Storytelling with Data: A Data Visualization Guide for Business Professionals

By Cole Nussbaumer Knaflic

10 min read

Why does it matter? Because the chart you make in thirty seconds takes your audience three minutes to misread.

You already suspect your charts aren't working. We've all felt it — the glazed eyes mid-presentation, the follow-up email asking you to "just send the numbers." The polite nod that means nobody changed their mind. Here's the part nobody said out loud: the problem didn't start when you opened Excel. It started three decisions earlier, when you skipped straight to the tool without asking who you're actually talking to, what you need them to do, and which of several true stories your data can tell. Most analysts never make those decisions at all. They inherit whatever the software defaults to — and the software has no idea what's at stake. What follows isn't a design manual. It's a corrective for a process flaw built into how most of us were trained: we learned analysis, but nobody taught us curation. That gap is the whole problem, and closing it changes everything downstream.

The Three Questions Nobody Told You to Ask Before Touching a Chart Tool

What if the reason your charts fail to drive action has nothing to do with the chart?

Most people treat data visualization as a craft problem: pick the right graph type, clean up the colors, add a title. That approach assumes chart-making begins when you open Excel. It doesn't. Three decisions determine whether any chart can succeed, and they all happen before you touch a tool.

The decisions are deceptively simple: Who is your audience, specifically? What do you need them to know or do? And only then — what data will serve as evidence? That sequencing matters more than it sounds. Most of us skip straight to the third question. We were trained to analyze; nobody ran a session on curation.

Consider a fourth-grade science teacher who ran a pilot summer program and has survey data worth sharing: student interest in science jumped from 40% to nearly 70% after the program. That's a compelling result. But compelling to whom? The same finding lands completely differently depending on who's in the room. Parents of participants want reassurance about their child's experience. Other teachers want methodology they can replicate. The budget committee (the people who control whether the program runs again) want one thing: enough evidence to justify signing a check.

The teacher who walks into the budget meeting with a chart built for all three audiences has built a chart for nobody. Commit to the budget committee, and everything sharpens: the "what" becomes a specific funding request, and the data becomes evidence for that ask.

Analysis means opening a hundred oysters. Communication means knowing which two pearls to put on the table. That's the exploratory-versus-explanatory distinction — and it's why presenters who show all hundred oysters are outsourcing their analytical work to the audience. The audience walks away thinking "interesting" rather than making a decision.

The tool for making that choice explicit is what Knaflic calls the Big Idea: a single sentence that states your point of view, makes the stakes clear, and is grammatically complete. For the science teacher: "The pilot program improved students' attitudes toward science; please approve our budget to continue it." Write that sentence, and it tells you exactly which chart to build.

Pie Charts Are Evil, and the Fox News Bar Chart Proves It

Fall 2012. A Fox News graphic fills the screen: two bars — the current top tax rate at 35% on the left, and what it will become once the Bush tax cuts expire, 39.6%, on the right. The visual gap looks enormous. Any viewer who caught that segment would feel genuine alarm.

Zoom in on the vertical axis. It doesn't start at zero. It starts at 34.

That single digit transforms the bars. When the baseline is 34, the left bar stands 1 unit tall and the right stands 5.6, instead of 35 and 39.6. An actual increase of 13% becomes a visual leap of 460%. The numbers on screen are real. The picture is a fabrication.

Bar charts rest on a specific perceptual mechanism: our eyes compare bar endpoints and read their lengths as the quantities being communicated. Move the floor and you move the lengths, and with them, what the audience sees as true. Bar charts must start at zero because the alternative doesn't produce a misleading chart — it produces a different chart that happens to share the same numbers. Anyone who catches the axis will throw out your argument and your credibility with it.

Pie charts fail the same perceptual test, and they do it without anyone fudging the baseline. They're evil by design. A market-share chart with four suppliers, rendered in 3D, showed Supplier B at 31% as clearly the largest slice. The bottom position in the tilted pie made it look closer to the viewer, and therefore bigger. Supplier A, which was actually larger, read as smaller. The chart gave the wrong competitive ranking before a single label appeared. Flatten it to 2D and the problem persists: human eyes cannot reliably convert angles or areas into numbers. You can tell that one slice is larger, vaguely. You cannot tell by how much, which means accurate comparison is the chart's only real job, and it can't do it.

