Most charts in most products exist because a chart felt appropriate, not because anyone asked what question it answers. The result is a dashboard that looks analytical and informs nobody: six graphs, no conclusion, and a user who exports to a spreadsheet anyway. Choosing well is a small discipline with an outsized effect on whether a product feels intelligent.
Start from the question, not the chart type
Every useful chart answers one of a handful of questions, and the question determines the form.
How has this changed over time? A line chart, always, because our eyes read slope as trend. How do these categories compare? Bars, sorted by value rather than alphabetically, unless the labels are dates. What is the composition of a whole? A stacked bar or, if you must, a single pie with three slices at most. How do two variables relate? A scatter plot, which almost no product uses and many should. Where is the outlier? Often a table with conditional formatting beats every graph on the page.
If you cannot state the question in one sentence, the chart is decoration. Delete it and reclaim the space for something with a conclusion.
The single number is underrated
The best visualisation of “revenue is up 12 percent this month” is the sentence “revenue is up 12 percent this month”, set large, with a small sparkline beside it for texture.
Product teams reach for graphs when users want answers. A dashboard’s top row should be plain figures with their direction of travel and their comparison period, and the graphs should sit below for the people who want to interrogate the number. This is the difference between a screen you glance at and a screen you study, and most users are glancing. It is also why the ten second readability test matters more than chart variety.
The choices that quietly mislead
A chart can be technically accurate and still leave a false impression, which in a product handling money is a genuine liability.
Truncated axes are the classic: starting a bar chart’s y axis at 90 turns a two percent difference into a visual landslide. Bars need a zero baseline. Line charts do not, since they show change rather than magnitude, but the axis should be labelled clearly enough that nobody misreads the scale. Dual axes are almost always a way of implying a correlation the data does not support. Cumulative totals always rise, which makes them flattering and nearly useless for spotting a decline.
And every comparison needs a stated basis: against last month, against the same month last year, against target. A number without a comparison is trivia.
Colour carries meaning, so spend it carefully
Colour in a chart should encode something. When everything is coloured, nothing is signalled.
The practical version: one accent for the series that matters, neutral greys for context, and semantic colours reserved strictly for semantic meaning (red for loss or error, green for gain, nothing else). Categorical palettes should stay under seven values, because beyond that nobody can match a legend to a line. Sequential data wants a single hue ramp, not a rainbow.
Then check it twice: once for contrast and colour blind safety, since red against green is the most common encoding and the most commonly invisible, and once in the other theme, because a palette tuned on white often turns muddy or fluorescent on dark. Pair colour with a second cue (label, shape, position) so it is never the only carrier of meaning, and build it from the palette the rest of the product uses.
Label directly, and let people see the rows
Legends make eyes travel. Direct labels at the end of a line, or above a bar, cut that work entirely, and tooltips should confirm a value rather than being the only way to read one.
Give people the underlying table too. Serious users always want the rows, whether for a spreadsheet, a board deck, or a suspicion that the chart is wrong. A visible path from a graph to its data is one of the strongest trust signals a product can send, and it is what operational dashboards live and die on.
Beyond that: no 3D, no animated entrances that delay reading, no chart junk. Motion should be reserved for moments that mean something, and a chart loading is not one of them.
The test that settles arguments
Show the screen to someone outside the team for ten seconds, take it away, and ask what the number was and whether it was good news. If they cannot say, the problem is not the data.
We run that test on every dashboard we design, and it changes more layouts than any opinion in the room.
Want your dashboard reviewed against these? hello@beconfidency.agency.
If you want the whole product surface designed by one accountable team, that is exactly what our product design service is for.