Predictive Modeling and You

The field of data science is booming, and your business stands to benefit. But many organizations have yet to harness their data to make big gains. Some assume that they are too small; or that they are in a field that doesn’t apply; or that they simply don’t have enough data to mine. For others, the field seems too complicated, expensive, or futuristic.

To all these doubts and misperceptions, we claim a hearty “not so!”

The basic principles of data science aren’t complicated, and taking the time to understand them can pay off with big dividends. This post is aimed at providing a lay-understanding of one of the field’s most powerful tools: predictive modeling.

How Does It Work?

The term “predictive modeling” refers to the practice of using historical data and algorithms, (like machine learning models) to predict future actions or outcomes. “Historical data” can be anything from the online movie ratings you’ve submitted (like on Netflix.com), to your bill-paying history (which informs your FICO score), to your browsing and purchasing habits. This behavioral information is aggregated and used to build a mathematical model, which enables companies to identify trends and patterns. Those patterns are then applied to current data to predict what you’ll likely watch, buy, or do.

Predictive modeling has ushered in a profitable new era for businesses via upselling, cross-promotion, and forecasting; and correspondingly, a new era of convenience for consumers. You know this from having visited an online retailer to check the price and reviews of a specific item, only to purchase something else that the merchant offered that better suited your criteria. Or perhaps you even bought both! Either way, you got what you wanted, and so did the business.

Perils and Potential

Of course, increasing your sales (or enrollments, or fundraising gifts) isn't as neat as plugging in predictive modeling software and walking to the bank. For one thing, your data must be vigilantly reviewed for relevancy, correctness, and appropriateness before it ever enters a model. Social media darling Pinterest learned this the hard way when it sent congratulatory emails to women with the assumption they were getting married. Many were not, and annoyed recipients took to Twitter to share their unhappiness. A company spokesman was later forced to apologize, saying that they came off “like an overbearing mother who is always asking when you’ll find a nice boy or girl.”

Another erroneous notion is that predictive modeling software can be left to function independently. As with any “machine”, predictive models require regular testing and adjustment. Who can forget the time big box retailer Target incensed an unsuspecting father by sending baby-related coupons to his teenage daughter who hadn’t yet disclosed her pregnancy? The lesson here is that models must be consistently tested for assumptions and built-in biases that could compromise the trust of your client base.

Both examples illustrate why it’s important to examine your data before, during, and after it enters your models. You needn’t be scared off by such misfires, though. Target, Pinterest, Netflix, and Amazon haven’t been, and neither have thousands of other retailers, universities, service providers, and sports teams! To the contrary, organizations of all types and sizes are using predictive modeling to imagine new ways of improving performance and increasing sales.

A great example of this comes courtesy of Sephora. The cosmetics giant recently launched Visual Artist, an app that enables their customers to easily try on different looks. Using past purchase history and demographics, Visual Artist also suggests new products based on individual skin type and beauty preferences. This innovative service strategy has launched Sephora into the #1 slot of cosmetics retailers world-wide, and there’s nothing scary about that (except maybe the pressure of maintaining such momentum)! That sounds like a good challenge to have, especially since data is continuously streaming in to suggest new opportunities.

In conclusion, predictive modeling can be a game-changer for your business. It can help you identify stronger leads and ideal customers (those most ready to buy), and in turn free up your time and resources to close new sales. Predictive modeling can also suggest new revenue sources, and help you better serve your customers. So long as you keep humans at the heart of your strategy (check those models early and often!), customer interest and loyalty will be as strong as your financial growth.