What it means
The theorem links two conditional probabilities that people constantly confuse. The probability that a fraudulent invoice triggers an alert is not the same as the probability that an alerted invoice is fraudulent.
Bayes' theorem converts one into the other by taking account of how rare the underlying event is. That base rate is the piece most people leave out.
If only 1% of invoices are fraudulent, then even a fairly accurate detector will produce far more false alarms than real catches, simply because there are so many more honest invoices to misclassify. Ignoring this is known as the base rate fallacy, and it is why teams so often overestimate the value of a new screening tool.
In practice the theorem gives a disciplined way to revise numbers rather than guessing. A credit team starts with a default rate for a customer segment, observes something new such as a late payment or a filed charge, and computes a revised probability of default.
The revision is proportional to how much more common that signal is among defaulters than among sound customers. The same logic runs behind Bayesian forecasting used in planning.
You begin with a prior view, such as last year's conversion rate, then update it as this quarter's data arrives rather than throwing away everything you knew. With small samples the prior does most of the work; as evidence accumulates, the data dominates.
The main caution is the quality of the inputs. A confidently wrong base rate produces a confidently wrong answer, and evidence that is correlated with something already in the model gets double counted.
The theorem is arithmetic, not judgement, so it improves reasoning without replacing it.
In practice
Real-world examples.
Example
A bank's transaction monitoring system alerts on 2% of card payments. Because genuine fraud affects a tiny share of payments, the compliance team uses Bayes' theorem to show the board that most alerts will be false positives no matter how good the model is, which reframes the debate from accuracy to review capacity.
Example
An external auditor samples 60 invoices and finds two errors. Rather than concluding the error rate is 3.3%, the audit manager combines the sample with the prior three years of results for the same client to produce a revised estimate that is less swayed by one small sample.
Example
A subscription business sees a customer downgrade their plan. The retention model treats the downgrade as evidence and updates that account's churn probability from the segment average of 8% to 26%, which moves it into the queue for a proactive call.
Formula
Calculation
P(A given B) = [P(B given A) x P(A)] / P(B), where P(B) = [P(B given A) x P(A)] + [P(B given not A) x P(not A)].
An accounts payable team runs an automated fraud screen. Historically 1% of invoices are fraudulent. The screen flags 90% of genuinely fraudulent invoices, and it wrongly flags 5% of legitimate ones. An invoice has just been flagged; what is the probability it is actually fraudulent?
The easiest way to see it is to work through 10,000 invoices.
Fraudulent invoices = 1% x 10,000 = 100. Of these, 90% are flagged: 0.90 x 100 = 90.
Legitimate invoices = 10,000 - 100 = 9,900. Of these, 5% are flagged: 0.05 x 9,900 = 495.
Total flagged = 90 + 495 = 585.
P(fraud given flagged) = 90 / 585 = 0.1538, or about 15.4%.
The same result comes from the formula: P(flagged) = (0.90 x 0.01) + (0.05 x 0.99) = 0.009 + 0.0495 = 0.0585, so P(fraud given flagged) = 0.009 / 0.0585 = 0.1538.
A screen that catches 90% of fraud still means fewer than one flagged invoice in six is genuinely fraudulent. That is not a failure of the tool; it is the arithmetic of a rare event, and it tells the team to size the review queue for roughly 585 cases per 10,000 invoices rather than 100.Case study
Seen in the real world.
The following is an illustrative and fictional scenario. Larkspur Logistics, an invented freight brokerage, deployed a supplier fraud screen and celebrated when it flagged 240 invoices in its first month out of roughly 20,000 processed. The operations director told the board the company had "found 240 fraud cases" and asked for three extra investigators.
The finance analyst applied Bayes' theorem instead. With an estimated base rate of 0.5% fraudulent invoices, a detection rate of 85% and a false positive rate of 1%, the expected numbers on 20,000 invoices were 100 fraudulent invoices of which 85 were flagged, and 19,900 legitimate invoices of which 199 were flagged, giving 284 flags in total. On those assumptions only about 85 of 284 flags, or roughly 30%, would be genuine.
That reframing changed the request. Instead of hiring three investigators to chase every flag, Larkspur added a cheap second check that only escalated flagged invoices from suppliers onboarded within the previous year, cutting the queue by more than half while keeping most of the true cases. The illustrative point is that the value of a detector depends as much on how rare the event is as on how accurate the detector claims to be.
Watch out
Common mistakes.
- Reading the accuracy of a test backwards, assuming that a 90% detection rate means a flagged item is 90% likely to be genuine.
- Leaving out the base rate entirely, which makes rare events look far more common than they are once a screen is applied.
- Feeding in two pieces of evidence that measure the same underlying thing, so the same signal is counted twice and the revised probability is overstated.
Questions
People also ask.
What is a prior?
It is the probability you assign before seeing the new evidence, usually taken from historical rates or a well-argued estimate.
Do I need statistical software to use it?
No; for a single yes or no test the natural frequency approach shown above needs nothing more than a calculator and a clear head.
Does the theorem work with a weak prior?
Yes, but the answer will be sensitive to it, so it is worth testing a range of plausible priors and seeing whether the conclusion changes.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%