Reaction: Netflix-like Recommendations for College
Earlier today I saw the following post:
Before I go deeper into my thoughts on the matter, I want to give a hat-tip to Dan Jarratt for actually deploying his idea. Well done sir.
Dan is on to something and I commend his approach. Trust me, if you provide services to highered, you should follow Dan immediately. He is studying recommender systems. What does that mean for you? Dan is able to synthesize actionable insights from Big Data. Yes, I am intentionally using a term that I loathe. He is doing the type of analyses today that highered will demand tomorrow.
Now, to my thoughts…..
I have long been pitching a “Netflix-type” algorithm for higher ed to my family, friends, well, anyone that would listen really. Highered desperately needs disruption. As a HUGE industry (we were expected to spend over a billion dollars in just paid advertising alone), I feel that we need to be more efficient. Period. If you follow me on the Twitters, it should not surprise you that I am very much against volume-based marketing. Yes, we can target students, but I would hazard a guess that many of us don’t do it well. Hell, I know that I personally can do better. My point? Recommendation engines are the future of student recruitment.
In the Chronicle article, Dan admits there is room for growth in his approach, but in my opinion, he is trying to get blood from a stone. Simply, he is doing the best he can with the data that are publicly available. For recommendation engines to work in this context, enrollment scientists need access to student-school data. Student-school…. what?!?!?
For example, the algorithms that Amazon and Netflix use are based upon user-movie and user-product pairs respectively. Simply, a user will rate a variety of products/movies. From there, the websites can recommend other products/movies based on other statistically similar users or products. There are many ways to do this, but that’s the art behind the science.
I have long wanted to get my hands on the database from services like Zinch or Cappex to explore this type of analysis, as they are well positioned to deploy this type of service. As much as we think each prospective student is unique, the reality is that we can deploy various statistical techniques to find groups of similar students, or in Dan’s case, statistically similar colleges.
From there, we can leverage the “college-search graph” to help students like themselves find the colleges they are looking for, easily. This type of student-school recommendation engine can help students narrow their list quickly to focus on the institution that fits them best. After all, satisficing is very real, and Barry Schwartz has exposed the Paradox of Choice.
Because Cappex and Zinch need to make money, the leads we (i.e. colleges) would get from this service would be more valuable because they are better qualified for our institution, as such, basic economics would suggest we should pay more for these prospects. If we can get a better ROI on our spend, then logic would suggest that is where we would allocate our budget.
The takeaway? Graphs (I mean this kind of graph) are the future of student recruitment.