Visualisation exists because people spot patterns in pictures far faster than in columns of numbers.
The eye picks out a rising line, a gap or an odd point almost instantly, while a table must be read cell by cell.
That speed is useful twice: once while exploring data, and again when explaining a finding to others.
Exploratory charts are quick and rough, made for yourself, and most of them are thrown away.
Explanatory charts are made for an audience, so every label, colour and title is chosen to carry one message.
A good explanatory chart has a title that states the finding, such as sales fell after the price rise, not just sales by month.
The best-known demonstration is Anscombe's quartet: 4 small datasets with nearly identical summaries and very different plots.
One is a straight-line trend, one a smooth curve, one a line with a single outlier, and one a vertical stack with one far point.
Summaries are still needed, because a chart shows shape well but exact values poorly.
Charts can mislead as easily as they inform, since design choices change what the eye notices first.
A chart also leaves things out, such as rows filtered away before plotting, which the reader cannot see.
Too many series, colours or labels make a chart slower to read than the table it replaced.
In practice, plotting each column before analysing it catches errors, skew and outliers early.
For explaining, one chart per message usually works better than one chart carrying everything.
The common misconception is that visualisation is decoration added at the end, when it is one of the first steps of analysis.