Replace the pie with a horizontal bar chart sorted largest to smallest. Bar endpoints align at a shared baseline. Your eyes do the comparison the way they were designed to. The ranking is immediate; the gaps are readable. The chart type you choose determines which perceptual system you're asking your audience to use. Some systems aren't equipped for the task.

Every Visual Element You Add Is a Tax on Your Audience's Attention

Every element on a chart — border, gridline, data marker, legend, diagonal label — charges a processing fee the moment it enters view. Perception is automatic; the audience has no choice whether to pay.

Cognitive load is the mental processing power required to extract meaning from information. Its implication for data design is blunt. Unnecessary complexity is expensive. Every element that doesn't carry information still costs something. What looks like a thorough chart is often just a higher invoice.

That cost becomes concrete in the IT ticket-volume chart Knaflic works through step by step. Picture a standard line graph tracking monthly support tickets received and processed over a year (the kind you might produce in five minutes from Excel defaults). The chart arrives pre-loaded: a border, background gridlines, data markers on every point, y-axis labels with trailing zeros, and months written diagonally because they won't fit straight. A legend sits off to the side, disconnected from the data it names.

Six removals fix it. Drop the border — audiences already perceive a chart as a cohesive object without one. Drop the gridlines, or at minimum render them in pale grey so they stop competing with the data. Drop the data markers: the lines already encode the trend; the dots charge twice for the same information. Strip the trailing zeros from axis labels: 1,000 carries exactly the same meaning as 1,000.00, with none of the noise. Abbreviate the months to fit horizontally; diagonal text is measurably slower to read. Finally, move the legend labels directly beside the lines they name and color each label to match its line.

Six steps. All deletions. The revised chart contains identical data, but every remaining element earns its place.

Your Audience's Eyes Have Already Decided What Matters Before Their Brain Starts Reading

Your audience's eyes settle before their mind engages. Iconic memory, the part of visual perception that fires first, operates in fractions of a second, well below conscious attention. You cannot turn it off, and neither can your audience.

Iconic memory responds to a specific set of visual signals: differences in size, intensity, color, position. They're called preattentive attributes because they register before deliberate attention kicks in — before your audience has decided to look anywhere. The practical implication: you aren't really competing for your audience's attention. You're working with or against a reflex.

Here's the proof. Take a paragraph-sized block of numbers with six 3s scattered through the sequence. Finding them requires patient, methodical scanning, a slow hunt that taxes working memory. Now bold those same six 3s and leave everything else unchanged. The task goes from laborious to almost involuntary. You don't find the 3s; they surface. Your brain detected the contrast before you decided to look for it.

That shift from hunting to surfacing is the whole game. It means you can guide what your audience sees in the first three seconds (before they've read a title, before they've processed a legend) simply by creating contrast in the right place. One line bolder than the others. One bar colored while the rest stay grey. The audience hasn't chosen to look there. Their brain has already gone.

A Google team building a dashboard had equal data on three metrics, but acquired only one first. That metric filled roughly 60% of the screen while placeholders held the other two. When the remaining data arrived, someone noticed: the sizing was sending a clear signal that this metric mattered most. The layout had been making an editorial claim for weeks before anyone caught it. They restructured to equal sizing before launch.

I've done the same thing. I spent a week on a chart I was proud of: muted tones throughout, one line in red that I'd placed as a secondary reference. My colorblind husband looked at it for two seconds. "So the red line is the important thing?" He hadn't read a label. His preattentive system had already decided, and the hierarchy I'd built didn't exist for him.

There is no neutral design choice. Every sizing decision, every color assignment, every positioning call is an implicit claim about importance, whether you made it consciously or not. Intentional design means taking authorship of those claims.

The practical rule is blunt: design in grey first, then add one bold color precisely where you want the audience to look. Not for aesthetics — for control. Grey as default means anything that isn't grey becomes a signal. One color means one focal point. When everything is colored, the preattentive system has nowhere to land, and the audience is back to hunting.

The Same Dataset Contains Three Different True Stories

Is there a correct way to visualize a dataset? Most people assume yes — that somewhere in the data, a single honest chart is waiting to be found, and the analyst's job is to locate it. The retail pricing walkthrough in Knaflic's final chapters dismantles that assumption with the same dataset shown three different ways, each version honest, each telling a completely different story.

The setup: a startup needs to price a new consumer product. They pull retail price trends for five competing products — one line chart, standard issue. Color the right lines, fade the others to grey, drop a marker at the key moment — and a completely different argument emerges.

