What descriptive statistics do you typically include in a business performance report, and how do you decide which measures of central tendency and spread to use?

Question

Accepted Answer

Choosing the right descriptive statistics for a business report depends on the nature of the data and the audience. The wrong choice can paint a misleading picture, so understanding when to use each measure is a core BI skill.

For measures of central tendency, I choose between mean, median, and mode based on the data distribution. The mean works well for normally distributed data without extreme outliers, such as average daily website sessions over a stable month. The median is better for skewed data, which is extremely common in business. Customer lifetime value, order values, and employee salaries are all typically right-skewed, meaning a few very large values pull the mean upward and misrepresent the typical experience. Reporting median order value gives a more accurate picture of what a typical customer spends. The mode is useful for categorical data, like identifying the most popular product category or the most common support ticket type.

For measures of spread, I include standard deviation or interquartile range depending on the same distributional considerations. Standard deviation pairs with the mean for normally distributed data. Interquartile range pairs with the median for skewed data and gives stakeholders a clear picture of where the middle fifty percent of values fall. I often present the IQR as a range statement, such as "most customers spend between forty-five and one hundred twenty dollars per order," which is immediately actionable.

Beyond these basics, I include percentiles when the tails of the distribution matter. For example, in operational metrics like page load time, the P95 and P99 values reveal the experience of your worst-served customers, which is often more important than the average. Revenue analysis benefits from showing the contribution of the top ten percent of customers.

I also include counts and completeness statistics. Knowing the sample size adds credibility, and documenting how much missing data exists prevents stakeholders from over-trusting incomplete datasets.

Visualizations accompany the statistics. A histogram or box plot shows the distribution shape, making it immediately clear why I chose the median over the mean. I find that pairing the right statistic with the right visualization eliminates most questions about methodology.

In practice, I build a standard template for recurring reports that uses the appropriate statistics for each metric based on its known distribution. This creates consistency and saves time while ensuring the reports remain statistically sound.