Zementis Industry Expertise: The 4 Types of Predictive Models

mark rabkin

Written by Mark Rabkin, Director of Business Development at Zementis.

For those of us in the Big Data space it is impossible not to notice the common use of terms like Data Science, Predictive Analytics, Prescriptive Analytics and Machine Learning. These terms refer to the use of dozens of different complex mathematical techniques used to improve business responses to hundreds if not thousands of common problems including but not limited to churn reduction, dynamic pricing, equipment utilization, fraud detection, health decisions, inventory management, offer optimization, predictive maintenance, risk management, and many, many more.

There is a lot of diversity in approaches to solving these problems from industry to industry as well as from company to company; however, when evaluating the financial value of a predictive model, our experience is that there are only four ways to monetize predictive models. This is important for Data Science teams to understand as they gain more and more attention from C-Suite executives and Line of Business owners.

The first way that predictive model deliver value is through Cost Savings. This is achieved by decreasing the amount the company is spending to provide its products and/or services to customers without reducing quality or efficacy. The benefit of this type of model is that every dollar saved either contributes to net income or is available for investment. Predictive maintenance and logistics, such as order routing, are common areas where predictive models are used to reduce cost.

The second way is by Growing Revenue. In this case, the result is improving quality of experience or quality of service in a specific way such that either current customers will spend more or the company will acquire more customers. The value created by these models is focused on the sales “pipeline”. Models can help increase the conversion rate between stages, such as increasing the rate at which prospective customer buys or a reduction in the time it takes for the sale to take place. The former is clearly growing revenue; however, the second one just changes the timing and is more like the next way to monetize predictive models. Recommendation engines are a common example of a revenue generation predictive analytic.

The third way a model creates value is Increasing Return on Investment (ROI),or Return on Assets (ROA), or a changing the timing of some other financial profitability measure. In this case the result is enabling something that was going to have a positive financial impact to happen sooner. These models are often combined with one of the first two types of models, since the amount of new revenue generated and/or cost savings realized also result in a differential in timing from what would have happened without the model. These kinds of models are often linked directly to financial events. A common example of this kind of model is predicting customers who may be slow payers. If a company can predict a slow payer it may able to take actions in advance to either alleviate the problem or at least reduce the days associated with outstanding balances on a consistent basis. This both reduces the overall expense for bad debt, improving operating cash flow, as well as shrinks the revenue cycle, driving up the firm’s ROI by reducing the need for either debt or equity financing.

The fourth use is targeted at Managing Risk, by which uncertainty in outcomes is reduced, resulting in more accurate knowledge of a new or ongoing business activity. This knowledge normally provides an estimate of the contingencies that are factored into ROI/ROA calculations. The full financial evaluation of these models can be a bit trickier depending upon the business opportunity to which they are being applied. It is difficult to evaluate the impact of not taking an action that would have otherwise been taken because given new information the risk is too high or the converse, making a bigger investment in a business activity because given new information the risk of investment is lower. One industry where this kind of predictive analytic is common and powerful is in the insurance industry. Insurance companies must be able to as accurately as possible with available information predict the risk of both existing and new policies and price accordingly.

I hope you have found this post informative, it was co-authored by Pasquale (Pat) Lapomarda, Assistant Vice President Advanced Analytics at Unum and I. Please share this post if you enjoyed it and we would enjoy hearing your comments.

If you are interested in learning more about how you can improve the ROI of predictive analytics please contact me at http://zementis.com/contact/. There is a tremendous amount of information at www.zementis.com. Zementis provides “write once, deploy anywhere – immediately” solutions for predictive analytics that drastically reduces the complexity, expense and time of implementation.