B) > 0.

3. Conditional probability

Conditional probability measures the likelihood of an event when another event is known to have occurred:

The notation P(A|B) is read as “the probability of A given B.” The event after the vertical bar is the conditioning event.

Worked example: customer response

A marketing database contains 1,000 customers. Of these, 300 opened an email, and 90 of the customers who opened it made a purchase. The probability of purchase given email opening is:

P(purchase | opened) = 90 ÷ 300 = 0.30.

This is different from the overall purchase probability, which would require the total number of purchasers among all 1,000 customers. Conditioning changes the reference group from all customers to customers who opened the email.

A useful operational principle is: always identify the denominator. Many business errors arise when a conditional probability is presented without making its reference population clear.

4. Bayes’ theorem and updating beliefs

Bayes’ theorem reverses a conditional probability. It determines the probability of a possible cause after observing evidence.

Here:

• P(A|B) is the posterior probability of A after evidence B;
• P(B|A) is the likelihood of observing B if A is true;
• P(A) is the prior probability of A;
• P(B) is the overall probability of the evidence.

If A₁, A₂, …, Aₙ are mutually exclusive and collectively exhaustive possible causes, then:

The denominator ensures that posterior probabilities sum to 1.

Worked example: fraud screening

Assume that 2% of transactions are genuinely fraudulent. A screening system flags 90% of fraudulent transactions and incorrectly flags 5% of legitimate transactions. Let F mean “fraudulent” and + mean “flagged.” Thus:

• P(F) = 0.02;
• P(+|F) = 0.90;
• P(Fᶜ) = 0.98;
• P(+|Fᶜ) = 0.05.

First calculate the probability of any transaction being flagged:

P(+) = P(+|F)P(F) + P(+|Fᶜ)P(Fᶜ)

P(+) = (0.90 × 0.02) + (0.05 × 0.98) = 0.018 + 0.049 = 0.067.

Now calculate the probability that a flagged transaction is genuinely fraudulent:

P(F|+) = (0.90 × 0.02) ÷ 0.067 ≈ 0.2687.

Therefore, only about 26.9% of flagged transactions are expected to be fraudulent. The system detects most frauds, but the low base rate of fraud means that false positives are substantial.

This is the base-rate effect: predictive accuracy depends not only on the performance of a test or model but also on the underlying prevalence of the event.

Frequency interpretation

The same result is often easier to communicate using 10,000 transactions:

The proportion of flagged transactions that are fraudulent is 180 ÷ 670 ≈ 26.9%. Frequency tables can make Bayesian reasoning more transparent to managers and stakeholders.

5. Counting methods

Counting methods calculate the number of possible arrangements or selections. They are useful for sampling plans, password security, scheduling, committee formation and experimental design.

Factorials

For a non-negative integer n, the factorial is:

By convention, 0! = 1.

Permutations

A permutation is an arrangement where order matters. The number of ways to arrange r objects selected from n distinct objects is:

For example, the number of ways to assign three distinct management roles to three people chosen from eight employees is:

P(8,3) = 8! ÷ 5! = 8 × 7 × 6 = 336.

Combinations

A combination is a selection where order does not matter:

The number of five-person project teams chosen from 12 employees is:

C(12,5) = 12! ÷ (5!7!) = 792.

The distinction is practical. Selecting a committee uses combinations; assigning chairperson, secretary and treasurer uses permutations because the roles differ.

Counting with probabilities

If all outcomes are equally likely, the probability of an event is:

This formula applies only when outcomes are equally likely. It should not be used automatically for observational business data, where outcomes may have unequal probabilities.

6. Random variables

A random variable assigns a numerical value to each outcome of a random experiment. It is usually denoted by X or Y.

A discrete random variable takes countable values, such as the number of returns in a day, the number of defective units in a batch or the number of sales generated by a campaign.

A continuous random variable can take any value within an interval, such as delivery time, transaction value or waiting time. For a continuous variable, probability is assigned to intervals rather than individual points. Under a continuous model, P(X = x) = 0 for any single exact value x, while P(a ≤ X ≤ b) may be positive.

Probability distribution

A probability distribution specifies the possible values of X and their probabilities. For a discrete variable:

and each probability must satisfy 0 ≤ P(X = x) ≤ 1.

