A chart exists to support a decision, not to show that you have data. The difference is easy to spot: a chart that starts from a question reads in a few seconds, while one that starts from "let's see what comes out" needs an explanation. This guide starts from a table of one hundred products and shows how a question picks the chart. It then covers design principles, a test to check any chart and the most common mistakes.
Question first, chart second
The principle is simple: you choose the shape of a chart from the question, not from the data you happen to have. The same spreadsheet can produce ten different charts, and usually only one answers what you need to know.
An example with made-up numbers. You have a table of one hundred products with three columns: units sold in the quarter, units refunded and stock on hand. In total, 12,000 units sold and 720 refunded: an average refund rate of 6.0%. You can scroll a table like that for hours and see nothing. The question is: which products sell a lot but come back often?
There are two measures per product (how much it sells, how much comes back) and you are looking for outliers. That is the job of a scatter chart: one dot per product, sales on the horizontal axis, refund rate on the vertical axis. You draw two lines, one at the average sales per product (120 units, that is 12,000 ÷ 100) and one at the average rate (6.0%). The top-right box holds the products that sell more than average and come back more than average.
In the example there are eight. Together they account for 3,200 units (more than a quarter of the total) and 400 refunds: a rate of 12.5% (400 ÷ 3,200). The other 92 products account for 8,800 units and 320 refunds: 3.6%. So eight products out of a hundred generate 56% of the refunds (400 out of 720). If a refunded unit is worth €40 on average, that is €16,000 returned in the quarter (400 × 40). The table did not show this. The chart shows it in five seconds.
From the chart you move to five actions:
- Read the refund reasons for the eight products. If reasons are not recorded, start recording them.
- Check the product pages: sizes, photos, description. A return often starts with a wrong expectation.
- Look at stock. If some of the eight carry a lot of inventory, pause reorders until the cause is clear.
- Take the eight out of promotions until the refund rate comes down.
- Repeat the chart every month with the same lines and see who enters and who leaves the box.
Two cautions. The lines are your own catalog's averages, not thresholds that apply to everyone. And a product with 3 units sold and 1 refunded has a 33% rate that means nothing: set a minimum volume below which a product stays out of the chart.
Which chart for which question
This table pairs each chart with the question it answers best and the case where it is better left alone.
| Chart | Answers | Don't use it when |
|---|---|---|
| Lines | How a value changes over time | You have very few periods, or categories with no time order. Beyond four or five lines it becomes unreadable |
| Sorted bars | Which category is higher or lower | There are very many categories and no cut-off on the top ones. Show the top ten and the rest in a single bar |
| Stacked bar | How a total is made up of parts | You want to compare the parts across bars: only the bottom part starts from the same line, the others are hard to read |
| Scatter | The link between two measures, and outliers | The audience does not know it and there is no time to explain it, or there are very few points |
| Table | Exact values and looking up one item | You need to see a trend or a comparison at a glance |
| Big number | A single value to keep an eye on | There is no comparison: an isolated number does not say whether it is good or bad |
A simple rule for choosing: write the question in one sentence. If it contains "over time", think of a line. If it contains "which" or "how much", think of sorted bars. If it contains "made up of", think of a stacked bar. If it contains "link" or "outliers", think of a scatter.
Four design principles
Each principle comes with a before and an after.
- The title informs. Before: the title says "Sales by month". After: "Sales have fallen for three months, mostly among new customers". The first describes what is in the chart, the second says what to conclude. Use a title with a conclusion only if the data supports it.
- Color has a job. Before: ten categories in ten colors, and the eye does not know where to go. After: everything in gray except the category you are talking about, in a single color. Color tells people where to look. It is not decoration.
- The order is decided. Before: bars in alphabetical order, so finding the tallest means comparing them one by one. After: sorted from highest to lowest. The exception is months and other natural sequences, which keep their order.
- The axis does not mislead. Before: the bar axis starts at 9,500, and two bars of 9,800 and 10,000 look like one is twice the other. After: the axis starts at zero and the real difference shows, about 2% (10,000 ÷ 9,800). For lines, an axis that does not start at zero can be useful, but say so on the chart.
A test to check a chart
Before sharing a chart, put it through four checks. It is close to a template you can print.
- Five seconds. Show it to someone for five seconds, then hide it and ask what they understood. If they cannot say, the chart is doing too much.
- Decision. Write down which decision changes if the number goes up or down. If you cannot write it, the chart is information, not a tool.
- Doubt. Listen to the first question you get. If it is "does this include refunds?", the definition of the number is missing, and it belongs next to the chart.
- Time. Measure how long it takes the user to get from chart to action. If understanding it takes a whole meeting, simplify it.
Common mistakes
- One chart for everyone. Management, product owners and buyers have different questions. A page that tries to answer all of them serves none.
- Two vertical axes on one chart. Each scale is chosen freely, and two lines can look related only because one was fitted to the other.
- The average alone. An average hides the extremes. In the example the 6.0% average hid eight products at 12.5%.
- Too many elements. Twenty slices in a pie or fifteen lines in a chart cannot be read. Group the rest as "other".
- Periods that cannot be compared. A 28-day month against a 31-day one, or a promotion week against a normal week, gives a false comparison. Always state what you are comparing.
- A handsome chart with no question. Without a question behind it, the most polished chart is still decoration.
Why visualization works better when it is custom
Roles in a company make different decisions, so they need different views. Management wants a few big numbers with a comparison. Whoever follows products wants the scatter from the example. Whoever runs the warehouse wants stock next to sales. With custom work you start from each role's decision, pick the right chart, build on the company's own data and write the definitions once, so everyone reads the same number.
It is not always needed. A ready-made tool is enough when the questions are the standard ones (sales, orders, customers), there is a single source and the people reading the numbers have similar roles. In that case building custom is extra work that does not change the decisions. Custom pays off when the questions are specific to how you work, there is more than one source and the roles reading the data are different.
Where Opalca Lab fits
Opalca Lab is the part of Opalca that works on custom projects: data source integration, custom API connectors and databases, custom dashboards, data visualization and data modeling. There is no standard price list: each project is assessed on its own case. For visualization we start from the questions and the roles, as in the example in this guide, and only then choose the chart. Among the cases, Turnover, an agency specializing in selling on Amazon, has custom Amazon reporting dashboards built with Opalca Lab.
Do you have an idea of what you would like to see in your data? Tell us about it, or contact us.
Keep reading
- Ecommerce dashboard examples: five templates to start from
- Data integration: what you need before building a dashboard
Frequently asked questions
What is the best chart for comparing products?
It depends on the question. To see which product sells the most, sorted bars. To see which sell a lot and come back often, a scatter with two measures per product.
Is a table or a chart better?
A table is for looking up an exact value, a chart for seeing a trend or a comparison. Often the right choice is the chart to see where to look, with the table beside it for the details.
How many charts should a dashboard have?
As many as it takes to answer one question per role. If reading a page takes more than a short glance, there are too many elements or too many questions mixed together.
Are pie charts always wrong?
No, they work with two or three very different parts. With many parts, sorted bars read better, because the eye compares lengths better than angles.
When does it make sense to build visualizations custom?
When the questions are specific, there is more than one source and different roles need to read the same data through different views. With standard questions and a single source, a ready-made tool is usually enough.