Blog
Students who leave usually signal it a semester in advance
Universities tend to learn who dropped out when they drop out, though the data said so much earlier.
Published
Most universities run on an assumption they have never examined: that students who leave leave because their grades fell.
The data usually says otherwise. Falling grades are the last symptom. What comes first is disappearing from certain classes, not submitting the small pieces of work nobody chases, logging out of the learning platform around mid-semester, and not answering the adviser's email.
Those signals appear weeks before any grade does, and they appear in systems the university already runs. They are simply scattered across separate systems, and nobody looks at them together.
Where the data sits
The registry knows about enrolment and grades. The learning platform knows about logins and submissions. The card system knows about building access. Student affairs knows about activities. Finance knows about unpaid fees.
No system sees the whole picture — and the academic adviser, who should, has the least of all, because they see only what the student chooses to tell them.
What a risk model can do
Combining those signals into a risk score is not technically hard. The hard part is deciding who sees the score, and what they are expected to do about it.
A system that works does not send a list of at-risk students to management. It sends it to the adviser with the reasoning attached — missed three classes across two weeks, no platform login in ten days — and leaves the adviser to decide how to open the conversation.
The line not to cross
A risk score is a dangerous instrument in the wrong place.
Used in admissions, it becomes automated exclusion, because a model learning from the past learns the past's biases too. Students from particular schools or provinces get scored as higher risk for reasons that have nothing to do with them.
Shown to the student, it becomes a self-fulfilling prophecy.
Set the rule at the start: the score allocates support, never opportunity. And review it periodically to see which groups it is scoring high, because a model nobody audits will drift somewhere nobody intended.
What often beats the model
At many institutions the biggest gain does not come from prediction at all. It comes from putting what already exists in front of the adviser in one place.
A single screen showing which of your advisees are missing classes, which are not submitting, which have unpaid fees, with a box to record what you have already discussed, turns advising from an annual meeting into something continuous. No AI required.
What GIPSIC does here
This is mostly integration work. Registry systems are usually old and cannot be changed, so the job is extracting data without disturbing what runs, then building a layer other systems can be built on.
We work waterfall when the scope has to be fixed and agile when the problem still needs iteration. Education integration is usually both: the side touching the legacy system has to be pinned down, while the side people use will change in the first month once advisers get their hands on it.
If you would like to talk about student data at your institution, get in touch.
About the author
Film — Wisit. A businessman who still does his own BA work more often than he probably should, and writes a fair bit of code, front and back. Runs two or three small businesses. Follows technology and business obsessively, in Thailand and everywhere else. Off the clock: physics, astronomy, DIY, and anything to do with networks. Music always on, though he cannot sing. Plays instruments anyway, badly. Plays a lot of sport, racket sports above all. Not much of a traveller by himself, but happy to take Mint anywhere in the world. A man who fears — sorry, loves — his wife. One flaw: he barely touches video games.
Written with Claude Opus 5