Like many buzzwords, business intelligence (BI) is often written about in glowing terms as a means to gain an unrivaled edge on your competition. The omnipotence of data driven decisions is often expounded upon without much accompanying detail. So while the utilization of data analytics can produce significant advantages, these advantages are abstracted by Big Data idealism.
To derive important insights through data mining and data analytics, you must first start with a question.
In the excellent Data Science for Business authors Fostor Provost and Tom Fawcett focus on two main types of questions that data mining can tackle. The first deals in discovery: "can we uncover significant relationships in our customer data that aren’t apparent without using data mining techniques?" The second relates to accuracy: "can we examine an action that either we or our customers repeat and find a way to reduce inefficiency or produce better results?"
These two broad questions cover the majority of what business intelligence software can prove or disprove. Thinking in these terms can make your BI efforts less abstract, and help you better pinpoint the problem you’re trying to solve.
A Question of Discovery
The most acclaimed BI examples are those that uncover seemingly invisible relationships between consumer behaviors, usually by mining historical data. Target’s prediction of pregnant women’s buying habits is one of the most famous (or perhaps infamous). The uniqueness of this discovery lays in the retailer’s ability to predict when its female patrons were expecting a baby, based on changes in buying habits. While buying diapers would be an obvious signal of pregnancy, Target’s analysts were able to identify expecting mothers from more subtle patterns, such as changes in lotion brand or food purchases.
No doubt you’ve read about the backlash and the subsequent questioning of Target’s ethics, but it’s undeniable that the campaign was successful in terms of numbers. And it worked by identifying consumer actions joined together by a central relationship, albeit a relationship that needed to be discovered through connecting data.
Like Target, Walmart used data mining to reveal unexpected purchasing behavior way back in 2004. Over a decade ago as hurricane Frances careened toward the northeast shore of the United States, Walmart wanted to predict how customers would respond to such a momentous situation. Why? So they could stock their shelves accordingly.
Looking past the expected flashlight and bottled water purchases, Walmart’s data analyst needed to identify unexpected purchasing behavior at a local level. They had the data to do so. Several years earlier, Hurricane Charley had swept the same region of the country, providing analysts with congruous data to examine.
After analyzing the historical set, analysts realized that the highest selling item during the pre-hurricane weeks was beer, and that the purchase of Pop-tarts, specifically strawberry flavor, increased sevenfold.
This type of data analysis often begins with a broader question: can we find unexpected buying relationships and capitalize? Target’s discovery connected a complex set of buying patterns to uncover a singular event that drove consumers to change their behavior. In contrast, Walmart’s was a bit more superficial, as it simply examined purchasing behavior as driven by a community-wide event.
The next question begins with a more granular focus, and emphasizes improvement over discovery.
A Question of Accuracy
While the previous discoveries are often brought up when discussing the use of business intelligence, such types of hypotheses are actually the far less common of the two. The second of our two questions deals with accuracy: the accuracy of marketing communications, the accuracy of customer retention efforts, and the accuracy of predicting customer behavior.
This question asks: How can we improve the outcome of an action that occurs repeatedly at a large scale?
This type of analytics helps businesses predict customer churn, or the likelihood a customer will leave their current service provider or downgrade their services to a less profitable package. Again, mining historical data forms the basis for decision making. Predicting churn requires building a statistical model that uses logistic regression and decisions trees – but in advanced cases can also utilize machine learning algorithms – to understand the myriad of causal factors that motivate customers to leave.
Churn models are particularly popular in saturated markets like finance, telecommunications, and increasingly Software as a Service. In these verticals, most of the adoption has been done, so it’s the acquisition of new customers, or on the reverse side the retention of existing customers, that drives profit margins. Consequently, developing a churn model has become a crucial part of staying competitive.
To relate this to our second question, these large organizations use data analytics to examine mountains of historical data that detail interactions between customers and the business to pinpoint the sequence of events (in this case referred to as causal relationships) that lead to a customer choosing to churn. By using data mining, these businesses can create predictive models that segment customers into groups based on their risk of switching providers.
Finally, business test these hypotheses in the field by changing the way their organization interacts with customers in specific instances. They also deploy salvage opportunities to high risk customers to convince them to stay. Since the data isn’t perfect, the reduction in churn won’t be 100 percent, but such efforts can improve the accuracy of customer service protocols as well as the effectiveness of marketing efforts to a tremendous degree.
When considering the data your business has stored within warehouses or spreadsheets, consider these two questions: "Can I uncover customer buying patterns in my data that aren’t obvious, or can I improve a process that happens repeatedly and is central to my business?" Your business intelligence will be better for it.
About the Author
Zach Watson is a content writer at TechnologyAdvice. He covers business intelligence, healthcare IT, and gamification. Connect with him on Google+.