Consult the data. This elementary idea shines forth as a beacon to guide people toward sound business decisions. Despite the simplicity of the sentence, the process of consulting the data turns on a more involved set of assessments, procedures, and interpretations. Even the word “data” carries some internal ambiguity; it’s not self-explanatory. What counts as data? Good or bad data? Reliable data? Enough data? Relevant data? Thinking through these kinds of questions determines whether the data will be useful or in what way it may be useful.

Data can be collected from a variety of sources. Websites collect data about the amount of time people spend on each page, the links they clicked, and whether they purchased any products. But companies also send out surveys, asking their clients and customers about their goods and services, and what they might like in the future. Even before collecting data, companies need to consider how they ask these questions. As many of us learned from the book Nudge by Richard Thaler and Cass Sunstein, people are influenced by the way questions are asked, the order of the options, and other details. A survey’s results may be skewed inadvertently toward a particular response. It’s important to begin wondering about potential biases in the first phases of data collection.

Then, we need to consider the amount of data. Running a small physical store could present easier conditions for collecting manageable data with a smaller set of customers. While small shops continue to exist, the collection of data grew exponentially with the rise of e-commerce, which dons the name “big data.” The number of customers logging into Amazon each day far outpaces physical stores because so many people can purchase from an online store at the same time. But a physical store can contain only a certain number of people, even if they are crammed in as close a possible. Further, since an online store is not confined by store hours or geography, the data collected by such a company is astronomical.

Not only the amount of data but which data matters presents problems for a company. It’s not sufficient to only collect data, but this data must also be curated to find the best pieces that will provide the most information. In a paper called “Data Interpretation in the Digital Age,” philosopher Sabina Leonelli writes, “it is hard to imagine how data could be used as evidence for a claim, or as a reason to set up a research project, in the absence of intuitions about what those data tell scientists about specific entities or processes.” Even in the beginning, bias guides people toward particular understandings of data. The act of curation involves making choices, but our intuition pushes us toward choosing some data as more significant. We need caution in the act of selecting which data matters for our purposes.

Doesn’t data speak for itself? The answer depends on the amount of detail you require. At a basic level, data provides largely uninteresting information without interpretation. If you make 1,000 units of a product and 850 units are sold, then the data tells you that you sold 85%. You may be happy if you expected to sell less or frustrated if you expected to sell more. But this leaves many significant questions unanswered, especially the question “Why?”. Answering these deeper questions requires additional steps, including analyzing and interpreting more data.

Since data needs to be interpreted, then we once again consider that biases occasionally leech onto interpretations. Biases may be implicit or explicit, but either way they often lead to harm or exclusion. We could imagine implicit biases being generated from well-intentioned people who neglected to consult a particular group, thereby inadvertently causing them harm. Take, for example, some of the issues with bias in AI. Some programmers are likely surprised that the AI they helped design led to overt racism or sexism. The problem here may not be intentions, but lack of diverse points of view. This can infiltrate data analysis and interpretation as well, which is maybe why it is best for many people in a company to gain comprehension about the nuances of data.

Brent Dykes, a data strategy consultant, warns companies, “This data interpretation skill gap is more widespread across companies than many people realize.” Data interpretation requires more than understanding statistics, analytics, and programming. Interpreting data is an art that requires someone to provide what those from the humanities have known for years: a narrative structure. Storytellers provide this structure.

Once you believe the data is relevant, accurate, and informative, then you need to present it to others: clients, partners, employees, and investors. Presenting raw data can be confusing, distracting, and largely ineffective. Companies collect data in order to gain insights. Dykes explains that in order for data to bring an insight, rather than just an observation, three criteria must be met. First, the data should enable a “shift in your understanding.” Second, the data should reveal an "unexpected reason.” Finally, an insight ought to inspire action, so Dykes asserts that it should “align with what you care about.” Dykes further explains, “It’s up to the person sharing the insight to make sure the appropriate connections are drawn.” Sharing insights and making connections from data means telling a good story.

If we look at another context, art, then we find a common theory that art is supposed to express the ideas or emotions of the artist. This view of art seems to have some intuitive weight, especially when it comes to much contemporary art. But without the aesthetics of the artwork drawing people toward it, most of these ideas would remain less impactful or even unheard. The same goes for data science. Storytelling provides a powerful, memorable, and necessary medium for sharing the data with your audience. Rather than abstract numbers, presenting the data as a  story provides a concrete context and a unified narrative arc through which to understand and think about its meaning. It’s more efficient and effective. Moreover, a story connects with people’s emotions, which drives them to action. For instance, Gerald Zaltman, a business professor at Harvard University, claims that 95% of our retail purchases come through our subconscious, emotional side. Without this connection, many pieces of information get lost in the storehouse of our mind.

While data provides valuable insight, we need to be careful not to rely on it to the exclusion of all else. Giving too much weight to data alone becomes an ideology that controls us. Dykes urges everyone involved in these procedures to have data skepticism and data curiosity. We need to question the assumptions of our data and challenge the conclusions. We may end up following a course of action we believe is derived from the data, but we shouldn’t merely succumb to the data without levying some pushback. The collection of data continues to grow rapidly. While large companies, like L’Oréal, probably have sophisticated data strategies, even though they should regularly reassess them, smaller companies starting out ought to think about their data strategy, even though they may not have much data to work with in the beginning. They should plan for the data before it gets too complicated and unmanageable. But we should temper our data analysis and interpretation with other facets. We need creativity, curiosity, and experience along with data to guide our decision making. Data provides information, but people make decisions.