An average order value of 75 does not tell a merchandiser whether most customers spend around 75 or whether small baskets coexist with a few very large purchases. A histogram can reveal that difference. But a chart can also invent a persuasive-looking pattern when its bins are selected carelessly.
EcomToolkit recommends designing the distribution chart around a specific trading question. This guide uses a small, hypothetical order dataset to show how boundaries, widths and denominators affect the story. It is a method for examining your own sales, not a claim about typical basket values across ecommerce platforms.
Table of Contents
- Define the order value before drawing the chart
- Make every boundary unambiguous
- Test the conclusion against another reasonable width
- Do not confuse frequency with revenue contribution
- Treat unequal widths carefully
- Compare periods without moving the goalposts
- The EcomToolkit view
Define the order value before drawing the chart
Start with one row per eligible order and one amount definition. Decide whether that amount includes discounts, delivery charges and tax, and how refunds affect the observation. A histogram of original checkout totals answers a different question from one showing the latest net amount after returns.
Choose a reporting currency and document the conversion date when orders originate in different currencies. Otherwise the chart can show artificial clusters caused by unit differences. Also separate genuine zero-value orders from missing amounts; replacing missing data with zero creates a bar that represents a processing decision rather than customer behaviour.
Use a closed, named period and record the inclusion policy. A report of paid orders should not silently include draft orders or failed checkouts because those records happen to carry an amount. Before analysing shape, reconcile the eligible order count and total with an independent control report.
The NIST histogram guide explains the basic role of a histogram in showing a numerical distribution. The ecommerce decision comes afterwards: are you investigating basket concentration, a delivery threshold, product mix, or the contribution of unusually large purchases?
Make every boundary unambiguous
Consider ten illustrative order values: 20, 25, 30, 45, 49, 50, 51, 70, 95 and 165. Their total is 600 and their mean is 60. The value 50 sits exactly where many teams would place a boundary, making this a useful test of the chart’s assignment rule.
Use intervals that include the lower bound and exclude the upper bound, then explicitly handle the final endpoint. A label such as 0–50 is ambiguous unless the chart explains whether 50 belongs there or in the following bin. The same amount must never appear twice or disappear between conditions.
| Order value interval | Count | Share of ten orders | Amount total |
|---|---|---|---|
| At least 0, below 50 | 5 | 50% | 169 |
| At least 50, below 100 | 4 | 40% | 266 |
| At least 100, below 150 | 0 | 0% | 0 |
| At least 150, below 200 | 1 | 10% | 165 |
| Total | 10 | 100% | 600 |
These bins reveal that half the orders are below 50, while one order occupies a distant upper interval. They do not establish separate customer segments. The sample is deliberately tiny, and a commercial decision would require a larger, relevant period and additional context.
Test the conclusion against another reasonable width
A very wide bin may hide a cluster around a free-delivery threshold. A very narrow bin can turn random variation into a series of apparent peaks. There is no universally correct width independent of the sample size, amount range and business question.
For the ten-order fixture, changing to intervals of 25 creates a different silhouette without changing a single purchase. That is a useful reminder to inspect the raw data and summary counts when a chart seems to reveal a sudden customer behaviour pattern. The picture is partly a consequence of its construction.
Choose a primary bin scheme before comparing periods, then test one narrower and one wider scheme as a sensitivity check. If a proposed action depends entirely on a particular boundary choice, the evidence is fragile. Document that fragility instead of selecting the version that tells the most attractive story.
For operational thresholds, include a complementary count based on the actual business rule. A delivery offer applying at 50 needs an explicit at-least-50 indicator even if the exploratory histogram uses another width. Keep the policy measure stable while allowing analysts to explore the distribution separately.
Do not confuse frequency with revenue contribution
In the worked example, the five orders below 50 represent half the purchases but contribute 169 out of 600, or about 28.2% of the amount. The single order at 165 represents 10% of purchases and 27.5% of the amount. Both views are correct and answer different questions.
An order-frequency chart helps describe how commonly basket sizes occur. A contribution chart helps show where revenue is concentrated. Neither establishes margin, because discounts, product costs, fulfilment and returns can vary across the same value bands. Add contribution margin only when its definition has been reconciled.
Label the vertical axis explicitly: order count, percentage of orders, revenue amount or percentage of revenue. A bar titled simply share forces readers to guess its denominator. If a dashboard switches measures interactively, update the label and explanatory text with the selection.
Use our order-value outlier guide when an extreme purchase drives the display. Do not delete it just to make the chart neater. Investigate whether it is valid, a separate business segment or a data error, then explain any exclusion.
Treat unequal widths carefully
Commercial bands often use unequal intervals, such as below 50, 50 to below 100, and 100 to below 200. A simple categorical bar chart can compare the counts in those named groups. A continuous histogram with unequal horizontal widths requires more careful interpretation.
If bar area is intended to represent relative frequency, height must account for width. In the hypothetical bands below, both the first and third groups contain 20 orders, but their widths differ. Drawing the wider group at the same density height would imply more area than its count warrants.
| Hypothetical interval | Width | Orders in interval | Orders per unit of width |
|---|---|---|---|
| 0 to below 50 | 50 | 20 | 0.40 |
| 50 to below 100 | 50 | 40 | 0.80 |
| 100 to below 200 | 100 | 20 | 0.20 |
| Total | 200 across the displayed range | 80 | Density is not added as a count |
For a probability density, divide those final-column values by the total of 80 as well. The resulting areas sum to one. If that representation is unfamiliar to the audience, use equal-width bins or a clearly labelled categorical chart instead of a technically correct picture that readers routinely misinterpret.
An open-ended group such as 200 and above has no finite bin width for this calculation. Display it separately as a tail count or choose a justified upper plotting boundary with an explicit overflow category. Never make the tail vanish without telling readers how many orders it contains.
Compare periods without moving the goalposts
Freeze the bin edges when comparing weeks, customer groups or channels. Automatic binning can produce different edges for every selection, so a bar’s movement may reflect chart configuration rather than trading behaviour. Keep the same amount basis and eligibility rules alongside the same boundaries.
Use percentages for shape comparisons when groups have different numbers of orders, while retaining the underlying counts. A small segment can display a striking percentage change based on very few purchases. Counts help the reader decide whether further investigation is justified before changing merchandising.
Check composition next. A wholesale order entering a mostly direct-to-consumer sample can change the upper tail without any change among retail shoppers. Segment only where the business distinction is meaningful and sample size permits it; endless slices can create patterns that will not repeat.
Our guide to average order value uncertainty complements this distribution view. A histogram describes the observed shape. Uncertainty analysis asks how stable a summary or comparison may be under an appropriate sampling model. One cannot substitute automatically for the other.
The EcomToolkit view
A useful order-value histogram exposes choices instead of concealing them. Define the amount, show the boundaries, preserve tail observations and label the denominator. Then test whether the commercial conclusion survives a reasonable alternative display.
Bring a confusing basket-value chart to an EcomToolkit analytics review. The goal is a clear decision about customer behaviour, supported by reconciled data rather than the most dramatic arrangement of bars.
For a related practical question, read our guide to relative order-value spread and the coefficient of variation.
