The idea
A visual is not decoration on top of a number. It is a claim about how to compare. A line says "these points are in order and the direction matters". Bars say "these things are separate and their length is what to compare". A card says "this one number is all you need". Pick a shape that makes a different claim from the one your data supports and the reader leaves with the wrong impression, however correct your measure is.
So the work happens before the Visualizations pane. Write down the question the reader would ask, name the comparison inside it, and let that decide the shape.
Figure 1
From a question to a visual
- The question What would the reader ask?
- The comparison Which two things must the eye set side by side?
- The shape of the data One number, categories, time, two categories
- The visual Card, bar, line, matrix
- The truth test Order, baseline, labels
- Start with the sentence the reader would say out loud, such as "is 2025 growing?" or "which category carries us?". The chart gallery comes last.
- Every question is a comparison: against last year, between categories, between months, or against a target. The visual exists to make that comparison easy.
- The comparison fixes the data shape. One number is a card. Categories are bars. Time is a line. Two categories at once are a matrix.
- Only now do you pick the visual. In Power BI Desktop it is the **Visualizations** pane, but the choice was already made.
- Finally test it for honesty: is it ordered by the thing that matters, does the baseline start at zero, can you read the labels at arm's length?
Why it exists
Harbor & Pine's managers open a report for a few seconds. They do not read a chart, they glance at it, and in that glance the eye recovers only a few things: which item is longest, which direction a line goes, which cell is darker. A visual works when the thing the reader needs is one of those things.
Without a method, charts get chosen by habit or taste: a pie because it looks friendly, a table because "all the numbers are in it". A table is correct and still slow, because the reader has to do the comparison in their head. The method turns the choice into a short checklist: question, comparison, shape, visual, truth test.
How it works under the hood
Match the shape of the question to the shape of the visual.
- One number (total sales this month): a card. If the reader needs to know whether the number is good, add a target and a trend and use the KPI visual, which needs three things: a value, a target and a time axis.
- Compare categories (sales per category): a bar or column chart, one bar per category, sorted by value.
- Follow time (sales per month): a line chart, in calendar order.
- Look up exact values (the sales of one product): a table, one row per item with the columns you want to read.
- Two categories at once (category by channel): a matrix, which is a table that groups rows and columns and shows subtotals.
The monthly sales show why time is a line.
Figure 2
Completed sales by month, 2025
- Time is the question, so time goes on the x axis, in calendar order. The line makes direction visible: a dip in February, a plateau through summer.
- The average line gives the eye something to measure against. Most months sit around it.
- November and December are far above it, about 50% over the average. This is the story, and it only shows because the months are in order.
And category sales show why order and baseline matter for bars.
Figure 3
Completed sales by category, 2025, sorted
- Categories have no natural order, so the order must come from the value. Sorted by sales, the ranking reads left to right without a legend.
- Furniture and Outdoor lead, and the columns start at zero, so their height is honest: Furniture is a little over twice Textiles, and it looks it.
- Sorted alphabetically, the same bars would hide the ranking, and a baseline cut at $200,000 would make Textiles look almost empty. Order and baseline are the two ways a chart lies by accident.
Table or matrix, card or KPI
Two choices trip people up because the visuals look alike. A table lists rows; a matrix puts one field on the rows and another on the columns and adds subtotals, so it answers "category by channel" in one block. Use a matrix when two things cross, and switch off the subtotals the reader never uses, because every extra total is another number to ignore.
A card shows one value and nothing else. The KPI visual shows the same value with a target and a small trend behind it, so it answers "is that good?" before the reader asks. If the question is only "what is the number", the card is enough and the plain version is calmer on the page. If the manager has a goal for it, the target belongs on the screen, not in their head.
Colour and labels
Colour should carry meaning once. One accent colour for the bar that matters, grey for the rest, beats six colours that each mean nothing. Titles should say what the chart shows ("Sales by category, 2025") so a reader who arrives from a screenshot knows what they are looking at. If you need a legend to decode a chart, check whether direct labels would do the job better.
The measures a visual needs
Some visuals ask for more than one measure. A KPI visual wants Sales, a target and the time axis;
attainment is a third measure, DIVIDE([Sales], [Target]). The best habit is to decide the visual
first, list the measures it needs, and only then open Home → New measure.
The truth test, in Power BI Desktop
For each chart check Format → Visual → Y axis (does the range start at zero for bars?),
Sort axis from the visual's ... menu (value for categories, the date for time), and Data
labels (are they readable without hovering?). More than five or six slices in a pie, a second
axis nobody asked for and 3D effects are the usual failures.
Custom visuals
AppSource offers hundreds of extra visuals. Before you add one, check who published it, whether it is certified, what permissions it needs and whether it still works in Power BI Service. A fancy visual that breaks on refresh costs more than a bar chart ever saved.
Common misconceptions
"A table has all the numbers, so it is the safest visual"
A table is the right answer to "what exactly was the figure?". It is the wrong answer to "which is biggest?" or "is it rising?", because the reader has to compare in their head. Use it when the reader will look up a value, not when they need to see a pattern.
"Sorting is cosmetic"
Order is information. Categories sorted by value show a ranking; months sorted by value show nothing. The default sort in Power BI is often alphabetical or by value, and neither is right for every chart. Choose the order on purpose.
"A bar chart can start wherever makes the differences visible"
Bar length encodes the value, so cutting the axis at $200,000 makes Textiles look almost empty beside Furniture, though the real ratio is a little over two to one. Lines can use a narrow range because they encode position, not length. Bars must start at zero.
Check yourself
One question. Answering it ticks this lesson off on the map; it never locks a level.
Not quite. Sorting by value suits categories. For time, the position along the axis is the information, so reordering destroys it.Try another answer.
Not quite. Columns can show months. The problem is the order, not the shape: a calendar-ordered column chart would still show the trend.Try another answer.
Not quite. A table would give exact numbers, but the question is about direction, which the eye reads from a sequence far faster than from twelve figures.Try another answer.
Right. Growth is a statement about sequence. Once the bars are ordered by size, January sits next to whichever month was biggest, and the story is gone.
Lesson done: it is ticked on the map.