Using Data Insight To Make Better Business Decisions
Business intelligence and predictive analytics use data to paint a picture of what has happened and what might happen. But even more valuable than these insights is an understanding of how to influence what will happen. It might be helpful, for instance, to know if retail stores are selling more or less this quarter than the same period last year. Another benefit might be knowing that when a customer makes more than two purchases a month they are likely to purchase higher margin products. More importantly, business intelligence can help predict how to get a customer to make more than two purchases a month.
“Look for patterns, try to explain them, determine the impact on the present and future, and formulate the moves you will take to make the most of that impact," says Suresh Acharya, a business professor at the University of Maryland, and an expert in statistics and optimizations for supply chains. "Taken together, this is what data analytics lets you do."
Using Data Analytics
The key phrase in Acharaya's assessment is “formulate the moves you will take to make the most impact." Seasoned analysts may be familiar enough with industry patterns, seasonal variations, and customer motivations to formulate hypotheses about what might influence customer behavior. Unfortunately, those experienced analysts are in limited supply. Instead, companies are turning to data analytics to meet the growing demand for more insights into optimizing business decisions, including how to build relationships with customers.
Personalization is a popular technique for improving the quality of a customer's experience and for cross-selling. Bindu Thota, vice president of technology at Zulily, an online retailer, notes that personalization is about “giving shoppers the right mix of categories, selections, price points, shipping times and other key services." While this is an accurate description of personalization, it leaves the retailers to determine what is the right mix.
To influence an outcome, it is essential to understand its context. A retail analyst may, for example, want to identify characteristics of shoppers who purchase high margin products and when they purchase them. Basic statistics can measure how well two events or characteristics correlate with each other. For example, high margin purchases may be highest in December and on Thursdays. Historical data can reveal that December is peak season for retailers, so that is not surprising. But understanding the increase in high margin purchases on Thursdays is more challenging.
Using machine learning and statistics, analytics software can help a business analyst understand what is driving Thursday buying sprees. Of course, analysts need to avoid what are known as spurious correlations. These are correlations that show up in the data but don't have a direct cause and effect relationship. For example, my favorite spurious correlation, the number of people who drowned in swimming pools during a year correlates with the number of films Nicolas Cage appeared in. Fortunately, there are ways to help recognize spurious correlations. But if someone turns on a Nicolas Cage movie, maybe stay away from the pool...You know, just in case.
Data Science to the Rescue
This is where data science and machine learning come into play. Analytics software can generate large numbers of hypothesis about relationships in the data. This data becomes the raw material for machine learning models and statistical algorithms that rank hypotheses based on their actual ability to solve a business problem.
This requires some understanding of causation, which is difficult, but not impossible, to understand just by looking at data. Researchers at the University of Amsterdam, for example, found a simple rule that works surprisingly well: “If one event influences another, then the random noise in the causing event will be reflected in the affected event." Unusual weather, for instance, may correlate with a drop in retail sales, but drops in sales do not lead to unusual weather patterns.
Using techniques, such as the influence rule, analytics software can measure the impact of one event on other events. From there, analysts can use machine learning to help them build simulations to better understand the effects of multiple causative events and help identify actions that can lead to desired outcomes. For example, knowing that when customers make more than two purchases a month then they are likely to purchase higher margin products can help analysts build analytics models. These models could suggest actions, such as increasing the number of coupons and other offers to likely purchasers.
Analytics software has evolved from business intelligence reporting, which tends to describe the past, to making predictions about future outcomes, to finally influencing those outcomes. This development is democratizing knowledge that, has historically been difficult for companies to acquire.
The pace, precision, and effectiveness of business decision making is improving as companies embrace the use of analytics tools. These tools not only interpret data about the past, they also identify causal relationships that can be helpful in suggesting actions that will influence outcomes.
Once you have identified potentially useful insights, it is important to validate them with people familiar with the topic as well as assess the relative impact and cost of implementing the insight. The combination of impact and cost can be used to prioritize the recommended actions, which can then be presented to decision makers who can the decide the best action to take. Analytics tools can support more informed decision-making but it they are most effective when they move beyond describing correlations to recommending actions.
LATEST POST : The Simplicity (and Complexity) of Auto-Feature Engineering