Higher education already has the data. It rarely has the answer.
We have built analytics for colleges and universities since 2018, from New Delhi. Our whole argument is that the hard part is not the model — it is agreeing what the numbers mean, then acting on them, then honestly checking whether it worked.
Four positions we argue for.
These are not values in the poster sense. They are choices that cost us money, which is the only way to tell whether a company means them.
A model fitted elsewhere is not evidence about your students
We measured the transfer penalty: roughly 19 points of precision lost when a model moves between institutions. So we fit on yours, which is slower and more expensive than shipping one model to everyone.
A null result is a result
We publish efficacy studies that find no effect, including on our own deployments. Two partner institutions saw no measurable change in year one and that is on our research pages.
Student data is not our asset
We are a processor under your instruction. No pooling across institutions, no benchmark product built from your records, no training models that serve someone else. This costs us deals with institutions who want the comparison.
The dictionary comes before the dashboard
If your registrar and your IR office do not agree what retention means, no amount of modelling will fix it. We pause projects at that step rather than proceeding on an unsigned definition.
Why we started.
In 2018 we were doing counselling and coaching work with students across Delhi, and the same thing kept happening. A college would tell us a student had dropped out “suddenly”. Then we would sit with the student and find four months of visible warning signs, each recorded in a different system, none of which had been looked at together.
The information existed. Attendance knew. The LMS knew. The fee office knew. Nobody had put them in one place, and nobody owned the act of noticing. That gap — between what an institution already knows and what it actually does — is the whole reason this company exists.
Every institution we have worked with was already collecting enough data to see the problem coming. Almost none of them were looking.
We started with data integration because that is where the projects were failing, not because it was interesting. The predictive modelling came later, and the research group came after that — once we realised we were making claims about effectiveness that we could not actually defend.
Today we serve institutions in India and fourteen other countries, we remain independent and deliberately small, and we publish our method so that a sceptical institutional researcher can take it apart. If you are that person, we would rather hear from you than not.
Lines we have decided not to cross.
Asked for often enough that it is worth writing down.
Automated academic penalties
We will not wire a risk score to a consequence — probation, funding withdrawal, or removal from a programme. A prediction is a prompt for a human conversation, not a verdict.
Cross-institution student tracking
We will not follow a student between institutions or build an inter-institutional record. Every deployment is isolated, which means we cannot offer that and will not.
Sentiment scoring of individual students
We do topic analysis on aggregate feedback. We will not produce a per-student sentiment score from written work or messages — the error rate is high and the harm from being wrong is concentrated on the most vulnerable students.
Come and disagree with us.
We are most useful to institutions that push back. If you have read our method and think it is wrong, that is the conversation we want.