Example:

The probabilities sum to 1.00, so this is a valid distribution.

Expected value

The expected value, or mean, of a discrete random variable is the probability-weighted average of its possible values:

The units of E(X) are the same as the units of X. For the sales distribution above:

E(X) = (0 × 0.10) + (1 × 0.30) + (2 × 0.40) + (3 × 0.20)

E(X) = 0 + 0.30 + 0.80 + 0.60 = 1.70 sales.

The expected value need not be an outcome that can actually occur. A business may not sell exactly 1.70 units in a particular period, but 1.70 is the long-run average per comparable period under the stated probabilities.

For a continuous random variable with probability density f(x), the expected value is:

The density must be non-negative and integrate to 1 over its support. At diploma level, the discrete formula is usually sufficient for practical decision tables, while the continuous form explains the general principle.

Variance and standard deviation

Expected value describes location, but not variability. For a discrete random variable:

where μ = E(X). An equivalent computational formula is:

The standard deviation is:

Variance is expressed in squared units, while standard deviation has the same units as X. A larger standard deviation indicates greater dispersion around the expected value, although it does not by itself describe asymmetry or extreme-tail behaviour.

7. Expected value in business decisions

Expected value supports decisions involving repeated or comparable uncertain outcomes. If an action has possible payoffs x₁, x₂, …, xₙ with probabilities p₁, p₂, …, pₙ, then:

The probabilities must be non-negative and sum to 1. Payoffs should be measured on a consistent basis, such as net profit rather than mixing revenue and profit.

Worked example: product launch

A firm is considering a product launch. Estimated net profits are:

Expected profit is:

E(X) = (0.30 × 120,000) + (0.50 × 50,000) + (0.20 × −40,000)

E(X) = 36,000 + 25,000 − 8,000 = $53,000.

If the alternative of not launching has a certain profit of $35,000, the launch has the higher expected monetary value by $18,000. However, this does not prove that launching is the correct decision. The firm should also consider risk tolerance, cash-flow constraints, the possibility that probabilities are inaccurate, strategic benefits and the downside of a loss.

Expected value of perfect information

Suppose a manager could know the market outcome before deciding. The expected value with perfect information is the probability-weighted value of the best decision in each state. The expected value of perfect information (EVPI) is:

EVPI is an upper bound on what the firm should pay for perfectly accurate information. Real market research is imperfect, so its acceptable cost is generally less than EVPI.

8. Applied case: credit-risk review

Background

This illustrative case concerns a digital lender reviewing applications for a small business loan. The lender wants to estimate default risk and decide whether an application should receive manual review.

Problem

Historical data indicate that 6% of applications default within 12 months. A predictive screening model flags 85% of eventual defaulters but also flags 12% of non-defaulters. Management initially proposes rejecting every flagged application.

Analysis

Let D mean default and F mean flagged. Then:

• P(D) = 0.06;
• P(F|D) = 0.85;
• P(Dᶜ) = 0.94;
• P(F|Dᶜ) = 0.12.

The probability of a flag is:

P(F) = (0.85 × 0.06) + (0.12 × 0.94)

P(F) = 0.051 + 0.1128 = 0.1638.

The posterior probability of default given a flag is:

P(D|F) = 0.051 ÷ 0.1638 ≈ 0.311.

Thus, approximately 31.1% of flagged applicants are expected to default. Most flagged applicants—approximately 68.9%—would not default. Automatic rejection would therefore deny credit to many potentially viable businesses.

Approach and solution

The lender adopts a three-stage policy:

1. Use the model as a screening tool rather than a final decision.
2. Send flagged applications to trained analysts for additional evidence, such as cash-flow statements and repayment history.
3. Monitor false-positive rates, approval rates, default rates and outcomes across customer groups.

Additional evidence can be incorporated through conditional probabilities or a revised Bayesian model. The lender also calibrates predicted probabilities, meaning that applications assigned a 20% risk should default at approximately that rate over a sufficiently large comparable group.

Outcome

The policy reduces automatic rejection, retains human review for higher-risk cases and provides a clearer basis for measuring model performance. It also acknowledges that the model’s output is probabilistic, not a fact about an individual applicant.

