Context Decay
The systems schools use aren't broken. They just don't talk to each other. And in an AI world, that's the difference between personalization that works and personalization that's pretending.
Last week, a team member shared an insight in Slack that has stayed with me.
His point was that the systems schools use aren’t broken. The decisions to buy them weren’t bad. Most tools do exactly what they were designed to do.
The issue is that they do it in isolation. In a silo.
A CRM was built to record, not to act. A texting platform was built to push messages, not to remember. None of these systems were built to talk to each other meaningfully because when most were architected, the connective tissue was a human. A counselor would log the call. A marketer would pull a list. The software didn’t have to be smart because the people were.
That worked for a long time. It doesn’t work anymore. In an AI world, it doesn’t just stop working; it actively breaks what you’re trying to build.
A typical four-year university runs between eight and fifteen separate systems that touch a prospective student: the CRM, SIS, CMS, marketing automation, texting, chatbot, name buy, application platform, analytics, and more. I’ve been on calls with VPEMs who can’t list their own stack from memory because there are too many vendors.
None of that was a mistake. Each tool was purchased for a real reason by someone solving a real problem at the time. If you stack good decisions for a decade, you end up with a Frankenstein nobody designed and everyone has to live with. It’s a real problem.
The crux of it all is what happens to the student in the middle of all that.
A high school junior fills out a form on Tuesday. That data goes into the CRM. On Thursday, she texts the school, but the texting platform doesn’t know about the form. On Saturday, she returns to the website and reads about the nursing program. The CMS doesn’t know what she texted. A week later, she gets an email from a list generated before any of this happened.
One student. The institution has effectively forgotten her three times.
That’s what I’ve started calling context decay. Every time a student moves between systems, context decays. The data still exists, but it’s stuck somewhere nothing else can use. The information is technically owned by the school but, functionally, is lost to the student. This is unacceptable today.
Before AI, context decay was an annoyance hidden in unexplained conversion rates. In an AI world, context decay limits how good your AI can get.
Most schools haven’t fully internalized that the AI itself isn’t the problem anymore. Even the best AI can only work with what you give it. If the AI doesn’t know the student filled out a form last Tuesday, texted on Thursday, and read about the nursing program on Saturday, it can’t write an email that sounds attentive. It writes a generic one, like the kind she’s already deleted from ten other schools.
Not because the AI is dumb. Because nobody handed it the backstory.
Personalization isn’t a model problem. It’s a context problem.
Your systems probably aren’t broken. They’re just operating in isolation, as designed in an era when isolation was acceptable. That era is ending fast. I believe the schools that pull ahead in the next few years won’t be those that bought the smartest AI tool. They’ll be the ones who solved the context problem first, then turned AI on.
The order matters. Context, then AI. Not the other way around.
If you’re spending money on AI without solving for context, you’re paying for the appearance of personalization. The students you’re trying to enroll use AI more than you do. Their filter for generic is sharper than any of us want to admit.