Content marketers have started to tell stories with data, and best practices are quickly emerging. The first step is to find the story you want to tell: how you approach data collection and analysis will determine what kind of content you’ll be able to develop. Once the story becomes clear, craft your message without letting the data overwhelm it.

Finding your story

Start with your dream headline. When I work on a data-driven content marketing project, I like to start by imagining my dream headlines or tweets: the discoveries that I would love my data to yield. When I was looking at child-related security risks, for example, I hoped to discover the security practices that led to the biggest reduction in online misdeeds—something like “good passwords cut hacks perpetrated by kids by 50%.” While the data rarely turns out to support that dream headline, starting there lets me figure out how to tackle my research. What data would I need if I wanted to produce that dream story? How would I go about getting it? Looking for the data that would yield my best-case outcome helps me figure out what kind of data is going to be relevant to my audience, and gives me a clear focus when I’m plowing through a mountain of survey results or social media analytics.

Recognize your bias. Part of the appeal of data-driven content is that we think of data as unbiased and objective. But when you’re using data for marketing purposes, you often do have a bias, because you want data that helps deliver your key message or that shores up your particular brand story. As a marketer, you can and should let that bias shape the questions you ask, the topics you pursue, and the parts of your data you highlight. To make sure that bias isn’t leading you astray, however, ask yourself whether your ultimate story and insights accurately reflect the data you’re working from: a good test is to think about what someone would conclude if they had access to your full data set. If they’d likely come to a different conclusion, you’ve done too much cherry picking, and need to rethink the basic story you’re telling.

Look for patterns. There are times when we just want to tell an interesting story—we’re not interested in it proving something specific. I doubt that Jawbone cares whether New Yorkers go to bed at 10 or at midnight, or that Facebook cares about whether we “LOL” or merely “haha,” but sharing data on those patterns lets brands catch the attention of the media and potential customers. Sometimes it’s the absence of a pattern that’s interesting, like OKCupid’s data showing that gay and straight people have the same number of sexual partners. The easiest way to get started with data storytelling is through stories like these: stories that are quirky and interesting, but where your brand has no particular stake in what the data shows.

Look for surprises. The most compelling data-driven content tells the reader something they don’t already know. Sometimes that surprise lies in finding an unexpected correlation: you might expect younger workers to be more likely to communicate online and less likely to meet in person, but actually, the reverse is true. If it’s not surprising, it’s going to be a boring story.

Telling your story

Once you know the story your data will tell, you need to structure that story in a way that makes it as clear and compelling as possible. To do that:

Choose the right format. Despite the impression you might get from looking at data on Pinterest, one long infographic isn’t always the best way to tell your data-driven story. A white paper, a blog post, or even a simple tweeted-out graphic can all be effective ways of telling a story with data, depending on your goal and audience. If you have a lot of data or a complex story to tell, a long piece that fully explains your results is more effective than trying to fit all that complexity into the margins of a single graphic; if your heart is set on a short visual post, release it as a highlight or teaser for your full-length piece.

Articulate your key message. Whether you succeed in finding the dream story you started with, or find something totally different when you dive into your data, your final story should clearly communicate one key message. How would you summarize your story in a single sentence or tweet? Articulate that story very clearly at the top of your post or document, and use each section or chart to build on that story—just as you would in any other piece of persuasive writing.

Lead with one or two numbers. One of the biggest pay-offs from data-driven content is the kind of media and social media attention it can earn. The surest way of attracting that attention is to highlight one or two surprising, memorable numbers. When Grant Thornton published its latest study on women in corporate leadership, it prominently noted the fact that close to a third of companies have no women in senior management. When Vision Critical released What Social Media Analytics Can’t Tell You About Your Customers, we emphasized the point that 85% of what you hear online comes from less than 30% of your social media audience. These are the kinds of facts that get tweeted out and picked up in news stories.

Balance text and visuals. A lot of data-driven content goes astray by sprinkling a few numbers into a big block of text, or conversely, by burying the reader in charts and graphics. The best content uses text and visuals synergistically: charts provide full context on the data you’re sharing, while text lets people understand how to interpret those charts, and why the numbers are relevant to their work. Make sure that text and visuals are balanced not only in terms of quantity, but in terms of quality: I often see beautifully designed infographics that are full of spelling errors, or that fail to explain what the colorfully presented numbers actually represent.

Illustrate your data with human examples. Whenever you’re telling a story with data, use real or hypothetical stories of specific people to translate the numbers into a human story. Our HBR article ”How Pinterest Puts People In Stores” showed that a third of people said that pinning their most recently purchased Pinterest item had “a lot” of influence on their decision to buy it—and that number was made a lot more tangible through the specific story of Claire, who got a sale alert based on an item she’d pinned. While “Claire” was an invented name, her story was based on the specific responses of a single survey respondent. That kind of example makes it easier for people to understand the story you are telling with your data, and also makes it more relatable.

Make recommendations. If you’re delving deep into a data set, you may see the relevance of your data to a range of business or consumer decisions—but that doesn’t mean the relevance will be obvious to your readers. Once you’ve done your data analysis, step back and think about how you would make difference business or purchasing decisions based on the data you’ve uncovered. Then spell out those insights in a separate “recommendations” or “key insights” section.

The more you follow these best practices, the more they’ll feel like a natural extension of the communications skills you’ve honed in other aspects of your work. And that’s exactly the point: if you’re doing a good job of telling stories with data, it’s the story—not the numbers—that will shine through.