Everyone can make a dashboard now. Who’s the editor?

Last week, Anthropic launched Claude Dashboards. I was excited to see them turn Claude’s dashboard capabilities into a formal artifact type. BI tools have promised self-serve since forever. This is the one that’s actually going to catch fire.

You connect your data, ask a question in plain language, and get a live dashboard. While it comes with much of the polish and best practices out of the box, consistency across all those artifacts is a separate problem.

Claude Dashboards gets a lot more people building. It also means hundreds of people drawing the same numbers in hundreds of different ways.

The binder

There are a few things I know in my bones about data visualization. Some of it came from the classroom, some from reading Tufte, some from building graphics in newsrooms early in my career. I have an acute sense of when to use a certain color, a dotted line, a dashed line, a solid line.

While I’m pretty good at applying that to an individual chart, for a broader system you need something else to actually be consistent.

At the Boston Globe’s graphics desk, some of this was written down. There was a binder of data graphics guidelines sitting on each of our desks. But it was tough to actually use.

You’re on deadline. You’re producing an entire spread for Sunday. The last thing you need is to get caught using the wrong line style, when every piece still makes sense on its own.

But for the reader, who’s looking across the paper every day, it might not make sense. Yesterday, the same colors and indicators might have meant something different. Every chart can be right, but the set can still be wrong if those rules aren’t consistently applied.

No graphics desk

Outside of major news organizations, very few companies have a graphics desk. It’s nobody’s job to make sure those charts or expressions agree with each other.

For a long time, that didn’t matter much. The only people who could build dashboards were either deep in Tableau and enterprise BI, or knew the data and could build for the web. At Upstatement, there was one other person anywhere near fluent. That scarcity kept things under control. One person, or a very small team, made most of what we looked at, so you got some consistency for free.

That scarcity ended last week. Everyone is going to be making things, and sending them around Slack and email.

One number, three pictures

Imagine three people actually manage to pull the same numbers. Finance draws revenue as a blue line and fades the forecast. BD stacks green bars with a pale cap for the distance to goal. The exec deck turns it all into a donut.

Wait, do these three charts mean the same thing? Are we working from the same data? That’s extra cognitive effort for the people who most need to use the numbers. It’s less time to spend on what the numbers actually say, and more opportunity for people to shrug and move on.

But this can work in our favor too. The CEO and CMO will inevitably send around screenshots from their new creations. If they draw from the same language, they can actually hammer the teams with the common picture of what they see. And from a clear picture (or pictures!), teams can fix those issues.

Most people can’t read your charts

I’m always shocked (and a bit saddened) when I walk someone through a chart or dashboard for them to gasp: “Oh, now I get it.” Wait, you didn’t before? I’ve been sending this out for months?!!?!

Of course, the people who make charts are, by definition, the most chart-literate people in the building. Our audience knows less about these conventions. We have to accept that.

It means we’re making kids’ shows compared to the art in our heads. Kids’ shows work because they repeat themselves. Every time red means we’re behind, the reader learns to look for red. The first time someone makes a client red because it was next in the palette, the lesson is gone.

What about brand guidelines?

But we have a design system! We have brand guidelines! Yes. But a chart is a more complicated thing to express, with much more nuance and specificity. Every graphics manual I’ve ever seen has its own data visualization colors, kept apart from the brand.

Brand guidelines exist to stand out. Chart colors mean something distinct. Those two jobs can work against each other. When a Coca-Cola executive looks at a chart in Coke red, does red mean something’s wrong, or that we’re selling a ton of Coke?

The rules lived with the editors

So what did the Globe have that most companies don’t?

I don’t remember actually ever consulting the binders. What I remember is the editors. They’d sit down with you, do a naive read of the chart, and talk you through what was clear and what wasn’t. They’d made enough charts over the years to have great judgment and a real eye for problems.

But they weren’t identical. James and Javier focused on different things, and sometimes applied those rules differently. They also knew better than to throw the book at you when you were trying to go to press late at night.

The rules lived in their heads. The binders were just a handy backup.

A newsroom could live with two editors’ heads because they sat in the same room. A company with hundreds of people making charts can’t. Writing the rules down forces you to pick.

Introducing CHARTS.md

That’s what CHARTS.md is. It’s a living file I’ve been using for the last couple of months to make sure that what I make stays true to a set of rules. It gets applied twice: when a chart is first built, so it starts from the right guidance, and again when it’s reviewed.

It’s helped me nail down specifics I’d only ever known in my bones, like when to use a dotted line versus a dashed one:

Line style says whether a number has landed. Solid is measured and final. Long dash is not settled: the future, or a number that can still move. Dotted is a marker: a pace, average, target or reference point. Computed, never measured.

It makes sure the same treatments and the same colors show up on every surface across a suite of dashboards.

The suite is bigger than just the dashboards, too. It’s every presentation and board deck that shows these financial numbers. That’s how the team, leadership, and the people above us all end up reading the numbers the same way.

Who owns how your numbers look

Today, before I share a chart to the organization, it goes through CHARTS.md. That’s my editor, doing the naive read every time. It also checks that the concepts I’m using today match the ones I’ve already developed, so this week’s dashboard agrees with last quarter’s. And as I work through it, the guidance keeps getting sharper, especially for the more complicated business concepts, like pace against a moving target or a weighted pipeline.

So who owns how your numbers look? At Upstatement, I’m taking that on. As executives and finance start building their own, CHARTS.md is how one person’s judgment reaches charts I’ll never personally see.

What are the rules? That’s what I’ve been encoding for the last couple of months, and I’ll share them here, one per post, over the next few weeks. If you want a preview, my CHARTS.md is on GitHub. It’s a working copy, and it’ll keep changing.

James and Javier carried the rules in their heads. The rest of us are going to have to write them down.