THE BOX MODEL

How It Works Today

Institutions with predictive modeling capabilities tend to use it in two ways:

1. to help target promising demographics for list purchases
2. to determine whether an applicant would deposit if admitted

Perhaps it’s no surprise that these are the very uses of predictive modeling I find most offensive.

Used in this way, predictive modeling encourages an institution to make a value judgment about a student based on a number, a judgment that may or may not be right, but which can result in a drastic–and often terminal–decision.

Simply put, if the student does not meet a threshold value, they are discounted.

That may help you build selectivity, but it could also hurt your institution’s competitive strength in geographic, social, ethnic, or financial arenas where it may already lag behind competitors.

How It Could Work

Instead of using a model at either end of the recruitment arc, try to apply it throughout the entire process.  Rather than discard students based on a number, use those numbers to discover the most meaningful messages and relationship-building opportunities.

Here’s a typical scenario.  Say you have students in the medium cohort you wish to see move into the high cohort.  In other words, you have a group of students considering your institution who have not yet decided to apply.  What can you do?

In the traditional enrollment funnel, you would push the students to apply by encouraging a visit, offering fee waivers, sending “VIP applications,” etc.

But in the box model, the goal is to identify the obstacles preventing the students from making the choice for themselves.

What a perfect opportunity for predictive modeling!  Rather than try to address every concern a student has, you can use modeling to try to determine what the concern might be in order to address it more personally and specifically.

This requires a much more active use of predictive modeling that either of the 2 ways identified at the start of this article.  Now, instead of taking action on a student based on their rating, you try to determine which of the next possible actions would most increase the student’s score.

It’s sounds complicated, but it is actually fairly straight-forward.

Take each student in the current cohort and compare her to all the students in the desired cohort to find which one most closely matches her profile.  Then answer the question: what did that second student do differently from the first student that had the largest impact on his transition?

Here’s a more concrete example:

Let’s say you have students in the medium cohort you want to move into the high cohort.  You select one of the medium-cohort students and find the closest matching profile within the high cohort.  Let’s pretend they are both out-of-state students from depressed high schools with low graduation rates who have called your office.  But, after his phone call, the high-cohort student (who at the time was in the medium-cohort) received a follow-up call from a faculty member in his academic field.  Your model believes that’s what made the difference.  At least now you have some idea how to make a connection with the first student that will help to remove an obstacle.

So, rather than judge a student, you use your model to help her.

That’s a use of predictive modeling I can whole-heartedly support.  And who knows — maybe, by helping her, you also end up helping your institution become more competitive in its weaker markets.  Then everyone wins.