First: "After Product C launched in 2010, the average retail prices of existing Products A and B declined." Color A and B post-2010, mark Product C's entry, fade everything else to grey. The argument: new entrants drive price erosion for incumbents.

Second: "When a new product enters this market, its price spikes at launch, then falls." Recolor to highlight the launch-point peaks for Products C, D, and E. Same lines, now reading as a predictable spike-and-decline pattern.

Third: "By 2014, prices have converged. The average sits at $223, ranging from $180 for Product C to $260 for Product A." Mark the 2014 endpoints. The story is where the market landed, not how it got there.

All three are true. None is more legitimate than the others. The difference is entirely editorial: which preattentive signals you activate, which lines you push to grey and which you bring forward. Same numbers. Three arguments.

The implication runs deeper than technique. If the same dataset honestly supports multiple stories, then every chart you've ever made was an editorial choice — even when it didn't feel like one. The chart you built "just to show the data" was showing one story while suppressing others. You just didn't name the decision.

Which story should you tell? Whichever one your audience needs to make the decision you need them to make. That's the criterion. The dataset doesn't make the choice. You do.

Facts Don't Persuade People. Stories Do.

In a data communications workshop, the instructor asks participants to close their eyes and recall Red Riding Hood — the plot, the twists, the ending. A few laughs. Someone confuses it with Three Little Pigs. But when she asks for a show of hands, roughly 85% can reconstruct the story well enough to retell it.

Then she shows what Red Riding Hood looks like as a business presentation. Five bullets reduced a story with a wolf, a grandmother, a girl, and a woodsman to inputs, outputs, and expected outcomes. Same information. Zero effect.

The argument comes from screenwriting teacher Robert McKee. He identifies two modes of persuasion. Conventional rhetoric (bullets, statistics) is processed intellectually. The problem isn't that audiences ignore it. The problem is they engage too actively: while you present, they're forming counterarguments in their heads. You might win the intellectual debate and still fail to move anyone, because people don't act from reason alone.

Story is the second mode. Where bullets produce internal argument, story produces emotional experience. A story expresses how and why life changes: balance disrupted, a protagonist struggling to restore it, a resolution. Conflict isn't drama you add for interest; it's the load-bearing structure. Strip it out and you don't get a quieter story. You get the five-bullet slide.

That maps onto the classic story structure: beginning, middle, end. In presentation terms: setup, development, call to action. The beginning establishes the imbalance: what changed, why it matters to this audience, what a good outcome looks like. The middle builds the case: evidence, context, the cost of doing nothing. The end makes an explicit ask. The science teacher's beginning is the funding gap. Her middle is the 30-point swing in student enthusiasm. Her end is the dollar amount she needs. That last part is where most data presentations fail: they conclude on findings rather than action, leaving audiences to draw their own conclusions. If you need a decision, the presentation must close with one.

One diagnostic: read only the slide titles in sequence. If they describe rather than argue, the narrative isn't there. Knaflic calls this horizontal logic. "Q3 Satisfaction Results" is a label. "Satisfaction Dropped 14 Points — Here's What We're Asking For" is a sentence in an argument. Only one tells a story.

Applying this is messier than it sounds. Brand color constraints, a split audience, a colleague who insists on pie charts: these are real obstacles, not reasons to abandon the approach. The framework still works.

The final shift is about whose story this is. Novelist Kurt Vonnegut's writing advice has one principle that overrides the others: pity the readers. Frame the imbalance as your audience's problem. Make the solution something they're positioned to act on. A presentation built around your analysis is about you. Built around their decision, it's a story they can enter — and move from.

The Moment the Chart Changes

The shift isn't technical. You already know how to open a chart tool. What changes is the question you ask first: not "what should this look like?" but "who needs to make what decision, and what would actually move them?" That reframe is what the book keeps returning to. Everything else — the grey-first palette, the deleted gridlines, the Big Idea sentence — follows once you're genuinely trying to serve a specific person's choice rather than demonstrate your analysis.

None of this requires a blank canvas. Brand colors, a split audience, a boss who wants the pie chart back — these are real. The sequence (audience, ask, evidence) still works; it just clarifies which compromises matter and which ones you can push back on.

You don't need to overhaul how you work. Take one dataset you've already touched. Write one sentence about who needs it and what they should do. If the sentence comes easily, your chart probably already works. If it doesn't (which is more likely), you've found exactly where to start.