Navigating Big Data for Big Profits
TO THE INSIGHTFUL GO THE SPOILS
"One bit of insightful information may be the difference between victory and defeat," Wu says.
Wu identifies three layers of Big Data analytics, two of which lead to insights. The first of these, and the most basic, is descriptive analytics, which simply summarize the state of a situation. They can be presented in the form of dashboards, and they tell a person what's going on, but they don't predict what will happen as a result. Predictive analytics forecast what will likely happen; prescriptive analytics guide users to action. Predictive and prescriptive analytics provide insights.
It may seem simple, but Bhargav Mantha, a manager at ZS Associates, says that presenting the analytics on a clean, readable user interface is vital but sometimes ignored. "Users get frustrated when they see content that they can't decipher," Mantha says. "A canned dashboard just won't cut it; people need to know what action they have to take." Mantha says that users demand a "sophisticated alert engine that will tell them very contextually what actions to take."
Using such analytics, Target was able to uncover this insight: Women who bought certain products such as cotton balls, unscented lotions and soaps, zinc, and calcium were either pregnant or likely to become pregnant. Equipped with such information, the company was able to design coupons geared toward expectant mothers at specific stages of their pregnancy. It's this type of insight that helped Target increase revenue from $44 billion a year in 2002 to $67 billion in 2010.
Similarly, ZestFinance was able to glean this insight: Those who failed to properly use uppercase and lowercase letters while filling out loan applications were more likely to default on them later on. Knowing this helped them identify a way to improve on traditional underwriting methods, pushing them to incorporate updated models that took this correlation into consideration. As a result, the company was able to reduce the loan default rate by 40 percent and increase market share by 25 percent.
Software from the likes of InsideView can help companies uncover insights by setting filters around particular business problems and notifying salespeople or marketers when relevant changes occur, like, say, a personnel change in a prospect's management structure.
Unfortunately, insights have a shelf life. "Insights must be interpretable, relevant, and novel," Wu says. "They have to be new." Once an insight has been incorporated into a strategy, it's no longer an insight, and the benefits it generates will cease to make a noticeable difference over time. To Target's competitors, for instance, the kinds of signals that indicate pregnancy are now common knowledge, so Target now has less of an edge in this department.
GETTING THE RIGHT DATA
To get the right data leading to truly beneficial insights, a company must employ a sophisticated plan for collection, Mantha says. "Having a business case around the usage of data is the first important step," Mantha says. A company should figure out what goals it would like to meet, how and why customer data is crucial to reaching them, and how this effort can help increase revenue and decrease costs, Mantha says.
Wu agrees, pointing out that "relevance is key," and what is germane to a company is "determined by the problem [it is] trying to solve." He distinguishes useful data as that which contains signal, and everything else he lumps under "noise." But "one man's signal can be another man's noise,” he notes. If the target demographic is 18- to 34-year-old male sports fans living in New Jersey, it would make sense that the company would exclude information that falls outside those parameters.
Before we had countless tools to do the work for us, it was a no-brainer that companies would look only for the most pertinent data. "They'd start with the question, and collect the data that is specifically needed to solve the problem," Wu says. Collecting more than that was impractical.
But today, the process can get confusing, because often data is accumulating before a set of goals has been outlined by stakeholders. "Data is being collected irrespective of any specific problem, question, or purpose," Wu says. He points out that data warehouses and processing tools provided by the likes of Hadoop, NoSQL, InfoGrid, Impala, and Storm make it especially easy for companies to quickly attain large amounts of data. Companies are also at liberty to add on third-party data sources to enrich the profiles they already have, from companies such as Dun & Bradstreet. Unfortunately, most of the data, inevitably, is irrelevant. The key is to find data that pertains to the problem.