Learning points

• Base rates strongly affect predictive results.
• A high sensitivity rate does not imply a high positive predictive value.
• Expected costs should include both missed defaults and unjustified rejections.
• Model governance requires monitoring, documentation and review for disparate impact.

9. Limitations, governance and ethical practice

Probability calculations are only as reliable as their assumptions, data and decision context.

Data quality and selection bias

Historical data may exclude informal businesses, contain reporting errors or reflect previous lending policies. A probability estimated from one customer segment may not apply to another. Analysts should define the target population, time period and measurement process.

Dependence and changing conditions

Events that appear independent may share common causes. Economic shocks can simultaneously increase default, unemployment and late payment. Probabilities estimated in stable conditions may fail during a crisis. Models should therefore be stress-tested under alternative scenarios.

Base rates and communication

Analysts should report both conditional probabilities and base rates where relevant. Confusing P(A|B) with P(B|A) is a common and consequential error. Results should be communicated using frequencies, confidence about assumptions and plain-language explanations.

Fairness and privacy

Risk models may reproduce historical discrimination if protected characteristics or proxy variables are used. Organisations should examine error rates across relevant groups, document legitimate purposes for data use and comply with applicable privacy and anti-discrimination requirements. A model should not be treated as neutral merely because it uses mathematics.

Human oversight and AI

Machine-learning systems often produce probability scores, but a score is not automatically a well-calibrated probability. Organisations should validate calibration, monitor drift, record model versions and establish escalation procedures. Human review is particularly important where decisions affect access to credit, employment, insurance, housing or essential services. Automation should support accountable decisions rather than obscure responsibility.

10. Best practices for analysts

1. Define the event and reference population precisely.
2. State whether probabilities are empirical, model-based or subjective.
3. Check that probabilities lie between 0 and 1 and sum correctly where required.
4. Distinguish “or” from “and,” and distinguish mutually exclusive from independent events.
5. Use the general multiplication and addition rules unless independence or exclusivity is justified.
6. Show the base rate when presenting a classification result.
7. Use expected value for repeated decisions, but also examine variability and downside outcomes.
8. Test sensitivity to changes in probabilities and payoffs.
9. Keep monetary units and time periods consistent.
10. Document assumptions, data sources and limitations.
11. Validate models on data not used to build them.
12. Review decisions for fairness, privacy and practical consequences.

Chapter Summary

Probability quantifies uncertainty about events and provides rules for combining information. The complement rule calculates the probability that an event does not occur. The general addition rule handles “A or B,” while the multiplication rule handles “A and B.” Mutually exclusive events cannot occur together; independent events do not change one another’s probabilities.

Conditional probability changes the reference population and is defined by P(A|B) = P(A ∩ B) ÷ P(B). Bayes’ theorem uses prior probability, likelihood and overall evidence to produce a posterior probability. It is especially important in classification, fraud detection and credit risk because base rates influence the meaning of a positive signal.

Permutations count ordered arrangements, whereas combinations count unordered selections. Random variables assign numerical values to uncertain outcomes. Their probability distributions describe possible values and probabilities. Expected value is a probability-weighted average and is useful for repeated decisions, although it should be considered alongside variability, risk tolerance and model uncertainty.

Professional probability analysis requires more than correct arithmetic. Analysts must define events, justify assumptions, assess data quality, communicate base rates, monitor changing conditions and address fairness, privacy and accountability. Used responsibly, probability helps organisations make more transparent and defensible decisions under uncertainty.

Key Terms

Bayes’ theorem: A rule for updating the probability of a cause after observing evidence.

Base rate: The underlying prevalence or prior probability of an event.

Combination: A selection in which order does not matter.

Complement: The event that an event does not occur.

Conditional probability: The probability of an event given that another event is known to have occurred.

Continuous random variable: A random variable that can take values throughout an interval.

Event: A subset of possible outcomes from a sample space.

Expected value: The probability-weighted average of the possible values of a random variable.

Independence: A relationship in which occurrence of one event does not alter the probability of another.

Intersection: The event that two events occur together.

Likelihood: The probability of observed evidence conditional on a proposed cause or state.

