What it means
The rule gives you a fast sense of what is normal variation and what is genuinely unusual, without any statistical software. Once you know an average and a standard deviation, you can immediately sketch the range that should contain most outcomes.
In business this turns into practical thresholds. If daily sales, production defects or call volumes fall outside two standard deviations from the mean, that is roughly a one-in-twenty event and usually worth investigating rather than ignoring as noise.
The rule underpins a lot of everyday management tooling. Statistical process control charts in manufacturing, anomaly alerts in finance systems and outlier flags in audit sampling all use two or three standard deviation bands drawn straight from this idea.
The critical condition is that the data must be roughly bell-shaped and symmetrical. Many financial series are not: share returns have fatter tails than a normal distribution, so events the rule would call one-in-a-thousand happen considerably more often, which is exactly the failure that catches out badly built risk models.
Skewed data breaks the rule in a different way. Customer order values or insurance claims often have a long tail on the high side, so the mean sits above the typical value and the neat symmetrical bands simply do not describe reality.
In practice
Real-world examples.
Example
A contact centre manager finds that daily call volume averages 1,200 with a standard deviation of 150. She staffs to handle 1,500 calls, two standard deviations above the mean, knowing this covers roughly 97 or 98 days out of 100 on the upside.
Example
A food manufacturer fills jars to an average of 500 grams with a standard deviation of 4 grams. The quality team sets control limits at 488 and 512 grams, three standard deviations either side, and treats any reading outside as a machine fault rather than random variation.
Example
An internal audit team reviewing expense claims calculates a mean claim of $340 with a standard deviation of $90. Any claim above $610, more than three standard deviations out, is automatically pulled for manual review.
Formula
Calculation
For data that is approximately normally distributed, with mean m and standard deviation s:
About 68% of values fall between m - s and m + s
About 95% of values fall between m - 2s and m + 2s
About 99.7% of values fall between m - 3s and m + 3s
A retailer's daily sales average $50,000 with a standard deviation of $6,000, and daily sales are approximately bell-shaped.
One standard deviation: $50,000 - $6,000 = $44,000 to $50,000 + $6,000 = $56,000. About 68% of days, or roughly 5 days in every 7, fall in this range.
Two standard deviations: $50,000 - $12,000 = $38,000 to $50,000 + $12,000 = $62,000. About 95% of days fall here, so only about 1 day in 20 lands outside it.
Three standard deviations: $50,000 - $18,000 = $32,000 to $50,000 + $18,000 = $68,000. About 99.7% of days fall inside, so a day outside this band happens roughly 1 time in 300.
A day with sales of $70,000 sits ($70,000 - $50,000) / $6,000 = about 3.3 standard deviations above the mean, which is unusual enough to look for a specific cause such as a promotion or a bulk order.Case study
Seen in the real world.
This is an illustrative, fictional example. Kelso Precision Components made machined parts to a tolerance specification and had been inspecting every hundredth unit by hand, a slow process that still let occasional bad batches reach customers.
A new quality engineer measured 2,000 consecutive parts and found the critical dimension averaged 40.00 millimetres with a standard deviation of 0.02 millimetres, in a clean bell-shaped pattern. Applying the empirical rule, she set control limits at 39.94 and 40.06 millimetres, three standard deviations either side, and had the measuring gauge alert an operator whenever a reading crossed those lines.
Within four months, scrap fell by about a third, because drifting machines were caught in hours rather than at the end of a production run. The engineer was careful to note in her report that the approach depended on the measurements staying bell-shaped, and she scheduled a check of that assumption every quarter.
Watch out
Common mistakes.
- Applying the rule to data that is clearly skewed, such as customer order values or insurance claim sizes, where the symmetrical bands do not describe reality.
- Treating a value three standard deviations out as impossible rather than merely rare, which is how financial models underestimate tail risk.
- Calculating the mean and standard deviation from too few observations, so both figures are themselves unreliable before the rule is even applied.
Questions
People also ask.
How do I know whether my data is normally distributed enough?
Plot a histogram first; if it looks roughly symmetrical with a single peak and no long tail, the rule is a reasonable approximation.
Why do financial returns break the rule?
Market returns have fatter tails than a normal distribution, so extreme days occur far more often than 99.7% coverage implies.
What is the difference between the empirical rule and Chebyshev's inequality?
Chebyshev's inequality gives weaker bounds but works for any distribution, while the empirical rule gives tighter figures that only hold for roughly normal data.
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%