Data Storytelling: From Numbers to a Clear Point

A data story is a chart, or a short series of charts, built around one point the reader should leave with. Find that point before you design anything. Show it next to the comparison that makes it meaningful, explain why it matters to this audience, and remove everything on the page that doesn't help make it.

Most dashboards and report appendices show data without making a point. That is fine for monitoring. When you need someone to understand something or make a decision, a story works better than a wall of charts, because it does the first round of interpretation for them and shows the evidence for it.

Find the one point

Start by writing the point as a single sentence, the way you'd say it to a colleague in a hallway. "Mobile checkout fails twice as often as desktop." "Most of our growth this year came from one city." "The new onboarding didn't change retention." If you can't get it into one sentence, you probably have two stories or none yet.

Getting there usually means exploring the data first with quick, ugly charts nobody else will see. Look for what changed, what is unusual, what differs between groups, and what surprised you. Then ask what the audience needs to know or decide. A finance team and a product team looking at the same churn data need different points.

The point decides the chart. A claim about change over time needs a line chart. A claim about one group versus others needs bars with a highlight. The chart selection guide maps questions to chart types.

Context and comparison

A number on its own means almost nothing. "We had 4,200 signups in March" leaves the reader with no idea whether to be pleased. Compared with what? Last March, the plan, the industry, the month before?

Fig. 1 The left side gives a number. The right side gives the same number with three earlier years, and the point becomes obvious.

Pick the comparison that answers the reader's real question. Year over year removes seasonal effects, which matters for retail sales or website traffic in December. Against target answers "are we on track?" Against a peer group answers "is this us, or everyone?" Against a baseline before a change answers "did it work?" Usually one comparison is enough. Showing all four in one chart buries the point again.

Context also means scale. A 50% jump sounds big until the reader learns it went from 4 complaints to 6. Give the absolute numbers alongside percentages when the base is small.

Setup, finding, so-what

A clear structure for most data stories has three parts. The setup tells the reader what they are looking at and why it matters now: "Checkout completion is our biggest lever on revenue this quarter." The finding is the point itself, shown in the chart: "Mobile checkout fails twice as often as desktop." The so-what says what follows: "Fixing the mobile payment form is worth more than any other item on the roadmap."

This structure works at every size. In a single chart, the title carries the finding, the subtitle carries the setup, and an annotation or a sentence below carries the so-what. In a slide deck, each part might get its own slide. In a written report, each is a paragraph with a chart attached to the finding.

Keep the so-what proportionate to the evidence. A chart can show that two things happened together. It usually can't show that one caused the other, and the conclusion should say so when that matters.

Sequencing charts

When one chart isn't enough, order the charts so each one answers the question the previous one raised. A common sequence goes from the big picture to the detail: total revenue is flat, but one region is growing fast, and inside that region the growth comes from one product line. Each chart narrows the focus.

Keep the visual language steady across the sequence. Use the same color for the same category, the same scales where charts should be compared, and the same highlight color for whatever the current chart is about. When the reader turns to the next chart, they should only have to look for what changed.

Resist adding charts that are interesting but off the line of argument. Put them in an appendix or a link. Every extra chart costs the reader attention that the main point needed.

Remove clutter

Clutter is anything on the chart that takes attention without carrying information: heavy borders, dense gridlines, a legend for a single series, a different color for every bar, value labels on points no one will discuss. Each one is small, but together they hide the data.

Fig. 2 The same six months of revenue. Removing the border, gridlines, legend and extra colors leaves room for the one thing worth saying about June.

A good order for cleaning up is to remove first, then add back only what is missed. Take out the border and background. Lighten or remove gridlines. Replace the legend with direct labels. Turn the data gray and color only the point. Then add the title with the takeaway and an annotation where the point is. The labels and annotations guide goes into the text side of this.

Honest framing

A story that misleads, even by accident, costs more trust than it gains. The most common problems are choices that look technical but change the message.

Choice	How it misleads	What to do instead
Starting a time range right after a dip	Makes a normal recovery look like growth	Show a range that includes the usual pattern
Truncated bar chart axis	Turns a small gap into a large one	Start bars at zero, or plot the change
Percent change on a tiny base	4 to 6 becomes "+50%"	Show the absolute numbers too
Hiding the comparison that disagrees	The point only holds against one baseline	Mention the other comparison in a note
Implying cause from timing	Two lines move together by chance	Say "after", not "because", unless you tested it

Framing also includes what you leave out on purpose. If you dropped outliers, test accounts or a region with bad data, say so in a note. Readers forgive a caveat far more easily than a surprise they find later.

Check it on someone else

Before you share the story, show the main chart to someone who hasn't seen the data and ask them what it says. If they repeat your point, it works. If they describe the chart ("it's a line going up") or land on a different point, go back to the title, the highlight and the comparison.

Questions

Is data storytelling just a way of spinning numbers?

It shouldn't be. A good data story picks one point the data supports and shows it clearly, along with the context a reader needs to judge it. If the point only survives by hiding context or cherry-picking a time range, it is spin.

How many charts should a data story have?

As few as the argument needs. One chart is often enough for a single finding. A longer report might use three to six, each making one step of the argument, in the order the reader needs them.

What if the data has no clear story?

Then say so. "Nothing changed after the redesign" is a finding, and often a useful one. Show the flat line with enough context to make clear that the absence of change is real and not a measurement gap.