8 Quantitative Business Decision Making

Frame the decision

Begin by making clear what the business can choose, what conditions may occur outside its control, and what follows from each choice under each condition. Possible states of the world include uncertain factors such as demand or input costs. Measure consequences in a unit relevant to the decision, such as profit, cost, time, or customer retention.

State the objective and constraints as well, including budget, risk tolerance, and legal or operational limits. A payoff table makes the comparison explicit. Include costs and benefits that change between alternatives, and exclude sunk costs that cannot be recovered.

Compare uncertain outcomes

When probabilities for possible outcomes are known or reasonably estimated, expected value compares alternatives using their probability-weighted payoffs:

Here, $p_i$ is the probability of state $s_i$, and $V(a,s_i)$ is the payoff from alternative $a$ in that state. Probabilities must be nonnegative and sum to $1$. Expected value represents a long-run average across comparable decisions; it does not guarantee the result of one decision.

Comparing the alternatives

A company is choosing between expanding capacity and staying conservative. Payoffs are in thousands of dollars. High demand has probability $0.40$, and low demand has probability $0.60$.

• Expand: Payoffs are $120$ under high demand and $-40$ under low demand. Its expected value is $0.40(120)+0.60(-40)=24$.

• Stay conservative: Payoffs are $45$ under high demand and $25$ under low demand. Its expected value is $0.40(45)+0.60(25)=33$.

On expected monetary value alone, staying conservative is preferred: its expected payoff is $\$33{,}000$, compared with $\$24{,}000$ for expansion. This does not mean the conservative option has the higher payoff in every scenario.

A decision tree represents decisions as branches and uncertain outcomes as chance branches. Calculate expected values at chance points, then work backward to compare choices.

Assess risk and information

Expected monetary value does not capture every business concern. A firm with limited cash may reasonably avoid an alternative with a small chance of a severe loss, even if that alternative has a higher expected monetary value. Risk tolerance, liquidity, strategic fit, and nonfinancial outcomes may lead a decision-maker to consider utility or add constraints to the analysis.

Test sensitivity to assumptions

Sensitivity analysis tests whether a recommendation changes when uncertain inputs change. In the capacity example, let $p$ be the probability of high demand. In thousands of dollars, expansion has expected value $160p-40$, while the conservative option has expected value $20p+25$. Expansion is preferred only when $p>\frac{65}{140}$, or about $0.464$. If a plausible change in the demand estimate crosses this threshold, the recommendation is sensitive, and the probability estimate deserves scrutiny.

Value additional information

The expected value of perfect information (EVPI) is the expected payoff with advance knowledge of the true state minus the best expected payoff without that knowledge. In the example, perfect information would produce an expected payoff of $0.40(120)+0.60(25)=63$, in thousands of dollars. Therefore, EVPI is $63-33=30$ thousand dollars. This is an upper bound on what the firm should pay for information that perfectly reveals demand.

Real market research is imperfect. Its value depends on how much it is expected to improve the decision, so compare that value with the research cost. Decision analysis can help determine whether acquiring additional information is worthwhile.

Interpret statistical evidence

Business probabilities and payoffs often come from data. Descriptive statistics summarize observed results: the mean describes a typical level, while the median is less affected by unusually large or small values. The range and standard deviation describe variation. Pair averages with relevant measures of spread, and examine results across important groups or time periods because an overall average can hide meaningful differences.

A sample supports a conclusion only to the extent that it is appropriate for that purpose. Random sampling, adequate coverage of the target population, and careful measurement improve the credibility of estimates. Nonresponse, selection effects, small samples, and changing conditions can make a result that appears precise misleading. A sample estimate is uncertain; a confidence interval communicates a range of plausible values under the method's assumptions.

Hypothesis tests assess how compatible sample data are with a specified null hypothesis. A small p-value is evidence against that null under the test assumptions. It is not the probability that the null hypothesis is true, and it does not measure the size or business importance of an effect. Failure to reject a null hypothesis is not proof that there is no effect. Interpret tests alongside effect sizes, confidence intervals, study design, and the costs of false alarms or missed opportunities.

Use regression carefully

Regression can quantify an association and support forecasts. For example, a sales model may estimate how sales vary with price while accounting for other measured factors. Check whether the relationship is appropriate for the data, inspect residual patterns and unusual observations, and avoid extrapolating far beyond the observed range. A regression association alone does not establish that changing one variable will cause the predicted change; causal claims require a suitable design and assumptions.

Turn analysis into action

Translate statistical findings into business consequences. Ask whether an estimated effect is large enough to change revenue, cost, service quality, or risk; whether its uncertainty could change the preferred alternative; and which assumptions drive the result.

Set decision thresholds in advance when possible, report uncertainty and limitations plainly, and monitor outcomes after implementation. Quantitative analysis informs judgment but does not replace it. A sound choice depends not only on a numerical average, but also on risk, constraints, evidence quality, and business priorities.