The term Predictive Analytics most often refers to the application of statistical data modeling techniques, intended to provide a predictive model for an important business metric. It involves a “dependent variable”, which is the metric you want to predict. Secondly, it involves a set of “independent” , or sometimes called “driver” variables. The independent variables are those that influence the level of the measure we’re trying to predict.
The traditional, commonly-used statistical technique has been regression modeling, or one of its closely-related variations. Also, there is a wide array of more sophisticated methodologies such as neural networks and similar “data mining” algorithms. These more advanced methods are often used for very large data sets by companies who have massive databases, such as retailer scanner data. But for the majority of business applications and more common data set sizes, traditional regression modeling serves well.
Here are a few examples of some common types of predictive models:
- Market share prediction based on changing decisions within the “marketing mix” (prices, advertising spend, in-store merchandising, etc.)
- Sales volume forecasting model based on altering assumptions about company decisions, market conditions, prices, etc.
- Economic predictive model (e.g. GDP, unemployment rate, etc.) based on alternative levels of “driver” metrics
In our experience with developing web-based simulation models, the traditional sales forecasting model has been a staple application. This entails a professionally developed regression type model of sales, based on a series of data variables like marketing budgets, field sales calls, pricing versus competition, new product introductions, and a host of others.
Some of the most powerful (and consistently used) process simulation tools we have been involved with are these types of forecasting models. For example, it may have a predictive model for sales volume and revenue on the front end of the process. That might be followed by a “what-if” testing model to investigate several alternative policies for production, distribution and order fulfillment, given a set of assumptions about sales volume.
Finally, that can all be encapsulated in a financial modeling wrapper to emulate P&L and balance sheet effects, along with key profitability metrics. This type of approach can help build a “50 thousand foot” view for decision making. It can help lower and mid-level managers share that comprehensive perspective of the CEO’s office, so that functional areas are no longer constrained to their silo-oriented perspectives.
If you have such a process in your organization in mind, we’d love to discuss it with you to explore possible options for success with it. Please reach out to us here.
If you’d like to subscribe to future followup post on this subject, please submit the form below:Please select a valid form