Mutually exclusive events: Events that cannot occur simultaneously.

Outcome: A possible result of a random experiment.

Permutation: An arrangement in which order matters.

Posterior probability: The probability of an event after incorporating evidence.

Prior probability: The probability assigned before new evidence is considered.

Probability distribution: A specification of possible values and their probabilities.

Random experiment: A process whose outcome is uncertain.

Random variable: A numerical assignment to outcomes of a random experiment.

Sample space: The complete set of possible outcomes.

Review Questions

5 Short Answer Questions

1. What is the difference between a sample space and an event?
2. State the complement rule and explain when it is useful.
3. Distinguish mutually exclusive events from independent events.
4. What does the base rate represent in a classification problem?
5. Why might expected value not be sufficient for choosing between two business actions?

5 Long Answer Questions

1. Explain the general addition and multiplication rules, including the conditions under which their simplified forms may be used.
2. Describe Bayes’ theorem and explain the roles of prior probability, likelihood and posterior probability.
3. Compare permutations and combinations and provide one business example of each.
4. Explain discrete and continuous random variables and discuss how their probability descriptions differ.
5. Discuss the ethical and governance issues that arise when probability models are used for credit or employment decisions.

3 Analytical / Application Questions

1. A retailer estimates that 4% of orders are fraudulent. Its screening system flags 92% of fraudulent orders and 6% of legitimate orders. Calculate the probability that a flagged order is fraudulent and explain the business implication.
2. A company can launch a service with probabilities 0.25, 0.50 and 0.25 for profits of $80,000, $30,000 and −$20,000 respectively. Calculate expected profit and explain two additional factors management should consider.
3. A project team must select four analysts from ten employees, then assign one selected analyst as team leader. Explain whether the selection and leadership assignment should be counted using a combination, a permutation or a combination followed by a separate assignment. Calculate the number of possible outcomes.

Multiple Choice Questions with Answers

Question 1

If P(A) = 0.30, what is P(Aᶜ)?

A. 0.30

B. 0.70

C. 1.30

D. −0.30

Correct Answer: B. 0.70

Question 2

If P(A) = 0.40, P(B) = 0.50 and P(A ∩ B) = 0.20, what is P(A ∪ B)?

A. 0.20

B. 0.50

C. 0.70

D. 0.90

Correct Answer: C. 0.70

Question 3

Which condition defines independence of A and B?

A. P(A ∩ B) = 0

B. P(A|B) = 0

C. P(A|B) = P(A)

D. P(A) + P(B) = 1

Correct Answer: C. P(A|B) = P(A)

Question 4

A test has P(D) = 0.10, P(+|D) = 0.80 and P(+|Dᶜ) = 0.05. What is P(D|+), approximately?

A. 0.10

B. 0.20

C. 0.64

D. 0.80

Correct Answer: C. 0.64

Question 5

How many ways can three distinct roles be assigned to three people selected from six people?

A. 20

B. 56

C. 120

D. 216

Correct Answer: C. 120

Question 6

How many five-person teams can be selected from ten employees when order does not matter?

A. 25

B. 50

C. 120

D. 252

Correct Answer: D. 252

Question 7

A random variable X takes values 0 and 2 with probabilities 0.25 and 0.75. What is E(X)?

A. 0.50

B. 1.00

C. 1.50

D. 2.00

Correct Answer: C. 1.50

Question 8

Which statement about a continuous random variable is generally correct?

A. It can take only integer values.

B. The probability of every exact value is usually zero.

C. Its probabilities must be counted using factorials.

D. Its expected value must be an observed outcome.

Correct Answer: B. The probability of every exact value is usually zero.

Question 9

A project has outcomes of $10,000 with probability 0.6 and −$2,000 with probability 0.4. What is its expected value?

A. $4,000

B. $5,200

C. $6,000

D. $8,000

Correct Answer: B. $5,200

Question 10

Why can a screening system with high sensitivity still have many false positives?

A. Sensitivity is always equal to posterior probability.

B. The sample space has probability greater than 1.

C. A low base rate can make legitimate cases numerous relative to true cases.

D. Independent events cannot be screened.

Correct Answer: C. A low base rate can make legitimate cases numerous relative to true cases.