Choosing Between the Mean and Median for Business Measures

Hello. In the previous lesson, you learned that an average must respect the amount of activity represented by each row: for example, overall average selling price is total revenue divided by total units, not a simple average of category prices. This lesson adds a different decision: once you have the correct observations, should you summarise them with the mean or the median?

Both describe the centre of numerical data, but they answer slightly different business questions. By the end, you should be able to inspect a business measure, recognise when extreme values distort the mean, and clearly justify which summary best serves the decision.

---

Two ways to describe the centre

The mean is the arithmetic average:

$$\text{Mean}=\frac{x_1+x_2+\cdots +x_n}{n}$$

It includes every observation’s value. If five customer orders are worth $40$, $50$, $60$, $70$, and $80$, then:

$$\text{Mean order value}=\frac{40+50+60+70+80}{5}=60$$

The median is the middle value after sorting observations from smallest to largest. It divides the data so that roughly half of observations are below it and half are above it.

For those same five orders, the middle value is $60$, so the median is also $60$.

With an even number of observations, the median is the average of the two central values. For example, in the sorted list $40,50,60,70$:

$$\text{Median}=\frac{50+60}{2}=55$$

The key distinction is not merely how they are calculated:

• The mean is sensitive: changing any value changes the mean.
• The median is resistant, or robust: a very large or very small value usually has little effect on it.

A helpful interpretation is that the mean is the data’s balance point, whereas the median is its 50th-percentile position.

---

Why outliers change the choice

An outlier is an observation that is unusually far from most of the data. It might be a genuine but rare event, such as a large enterprise order, or it might be a data-quality issue, such as an extra zero in a transaction amount. Analysts must investigate rather than assume it should be removed.

Imagine ten delivery times, measured in days:

$$1,1,2,2,2,2,3,3,3,20$$

The mean is:

$$\frac{1+1+2+2+2+2+3+3+3+20}{10}=3.9$$

The median is:

$$\frac{2+2}{2}=2$$

If the question is, “What delivery time does a typical customer experience?”, reporting 3.9 days can be misleading. Nine of ten deliveries took three days or less; one 20-day delivery pulls the mean upward. The median of 2 days better captures the typical experience.

But the 20-day delivery still matters. If operations must plan total delivery effort, assess service failures, or estimate the average cost per shipment, the mean can be relevant precisely because it includes that costly exception. Choosing the median does not mean ignoring extreme cases.

The right question is:

Do we need a summary of a typical observation, or an average that includes the full numerical impact of every observation?

---

Distribution shape: symmetric, right-skewed, and left-skewed

A distribution is symmetric when values are balanced around the centre. In a roughly symmetric set of order values or quality scores, the mean and median are close. Either may be an acceptable central summary, though the mean is often convenient for further calculations.

A distribution is right-skewed when most values are lower or moderate, while a smaller number of high values create a long tail on the right. Revenue per customer, property values, salaries, and online order values commonly have this shape.

In the histogram, the mean is larger than the median because unusually high observations contribute heavily to the total used in the mean calculation. The median remains near the centre of the large cluster of ordinary observations.

A left-skewed distribution has a long tail of unusually low values. For example, a satisfaction metric may cluster near its maximum, with a few very dissatisfied customers pulling the mean downward.

Use this practical pattern as an initial diagnostic:

Shape of distribution	Typical relationship	Initial reporting choice
Approximately symmetric, no influential outliers	Mean and median are similar	Mean is often suitable
Right-skewed	Mean is greater than median	Median often better for “typical” experience
Left-skewed	Mean is less than median	Median often better for “typical” experience
Clearly multi-peaked	Either can conceal distinct groups	Segment the data before relying on one centre

“Initial” matters. Distribution shape guides the decision, but the business question decides it.

---

Mean versus median in business reporting

Consider an e-commerce business with these customer order values:

Customer	Order value
A	500
B	650
C	700
D	750
E	900
F	1,100
G	18,000

The mean order value is:

$$\frac{500+650+700+750+900+1,100+18,000}{7}=3,228.57$$

The median order value is $750$.

Both values are correct, but they communicate different realities:

• Mean order value: 3,228.57. Across these seven orders, this is the revenue per order on average. It is relevant for forecasting revenue when this mix of regular and enterprise orders is expected to recur.
• Median order value: 750. This is closer to the value of a typical customer order. It is more useful for describing ordinary customer purchasing behaviour.

A weak report says:

Average order value was 3,228.57.

A stronger report makes the distribution and decision implication visible:

Mean order value was 3,228.57, inflated by one 18,000 enterprise order. Median order value was 750, which better represents a typical order. Revenue planning should retain the mean, while customer-experience reporting should use the median and separately monitor enterprise orders.

