What it means
The curve is symmetrical, with its peak at the mean and tails that thin out on both sides. Shift the mean and the whole curve slides sideways; increase the standard deviation and it flattens and widens, without changing where the centre sits.
The useful shortcut is the 68-95-99.7 rule. About 68% of values fall within one standard deviation of the mean, about 95% within two and about 99.7% within three, which lets you convert a spread into a plain-English range in seconds.
Businesses use it to turn an average into a planning range. A sales forecast, a call centre staffing model or a manufacturing tolerance all become answerable questions once you have a mean and a standard deviation to work with.
The important caveat is that real financial data often has fatter tails than the curve predicts, meaning extreme events happen more often than the model says. Risk models that assumed normality have repeatedly understated the chance of a severe market move, which is why stress testing sits alongside statistical models rather than inside them.
The other misuse is human rather than mathematical. Forcing staff performance ratings into a fixed distribution assumes ability in a small team follows the same shape as a population of thousands, which it usually does not.
In practice
Real-world examples.
Example
A fastener factory machines bolts to a mean diameter of 10.00 mm with a standard deviation of 0.02 mm, against a specification of 9.95 mm to 10.05 mm. The limits sit 2.5 standard deviations out, so just over 1% of output fails; tightening the process to a 0.015 mm standard deviation pushes the limits to 3.3 standard deviations and cuts rejects below 0.1%.
Example
An operations director staffs a contact centre that handles an average of 900 calls a day with a standard deviation of 120. Rostering for two standard deviations means covering up to 1,140 calls, which handles roughly 97 days out of every 100 without overtime.
Example
A company applies a forced rating distribution of 10% top, 70% middle and 20% bottom to a 40-person engineering team, producing 4, 28 and 8 people respectively. Two strong engineers are rated in the bottom group purely because the quota required it, and both resign within the year.
Formula
Calculation
Range covered by k standard deviations = mean +/- (k x standard deviation)
Z-score = (value - mean) / standard deviation
A regional distributor has averaged $50,000 of monthly sales over several years, with a standard deviation of $8,000.
One standard deviation: $50,000 +/- $8,000, giving $42,000 to $58,000, covering about 68% of months
Two standard deviations: $50,000 +/- $16,000, giving $34,000 to $66,000, covering about 95% of months
Three standard deviations: $50,000 +/- $24,000, giving $26,000 to $74,000, covering about 99.7% of months
Now test a specific month. Sales of $66,000 give a z-score of ($66,000 - $50,000) / $8,000 = 2.0, so under the rule of thumb roughly 2.5% of months should be that strong or stronger, or about one month in 40. That is a good month worth understanding, not evidence that the business has permanently changed gear.Case study
Seen in the real world.
Kelby Instruments is a fictional maker of laboratory scales that sells about 60,000 units a year with a two-year warranty. Historic data showed a mean return rate of 2.4%, which is 1,440 units, with a standard deviation of 300 units and an average repair cost of $85.
The central provision came to 1,440 x $85 = $122,400. Rather than book the mean, the board provisioned at two standard deviations, or 1,440 + 600 = 2,040 units, giving $173,400 and a $51,000 cushion above the central estimate. For three years running, actual returns fell comfortably inside the range.
In year four a single supplier shipped a faulty load cell, and returns reached 4,100 units at a cost of $348,500. That was far outside even three standard deviations, which would have been 2,340 units. The illustrative lesson is that a bell curve models routine variation around a stable process, and says nothing useful about a one-off systematic fault.
Watch out
Common mistakes.
- Assuming any dataset with a hump in the middle is normally distributed, without checking whether it is symmetrical or how heavy the tails are.
- Applying the 68-95-99.7 rule to strongly skewed data such as salaries, insurance claims or house prices, where a handful of very large values pull the average away from the typical case.
- Forcing performance ratings into a fixed curve in a small team, which manufactures underperformers who do not exist.
Questions
People also ask.
Which two numbers define a bell curve?
The mean, which fixes where the peak sits, and the standard deviation, which fixes how wide and flat the curve is.
Are share price returns normally distributed?
Not exactly, because real markets produce large moves more often than the curve predicts, which is why models built on it need stress testing alongside.
What is a standard deviation in plain terms?
It is the typical distance between an individual value and the average, expressed in the same units as the data itself.
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%