Bayes' theorem for business decisions, with a worked example

What does Bayes’ theorem say in plain terms?

The theorem reads P(H | E) = P(E | H) × P(H) ÷ P(E). In words, the probability of the hypothesis H given the evidence E equals the probability of the evidence if H is true, times the probability of H before the evidence, divided by the overall probability of the evidence. The result is the posterior probability.

The overall probability of the evidence adds two paths: evidence that appears when H is true and evidence that appears when H is false. Both paths matter. A decision maker who looks only at the first path overstates the posterior, which is the root of many costly misreadings of a test, a flag, or a signal.

What do the terms of Bayes’ theorem mean in a business setting?

The prior is the probability of H before the evidence, ideally taken from a base rate. The likelihood is the probability of the evidence if H is true. The false alarm rate is the probability of the same evidence if H is false. The posterior is the updated probability that drives the decision.

In business terms, H may be that a supplier is unreliable, a lead will buy, or a project will overrun. E may be an audit flag, a demo request, or a missed milestone. Naming H and E in plain words before any arithmetic keeps the team from reversing the conditional probabilities.

How do natural frequencies make Bayes’ theorem easier to apply?

Replace probabilities with counts out of 1,000. Suppose 4 percent of suppliers are unreliable, so 40 of 1,000 are. An audit flags 90 percent of unreliable suppliers, or 36 of the 40, and wrongly flags 10 percent of reliable ones, or 96 of the 960. Counting people or suppliers is far more intuitive than multiplying fractions.

The flagged group has 36 + 96 = 132 suppliers. Of these, 36 are unreliable. The posterior is 36 divided by 132, or about 27 percent. The count makes the logic visible: many reliable suppliers exist, so even a small false alarm rate produces more false flags than true ones.

What does Bayes’ theorem say about a flagged supplier?

The formula gives the same answer. The numerator is 0.90 × 0.04, which equals 0.036. The denominator adds 0.036 to 0.10 × 0.96, which equals 0.096, for a total of 0.132. The posterior is 0.036 divided by 0.132, or 0.273. A flag moves the chance from 4 percent to about 27 percent, which is a large rise and still far from certain.

This result surprises many managers, who expect a 90 percent test to mean a 90 percent chance. The 90 percent describes how often unreliable suppliers are flagged, not how often flagged suppliers are unreliable. A decision based on the flag alone would put many reliable suppliers on a watch list, so the flag is a reason to investigate.

How does Bayes’ theorem handle a second independent flag?

Odds make repeated updates simple. A prior of 27 percent equals odds of 0.375 to 1. The likelihood ratio of a flag is 0.90 divided by 0.10, or 9. Multiply 0.375 by 9 to get odds of 3.375 to 1, which equals a probability of 3.375 divided by 4.375, or about 77 percent.

This holds only if the second flag is independent of the first. Two flags from the same checklist, or from the same inspector, share a cause and count as one piece of evidence. A second flag from a different source, such as a financial check and a site visit, carries its own weight.

What does Bayes’ theorem say after a clean supplier audit?

The same formula handles a clean result. The chance of a clean audit for an unreliable supplier is 10 percent, and for a reliable supplier 90 percent. The posterior is 0.04 × 0.10 divided by the sum of 0.004 and 0.96 × 0.90, which is 0.864. The result is 0.004 divided by 0.868, or about 0.5 percent.

A clean audit pushes the chance from 4 percent to about 0.5 percent. The team can now set a rule: investigate when the posterior passes 20 percent, and skip deeper checks below 1 percent. The rule connects the posterior to an action and a cost, which is what turns a probability into a decision.

Which mistakes distort Bayes’ theorem in business decisions?

Three mistakes recur. Ignoring the base rate makes a rare event look likely after a flag. Reading the likelihood as the posterior confuses how often evidence appears with how likely the hypothesis is. Treating dependent evidence as independent inflates the posterior. A fourth habit, guessing likelihoods without data, turns the arithmetic into decoration.

Protect the calculation with sources. Take the prior from a base rate, take the likelihoods from past audits or pilot data, and state the independence assumption in writing. A sensitivity check that varies the prior and the false alarm rate shows how much the posterior depends on them.

How does the EQOS Asia decision engine connect to Bayes’ theorem?

Bayes’ theorem needs a prior and likelihoods that people supply. The EQOS Asia decision engine takes a deterministic path: its Codex Universalis core returns scores and verdicts by arithmetic, not by language-model inference, and each can be recomputed identically (EQOS Asia method page, as of October 2026).

A reader can try a first score in the simulator on a real decision and keep the Bayesian posterior beside it. The monthly Sentinel recompute refreshes the score when the facts change, in the same spirit as updating a probability when new evidence arrives.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile