Internal‎ > ‎

Predictive Modeling

The Challenge

Traditional analytical modeling tools have gained some popularity in enrollment management as a means to segment prospect populations based on historical data and then to recalibrate ad hoc communication strategies that support institutional market positioning and recruitment. However, these static methods are retrospective in nature and require several years of consistent historical data for implementation, limiting their appeal. More importantly, the current methods are not responsive to a college’s need to manage real-time communication processes and respond proactively to changing competitive pressures. In other words, they fail to realize the potential inherent in a real-time combination of data mining, rules-driven communication, and business process analysis.

The Solution

In light of this situation, the National Science Foundation (NSF) awarded a grant to Admissions Lab (now the Enrollment Technology unit within RuffaloCODY) to “develop an empirically-based adaptive method for automating the response logic needed to successfully negotiate critical decision-making steps in the communication process between colleges and prospective students.”  What resulted is a modeling process that goes far beyond a traditional regression analysis to utilize sophisticated pattern matching analysis.
This analysis takes into account (1) the broad range of data elements available for each prospective student, as well as (2) sequencing and timing variables related to the unique set of interactions that have occurred between each student and the institution. In awarding the grant to Admissions Lab, the NSF recognized that an innovative predictive modeling technology would allow colleges to “more rapidly adapt to competitive market pressures and become leaders in a growing global education market.” 

The Process

Every time the modeling process runs, a new “adaptive” model is created for each student that examines a matrix of geodemographic variables and interactions at that specific point in time—and compares them to the matrix of variables and interactions that existed for previous students at similar points in previous recruitment cycles. Resulting patterns are then used to predict--with great accuracy--the likelihood that each student will achieve a specific outcome—such as applying for admission or accepting an offer of admission.
Modeling scores are displayed in the Scores Table (see illustration below) and are accessible through Advanced Finds, workflow rules and reports. Because the modeling and scoring process is embedded in the Enrollment Manager platform, models and scoring can be run throughout the admission stream without having to export data or import results. As a result, Enrollment Manager clients have the ability to:
  • More accurately predict enrollment outcomes based each student’s unique characteristics and interactions
  • Generate modeling scores without the overhead of exporting data files or importing modeling results
  • Automatically incorporate modeling scores into segmentation strategies and communication tactics

Currently, predictive models are updated on a monthly basis using point-in-time historical data, and predictive scores for students in the current recruitment cycle are created or updated on a weekly basis.