What it means
There are two ways to arrive at a probability. Theoretical probability comes from the structure of the situation, such as knowing a fair coin has two equally likely sides; empirical probability comes from watching what happened and counting.
Most business questions have no clean theoretical structure, so empirical probability is the practical tool. A finance team asking how likely a customer is to pay late has no formula for it, but it does have three years of invoice history, and that history is the answer.
Reliability depends almost entirely on sample size. A rate calculated from 20 observations swings wildly with each new data point, while one calculated from 2,000 is stable enough to plan around, which is why analysts quote the sample size alongside the probability.
The other condition is that the past has to resemble the future. If a company changed its credit terms, entered a new market or replaced its collections team, the historical rate describes a business that no longer exists, and the honest response is to weight recent data more heavily or start a fresh count.
Empirical probability feeds directly into credit loss provisions, warranty reserves, staffing models and insurance pricing. In each case an observed frequency is turned into an expected cost by multiplying the probability by the amount at stake.
In practice
Real-world examples.
Example
A subscription software firm counts that 84 of its 1,200 monthly subscribers cancelled last month, giving an empirical churn probability of 7%. Finance uses that rate to forecast recurring revenue for the next four quarters.
Example
An appliance manufacturer records 315 warranty claims against 21,000 units sold, an empirical claim probability of 1.5%. The figure drives the warranty provision booked in the year-end accounts.
Example
A logistics operator finds that 42 of its 700 scheduled deliveries in a quarter missed their promised window, a 6% failure rate. Operations uses this to negotiate realistic service level penalties in a new client contract.
Formula
Calculation
Empirical probability = Number of times the event occurred / Total number of observations
A wholesaler wants to know the likelihood that a customer invoice is paid more than 30 days late. Over the past year the company issued 480 invoices, and 96 of them were paid more than 30 days past the due date.
Empirical probability = 96 / 480 = 0.20, or 20%
Now apply it. If the company expects to issue 150 invoices next quarter, the expected number paid more than 30 days late is:
150 x 0.20 = 30 invoices
If the average invoice is $8,000 and late payment costs the business roughly 2% of the invoice value in financing and chasing costs, the expected cost is 30 x $8,000 x 0.02 = $4,800 for the quarter.Case study
Seen in the real world.
This illustrative, fictional case concerns Braewick Equipment Hire, a business renting construction plant to small builders. The finance director had been provisioning for bad debts using a flat 3% of revenue, a figure inherited from a predecessor with no record of where it came from.
She pulled five years of transaction history and counted directly: of 6,000 rental contracts, 240 had ended in a write-off, an empirical probability of 4%. Split by customer type, the picture was sharper still, with new customers defaulting at 9% and customers of more than two years' standing at 1%.
Braewick rebuilt its provisioning on those observed rates and changed its credit policy at the same time, requiring a deposit from any customer in their first year. Two years later the new-customer default rate had fallen to 5%, and because the provision was now based on counted evidence rather than an inherited assumption, the auditors accepted it without challenge.
Watch out
Common mistakes.
- Calculating a probability from a handful of observations and treating it as reliable, when small samples produce wildly unstable rates.
- Applying a company-wide average to a segment that behaves very differently, such as using the overall default rate for a brand new customer group.
- Continuing to use a historical rate after a major change in process, pricing or market, so the number describes conditions that no longer apply.
Questions
People also ask.
How many observations do you need for a usable empirical probability?
There is no hard rule, but a few hundred observations give a much steadier figure than a few dozen, and rare events need far more.
How is it different from theoretical probability?
Theoretical probability is derived from the structure of the situation, while empirical probability is counted from what actually happened.
Can empirical probability be zero?
Yes, but a zero count only means the event has not been seen yet in your sample, not that it cannot happen.
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%