Business Intelligence Network
business intelligence resources

Blog: Craig Schiff

« Build vs. Buy | Main | EPM - The Next Big Thing? »

The Essential Elements of Predictive Analytcis

Numerous vendors lay claim to the ability to provide predictive analytics. These firms include Business Objects, Computer Associates, and OutlookSoft as well as several others. While they all seem to agree on the basics - predictive analytics enable a company to more accurately determine the potential outcomes of today's business decisions - they disagree on how you get there. In other words, much like business performance management itself several years ago, vendors and consultants are taking the opportunity to define the requirements of predictive analytics to fit their own capabilities. To begin the march towards a more standard definition, and therefore more stable ground for end users (as well as consultants and vendors), we would like to propose a list of 12 key elements to look for in a predictive analytics system.

The list that we are about to share was developed by a firm we have recently become aware of called Isis Solutions. They focus almost exclusively on predictive analytics and their executives have a strong grounding in advanced mathematics. While we think it is a good, fairly inclusive list, it certainly is open to debate if this is the last list we'll ever need of required predictive analytics components. We are looking for your feedback to help further develop and evolve the list. Now on to the list. Before you can predict you need to analyze and forecast. The first 7 elements focus on analysis: Variance Analysis - to identify deviations, Cost Allocation - to calculate operating income, Activity-Based Costing - to determine efficiencies, Slope - leading indicator of future trends, Mean - key indicator of performance, Standard Deviation - measure of variability in the business, Statistical Process Control - identify processes that are in and out of control. Next comes 3 elements related to forecasting: Holt Winters - seasonal forecast, Linear Regression - linear forecast, Flex to Historical - volume forecast. Lastly come the 2 elements that enable you to predict: Monte Carlo Simulation - calculate the confidence in the forecast, and R Squared - identify the hidden relations between actions. This list of 12 essential business analytics is intended to provide the key elements necessary to enable an organization to more accurately predict its future. Do you agree? We'd like your input.

  Posted by Craig Schiff on July 6, 2006 8:17 AM |

Comments

Hi Craig

Presumably you've got your tongue at least part way in your cheek when you talk about "the last list we'll ever need of required predictive analytics components". It's not that long ago that the mention of Activity-Based Costing would have raised eyebrows, but I don't expect there are many now who would dismiss it. Who knows what might be added to the list in the next few years?

I'm not a stats guy, and probably the list does cover this somewhere, but which element would you say covers cause-and-effect across the business? For example, I might want to build a predictive model where future sales are (partly) dependent on repeat business, and I build a rule that says that repeat business is directly related to cusomer satisfaction. (It's kind of like the relationships you might build in a Strategy Map.) What I'd like my predictive model to do is forecast based on those rules, BUT ALSO monitor those relationships and give me an alert if my assumptions don't hold up as the actuals roll in over time.

Sorry for the long-windedness! Would one of the elements on the list cover that kind of requirement?

Posted by: Steve Mainprize | August 16, 2006 7:08 AM

So what are the basic components or inputs to be considered in setting up a predective model? I'm writing a paper relating demand of medical offices based on hospital size, affiliated physicians, etc

Posted by: Ahned | October 18, 2006 8:46 AM

Post a comment