What it means
Almost every business number you see is an estimate built from a sample: a survey of 400 customers, one month of transactions, a pilot in three stores. A confidence interval attaches an error bar to that estimate so decisions get made with the uncertainty visible rather than hidden behind a tidy-looking average.
The phrase "95% confidence" has a specific meaning that is easy to get wrong. It describes the method, not the individual answer: if you repeated the same sampling exercise many times, roughly 95% of the intervals produced would contain the true underlying value.
Three things drive how wide the interval is. Variability in the data widens it, a larger sample narrows it, and demanding 99% rather than 95% confidence widens it again.
The sample size effect is the one that matters commercially, because precision improves only with the square root of the sample. Quadrupling the number of respondents merely halves the width of the interval, which is why research budgets hit diminishing returns quickly.
In practice the most valuable use is comparison. If the interval around a test group's conversion rate overlaps heavily with the interval around the control group, the difference everyone is excited about may be nothing more than noise.
In practice
Real-world examples
Example 1
A subscription business surveys 900 customers on willingness to pay and finds an average acceptable price of $29.00, with a 95% confidence interval of $27.40 to $30.60. The pricing team drops a proposal to charge $32.00 because it sits outside the range the data supports.
Example 2
A retailer pilots a new store layout in six branches and sees sales up 4%, with a confidence interval running from -1% to 9%. Because the interval includes zero, the board funds a larger pilot rather than a national rollout.
Example 3
An insurer estimates average claim cost at $3,200 with a confidence interval of $3,050 to $3,350. The actuary prices the product using the upper end of the range so that a bad draw from the distribution does not wipe out the underwriting margin.
Think of it
Imagine throwing darts at a board while wearing a blindfold. Instead of claiming you will hit the exact bullseye every time, you draw a circle around the target area. You are confident that your throws will land inside that circle.
Formula
Calculation
For an average, the confidence interval is: Sample mean +/- (z-score x Standard error), where Standard error = Standard deviation / square root of the sample size. For 95% confidence the z-score is 1.96.
An online retailer samples 400 orders. The average order value is $50.00 and the standard deviation is $12.00.
Standard error = $12.00 / square root of 400 = $12.00 / 20 = $0.60.
Margin of error = 1.96 x $0.60 = $1.176, which rounds to $1.18.
Confidence interval = $50.00 - $1.18 to $50.00 + $1.18 = $48.82 to $51.18.
The retailer can therefore say, with 95% confidence, that true average order value sits between $48.82 and $51.18. Applied to a forecast of 100,000 orders, that is revenue of $4,882,000 to $5,118,000 rather than a single false-precision figure of $5,000,000.Case study
Seen in the real world
This is an illustrative example using a fictional business. Tallow Lane Coffee, a chain of eighteen cafes, tested a loyalty app in four sites for one month. Average spend per visit rose from $8.20 to $8.60, and the operations director immediately proposed a chain-wide rollout costing $340,000.
The finance analyst ran the numbers properly. With only 1,600 sampled visits and wide variation in basket size, the 95% confidence interval around the $0.40 uplift ran from -$0.05 to $0.85. In other words, the data could not rule out that the app made no difference at all.
Tallow Lane extended the pilot to twelve sites for three months instead, collecting around 15,000 visits. The uplift settled at $0.30 with an interval of $0.22 to $0.38, comfortably above zero, and the rollout went ahead on evidence rather than enthusiasm.
Watch out
Common mistakes
- Reading a 95% interval as "there is a 95% chance the true value is in this range". The confidence refers to the reliability of the procedure across repeated samples, not to a probability about this one interval.
- Ignoring the interval once it has been calculated and quoting only the midpoint in the board pack. The whole purpose is to carry the uncertainty into the decision.
- Assuming a bigger sample always fixes precision cheaply. Because width falls with the square root of sample size, halving the interval requires four times the data.
Questions
People also ask
What does the interval assume about my data?
Standard formulas assume the sample is random and reasonably representative; a biased sample produces a precise-looking interval around the wrong number.
Should I use 95% or 99%?
Use 95% for most commercial decisions and 99% where the cost of being wrong is severe, such as safety, regulatory or large capital commitments.
How does this relate to statistical significance?
They are two views of the same information: if a confidence interval for a difference excludes zero, that difference is statistically significant at the matching level.