Marketers and businesses are always looking to better understand their customers and potential customers, seeking to know what influences the purchase decision and what actions, triggers, or encouragements will accelerate sales. Organizations started relying upon data science to gain valuable information, including insights into customer behavior from the growing volumes and types of data – sometimes called “Big Data,” which the critics call an informal way of referring to lots of data. In the past few years, organizations have shown renewed interest in data science known as predictive analytics.
Traditional analytics is generally descriptive, providing statistics and reports on historical events or actions that have already occurred. On the other hand, predictive analytics is an area of data science where models are created using data, statistical algorithms, and machine-learning techniques to analyze current and historical facts to foretell about future events – the likelihood of a probable outcome in the future.
Predictive analytics focuses on actions on an individual level – per person, per campaign, per store, and the like. Massive amounts of data are available that encompass behaviors, characteristics, and outcomes for all kinds of individuals and activities, representing a vast array of experiences. Predictive analytic models use machine learning to analyze the volume and details of all this experience represented in the data in order to discern the predictable rules, patterns, propensities, and behaviors, subsequently used to predict the likelihood of certain behaviors, actions, or performance. These analytics are also deployed to recommend specific actions to influence the outcome.
These statistical models are tested using more data representing more historical experience, leading to more precise predictions. While this area of data science is empirical, it is more precise than guessing and therefore is of significant value.
Common applications for predictive analytics
A 2014 TDWI report found that the top reasons organizations are using predictive analytics are the following:
- Identify trends.
- Understand customers.
- Improve business performance.
- Drive strategic decision-making.
- Predict behavior.
According to a report by Forrester Research, the most common applications for predictive analytics are cross-selling, upselling, determining customer profitability, promoting customer loyalty, and credit scoring.
An enterprise might use predictive behavioral analytics to predict purchasing behavior to better target its marketing efforts; predict the probability of closure to prioritize leads; predict order cancellations to implement ways to improve customer retention and loyalty; or predict a customer’s product choices based on views or past purchases to make product recommendations personalized to that customer.
Frequently cited examples include the following:
- Target uses a customer’s shopping patterns to predict the customer’s pregnancy, and then direct product offers specifically useful or necessary to newborns, identifying 30% more prospects.
- Hewlett Packard use analytics to predict employees who were likely to leave their jobs so that managers could take actions to retain them or otherwise be prepared.
- Google uses predictive analytics to provide search users with high-quality pages in their search results.
- Amazon uses predictive analytics to offer personalized recommendations for other products based on the visitor’s product views which comprise 35% of its sales.
Uplift Modeling (sometimes called Persuasion Modeling) is a specific predictive analytics technique that has gained recognition in recent years. The technique is used to find members of a target audience who are “persuadable.”
It gained publicity after the statistical modeling team for the Obama for America 2012 campaign used an uplift modeling program to precisely identify voters who were leaning Republican but were likely to be receptive to the Obama message. The models used demographic, geographic, and political data to identify the characteristics of persuadable voters in swing states statistically; and used the models to determine which voters should be targeted with television ads, which with door-to-door solicitations, calls or mailings.
Uplift modeling has obvious applications in a marketing context. The idea is that an audience of potential customers includes the following:
- Those who have already decided to purchase a specific product regardless of any contact
- Those who will absolutely not purchase the product even if contacted
- Those who would react negatively to being contacted
- Those who can be convinced or persuaded to purchase with contact.
Clearly, it is more efficient for a marketer to target the group that can be persuaded. Marketing resources are wasted on those who have already decided one way or the other, and it is counterproductive to expand on those who would react negatively to contact.
Uplift modeling allows marketers to direct resources more efficiently and potentially accelerate sales by targeting ads and other efforts at persuadable consumers.
For example, a company undertaking a direct mail campaign might use uplift modeling to predict who on the mailing list are unpersuadable, unlikely to respond to the company’s offer, and remove them from that mailing. The company would save marketing dollars, have a better-targeted audience, and increase the response or conversion rate.
Predictive analytics is an exciting area of data science with wide applications. For business, it has significant potential to allocate resources more efficiently, accelerate sales and increase revenues. It allows vast amounts of historical data to be analyzed, which can predict the likelihood of certain behaviors, and then recommend specific actions to influence the outcome in the desired direction.