The number that looked fine
I once pulled up average order value for a small store I run and it looked completely normal. Somewhere in the middle of what I expected, nothing alarming, nothing to act on. I almost closed the tab.
Then I sorted the same orders by size instead of averaging them, and the picture changed. A handful of customers were buying two or three times a typical order, and a much larger group was buying barely anything at all. The average sat in the middle of those two groups like it was describing a customer who did not actually exist. Nobody in my order history looked like the average. The average was a compromise between two very different kinds of buyers, and it hid both of them.
That is the problem with an average. It compresses a distribution, which is just the technical name for how your values are spread out, into one number. Compression is useful when the underlying values are roughly similar. It is misleading when they are not, and in small businesses they usually are not.
What an average actually does
An average, or mean, is a sum divided by a count. Add up every customer’s spend, divide by how many customers there are, and you get one figure that represents all of them. That is the entire calculation. It does not know or care whether your customers are clustered close together or scattered far apart.
Here is the part that matters: a mean is sensitive to extreme values. If you have 50 customers who each spend about $40, and one customer who spends $4,000 in a year, your average customer value jumps noticeably even though 50 of your 51 customers are nowhere near that number. That one high spender is doing more to move your average than every other customer combined. This is not a flaw in arithmetic, it is just what averaging does with skewed data, meaning data where values are not spread evenly on both sides of the middle.
Business data is skewed constantly. Revenue per customer, time on page, days between purchases, support ticket resolution time, refund amounts. A small number of accounts or events tend to sit far out on one end, and that tail drags the average with it. If you only look at the average, you are looking at a number shaped by a few outliers, not a number that describes your typical customer.
The distribution is the real picture
A distribution just means the full spread of individual values, not one summary of them. Instead of “average order value is $85,” a distribution shows you that 30 orders were between $20 and $40, 15 were between $60 and $90, and 5 were over $300. That is a very different starting point for decisions than a single $85 figure.
You do not need statistical software to see a distribution. A simple histogram in a spreadsheet, which just groups your values into buckets and counts how many fall in each one, will usually show you the shape immediately. I do this with a pivot table: bucket order values into ranges, count orders per range, and look at the shape. Is it one hump in the middle, which statisticians call roughly normal? Or is it lopsided, with a long tail stretching toward the high end? Most customer spending data I have looked at across different businesses leans toward the second shape.
Once you see the shape, the average stops being a mystery number and starts being a data point you can interpret correctly. If the average is $85 but most orders cluster around $40, you now know the average is being pulled up by a smaller group of bigger spenders. That tells you something actionable. The average alone told you nothing you could act on.
Median and percentiles fill the gap
Two other numbers help when an average is doing too much on its own.
The median is the middle value when everything is sorted from smallest to largest. Half your customers spend more, half spend less. Unlike the mean, one enormous outlier does not drag the median around, because the median only cares about position, not magnitude. If your mean order value is $85 and your median is $42, that gap is telling you the mean is being inflated by a tail of high spenders. That gap itself is a useful signal, and it costs nothing to check since both numbers come from the same data you already have.
Percentiles go a step further and tell you where a specific point sits in the whole spread. Your 90th percentile customer is spending more than 90 percent of your customers and less than the top 10 percent. Looking at your 10th, 50th, and 90th percentiles together gives you a rough shape of the distribution in three numbers, without building a full histogram. I use this when I want a quick gut check on customer lifetime value, which is an estimate of what a customer is worth over the time they stay with you. A CLV average can hide the fact that your top 10 percent of customers are worth many times more than your bottom half. If that gap is wide, the business decisions for retaining your top decile look nothing like the decisions for your median customer, and an average would have quietly merged those two very different problems into one number.
Where this actually changes a decision
Say you are deciding whether to raise a shipping threshold for free delivery. You check average order value, see it comfortably above your proposed threshold, and figure most customers already qualify. But if the distribution is skewed by a few large orders, the median might sit well below that threshold, meaning most individual customers would not actually qualify under the new rule. Checking the median next to the mean here takes a couple of minutes and can change what you decide, because it tells you whether you are describing your typical customer or a handful of unusual ones.
Same logic applies to churn analysis. An average subscription length can look healthy while hiding a distribution with two peaks, a group that cancels almost immediately and a separate group that stays for years. That shape, sometimes called bimodal because it has two humps instead of one, means you actually have two different customer behaviors bundled into one average. Treating them as one group with one average retention tactic misses both.
None of this means averages are useless. They are a fine starting point and a compact way to track a trend over time, like watching whether average order value moves up or down month over month. The mistake is treating the average as a full description of your customers rather than a summary that can hide meaningful splits underneath it. A summary is not a substitute for looking at the actual spread, and no single number, however you calculate it, can tell you everything a distribution can show at a glance.
A simple habit worth building
Whenever a business metric is reported as a single average, whether it is CLV, order value, session length, or support response time, ask what the median and the top and bottom deciles look like. If they are close to the average, the average is doing its job and you can move on. If they are far apart, the average was hiding something, and that gap is worth a closer look before you act on the average alone. This does not require new tools in most small businesses. A spreadsheet, a sort, and a few minutes are usually enough to check.
The numbers will not make the decision for you either way. What they do is make sure the decision is based on how your customers actually behave, not on a single figure that happens to sit between two very different groups.
If you want more explainers like this on the metrics and methods behind running a small business on your own numbers, you can find the rest of them on the home page.
Get new guides and videos first — join the Telegram channel.