This is a core analyst habit: do not report a central measure without considering what it obscures.

Common business choices

Business measure and question	Usually prefer	Why
“What is the typical customer’s order value?”	Median	Large orders can make the mean unrepresentative of ordinary orders.
“What revenue per order should we expect across all orders?”	Mean	Total revenue divided by total orders needs every order value.
“What is a typical employee salary?”	Median	A few executive salaries can pull the mean far above most salaries.
“What is average payroll cost per employee?”	Mean	The full payroll budget includes all compensation amounts.
“What is the typical home price in an area?”	Median	Housing prices are often right-skewed.
“How long does a typical delivery take?”	Median, plus a high percentile if available	Late deliveries may distort the mean; the median describes the middle experience.
“What is the average handling time for staffing forecasts?”	Mean, checked alongside the median	Long cases consume real staff time and affect required capacity.

The last example shows why a simplistic rule such as “outliers mean median” is insufficient. Long support calls may be rare, but they still consume staff time. For staffing, the mean answers an operational workload question. For describing what a normal caller experiences, the median is more informative.

---

A decision process you can use

When deciding between the mean and median, work through these checks in order.

1. Confirm the measure and observation grain.
   Are you summarising one row per transaction, customer, employee, product, or day? If the data are already grouped, revisit the previous lesson: a mean of group-level means may require weighting.

2. State the business question precisely.
   Words such as typical, ordinary, middle, and customer experience usually point toward the median. Questions about total cost, expected amount, resource requirements, or revenue per observation often need the mean.

3. Inspect the distribution before selecting a single number.
   Sort the values, calculate both measures, and use a histogram or box plot when possible. A large gap between mean and median is a signal to investigate.

4. Check extreme values.
   Ask whether they are valid, incorrectly entered, duplicated, or caused by a different business process. Do not silently delete valid observations merely to produce a cleaner average.

5. Choose, justify, and disclose.
   Name the measure, population, period, and reason for the choice. If skew is meaningful, report both values.

In Excel, the basic formulas are straightforward:

=AVERAGE(B2:B101)

=MEDIAN(B2:B101)

The difficult part is interpretation. Excel can calculate either statistic instantly, but it cannot decide whether an 18,000 order is a data error, a legitimate enterprise sale, or an observation that needs separate analysis.

---

Do not confuse weighted means with medians

The previous lesson’s weighted average remains important here. A weighted mean and a median solve different problems:

• A weighted mean ensures observations or group summaries influence the average in proportion to their underlying volume.
• A median identifies the middle observation and reduces the influence of extreme magnitudes.

For example, suppose a dashboard has one row per product category, with category-level average order values. You should not take the median of those category averages to describe the typical order. Each category row represents a different number of orders, and the median would give every category equal influence.

To report a typical order value:

• Use order-level data and calculate the median across orders.
• If only aggregated data are available, be explicit that you cannot recover the exact order-level median from category averages alone.

To report overall revenue per order:

• Use total revenue divided by total orders, which is a weighted mean when the data are grouped.

This distinction protects you from a common reporting error: treating a summary-table row as though it were an individual customer or transaction.

---

Reporting with appropriate caveats

A good analytical statement contains the result and the interpretation.

Mean chosen

Average handling time was 8.4 minutes across 12,480 calls in April. This mean includes long calls and is appropriate for estimating total agent workload. The median was 6.1 minutes, indicating that a smaller number of complex calls increased the average.

Median chosen

Median order value was 750 in April, meaning half of orders were below this amount and half above it. The mean was 3,228.57 because a small number of enterprise orders were substantially larger; it should not be interpreted as the typical customer order.

Avoid these statements:

• “The median is more accurate.”
  It may be more representative for a particular question, but both calculations can be accurate.

• “Outliers should be removed because they affect the mean.”
  First determine whether they are errors, legitimate exceptions, or a separate segment.

• “The mean is always better because it uses all the data.”
  Including every value is valuable for some questions, but it can misrepresent the typical case in skewed data.

---

Key takeaways

The mean uses every observation and is sensitive to extremes. It is often appropriate when the decision depends on total impact, expected amount, workload, revenue, or cost across all observations.

The median is the middle sorted value and is resistant to extreme values. It is often the better choice for communicating a typical customer, employee, transaction, or delivery experience in skewed data.

Before reporting either one:

• confirm the observation grain;
• inspect the data distribution and outliers;
• distinguish valid rare events from data-quality errors;
• match the statistic to the business decision;
• report both mean and median when their difference tells an important story.

Next, you will build on this idea by linking a business objective to measurable drivers in a KPI tree.