5 Probability for Business Analytics

Events and probability rules

Probability provides a formal way to describe uncertainty. In business analytics, it helps quantify uncertain outcomes such as customer demand, late deliveries, product defects, and investment returns, and supports comparisons between decisions under risk.

An experiment is an uncertain process, and an outcome is one possible result. An event is a set of outcomes. For example, if a shipment can arrive on time or late, “arrives late” is an event.

A probability ranges from $0$ to $1$: $0$ means an event cannot occur, and $1$ means it is certain. The complement of an event $A$, written $A^c$, is the event that $A$ does not occur:

For any two events, the probability that at least one occurs is given by the union rule:

The subtraction avoids counting outcomes in both events twice. If the events cannot occur together, then $P(A\cap B)=0$.

Conditional probability and independence

Conditional probability describes the chance of an event when another event is known to have occurred. For events $A$ and $B$, it is

Events $A$ and $B$ are independent when knowing that one occurred does not change the probability of the other. In that case, $P(A\cap B)=P(A)P(B)$. Independence is an assumption to check, not something to presume; for example, repeat purchases by the same customer may be related.

Random variables and distributions

A random variable assigns a numerical value to each possible outcome. For example, $X$ might be the number of defective items in a shipment, or $Y$ might be delivery time in days. A probability distribution describes the possible values and their probabilities. A distribution can be summarized by its shape, center, and spread.

A discrete random variable takes countable values, such as $0, 1, 2,$ and so on. Its probability mass function (PMF) gives $P(X=x)$ for each value. These probabilities are between $0$ and $1$ and sum to $1$.

A continuous random variable can take any value within a range, such as delivery time or transaction amount. Its probability density describes how probability is distributed over the range. Probabilities correspond to areas over intervals; for a continuous variable, the probability of any exact single value is zero.

The cumulative distribution function is

It is useful for questions such as the probability that demand is no more than $500$ units, or the delivery-time threshold that $90\%$ of orders meet.

Expected value and variability

The expected value is the probability-weighted long-run average. For a discrete random variable, it is

The expected value need not be a value that can occur in one observation. Suppose daily demand is $100$ units with probability $0.3$, $200$ units with probability $0.5$, and $300$ units with probability $0.2$. Then

This is the average implied by the model over many comparable days, not a guarantee that demand on a particular day will be $190$ units.

Variance and standard deviation describe spread around the expected value:

For the demand example, the variance is $4{,}900$ units squared and the standard deviation is $70$ units. Two opportunities with the same expected profit can therefore have different levels of uncertainty. Expected value describes the center; standard deviation helps describe typical spread.

Common probability distributions

Choose a distribution that matches the business process and its assumptions; a familiar distribution name does not by itself make a model appropriate.

• Bernoulli: Models one trial with two outcomes, such as a customer converting or not converting. If the conversion probability is $p$, the outcome is $1$ with probability $p$ and $0$ otherwise.

• Binomial distribution: Counts successes in a fixed number $n$ of independent trials with the same success probability $p$. It can model sales among a fixed set of prospects if those assumptions are reasonable. Its mean is $np$, and its variance is $np(1-p)$.

• Poisson distribution: Models a count of events in a fixed interval when events occur at a roughly stable average rate and the model’s other assumptions are suitable. It may be used as a starting point for calls per hour or defects per batch. Its mean and variance are both $\lambda$, the expected count per interval.

• Normal distribution: A continuous, symmetric, bell-shaped model, often used for measurements or errors when the data and context support it. It can help estimate the likelihood of values near or far from the mean, but may be unsuitable for strongly skewed or bounded quantities such as nonnegative demand.

These distributions provide compact models for calculating probabilities, comparing scenarios, and simulating possible outcomes.

Probability in business decisions

A retailer estimating demand at $100$, $200$, or $300$ units with probabilities $0.3$, $0.5$, and $0.2$, respectively, obtains expected demand of $190$ units. That estimate can inform a stocking plan, but it does not by itself identify the best order quantity. The retailer also needs to consider the costs of unsold inventory and lost sales, the range of possible demand, and any service-level goals.

More generally, a decision-maker can assign a payoff to each possible outcome and calculate its expected payoff. Expected profit, for example, is the sum of each scenario’s profit multiplied by its probability. The option with the greatest expected payoff is not automatically the best choice: risk tolerance, cash constraints, severe downside outcomes, and the reliability of the probability estimates may also matter.

Probabilities may come from historical data, experiments, or expert judgment. Historical estimates are most useful when the data are relevant to the current situation; changing customer behavior, operating conditions, or market conditions can make past frequencies less representative. Probability describes uncertainty under a model and does not guarantee a particular result.

• Independent events