Analytics Meets Mother Goose

Want to get your point across about data? You’d better learn to tell stories.

Analytics is not just about data. In fact, getting too caught up in the data can obscure what the data means. And what it means is what matters in business.

But data scientists often get more involved in the data. “When I ask analysts to tell a story, they feel like they have to talk about their data and their process. That’s what they should not be talking about,” says Meta (pronounced May-tah) Brown, an analytics consultant and speaker who just published Data Mining for Dummies.

Instead, they should focus on telling the business story their data shows. One exercise she gives her clients is to develop a 60-second story from their data. She uses this to force data scientists to talk not about their data and processes, but on one thing the data says about the business’s customers. Most of them struggle.

This is the “last mile problem” of analytics, says Arvind Karunakaran, a PhD student at MIT’s Sloan School of Management who is studying how companies can get better decision making from analytics. The phrase refers to the problem of getting connectivity from telecom and ISP endpoints over the “last mile” they travel to arrive in homes — a thorny bandwidth conundrum. In analytics, this last mile is making connections between the data and executives.

The good news is, you don’t need a PhD to communicate — just practice and a shift in focus. Brown tells students keep points short, talk in dollar terms, and don’t overdo the details. Pay attention to what interests your audience, and respect it. Remember that they are the ones with the power to act on your insights.

While that sounds simple, Brown says most data scientists go to graduate school, and “people who go to graduate school are being trained to write in a way that most people cannot understand.”

Brown experienced this first-hand when she was working on her MS in nuclear engineering at MIT. She walked in to her advisor’s office one day and found two chapters of her thesis in the trash. He told her to rewrite them in a more academic style — one she felt was less readable.

Brown, who spent more than a decade in various roles at SPSS, a maker of statistical software packages, says the number of requests she gets from companies to help train data scientists to communicate jumped in the last couple of years, as the phrase Big Data became widely used. Usually, she hears from business managers that they can’t understand the point their data scientists are trying to make.

Compounding their communications woes, data scientists tend to work in relative isolation and focus on their data rather than the “big picture” their data sheds light on. The people they talk with most often are probably other data scientists. In contrast, information technology workers will probably work every day with people in marketing and sales and operations, and will learn in a hurry if people can’t understand what they’re saying.

She has her clients practice talking about analytics by having them summarize news stories with an analytic angle. For lengthy presentations, she has her clients work on creating separate stories about the data for each section of their talk. They’ll still be talking about dry numbers and algorithms that may be beyond their audience, but as long as the data framed by a story, people will be likely to pay attention — and learn something, which is the point of communication.

Data visualizations are a popular tool, but Brown thinks they are used too much, and that visualizations often obscure the point. “People can understand straight lines, can compare them. When you take anything less straightforward, they can’t do it. If your aim is to get executives to take action, don’t overload them with visualizations, and don’t be fancy.”

Scientists often make “heavy” graphics, says Justin H.S. Breaux, an external communications specialist at Argonne National Laboratory. “Making something meaningful from data, that’s always a challenge.” At Argonne, the scientists are encourage to frame their data around what will matter to Jane Doe, and to brainstorm ways to tell that story.

While some people are natural storytellers, anyone can develop the skill of constructing a narrative. Brown recently had a student stand up and tell a story about a person using a new version of software for the first time, and becoming frustrated when their work seemed to disappear.

That story came out of technical support data for an important new product, highly touted by the executives at his company, who didn’t know their software had this significant problem. “I could feel the software user’s pain in the pit of my stomach,” she says. That’s a story she thinks executives will hear.

1 Comment On: Analytics Meets Mother Goose

  • Thomas Speidel | November 24, 2014

    I think the argument that the analyst should be able to tell stories from data fails to recognize that most stories we are supposed to glean from data are cognitively dissonant. Faced with this incoherence we tend to cherry pick what story to communicate: those that fulfill a vested interest, impress our bosses, get us publication in a scientific paper, or are elegantly simple to understand. We then have to wonder if we are doing a disservice to the organization by encouraging story telling rather than a more cold approach to evaluating evidence.

    The role of the analyst is not necessarily to be a good story teller, but rather as someone who can lay out the evidence as it relates to a problem and perhaps make recommendations. Not to discourage developing strong communication skills, but story telling creates all sorts of biases that we fought hard to remove or control in the first place thus impairing our ability to make optimal decisions. Story-telling promote a paternalistic approach, rather than focusing on evaluating the evidence, be it in favor, against or inconclusive.

    It is a myth that the truth is hidden in the data, that somehow every dataset contains a story waiting to be unlocked by the best and brightest. Give the same data to ten different analysts, and you will get ten different stories. Who’s right?

Add a comment