A long-established Israeli university was losing hundreds of first-year students to quiet dropout, and finding out only once they'd already failed or filed to leave. We moved the discovery weeks earlier, to when help still works.
A leading institution was losing hundreds of students (mostly in first year) to dropout, a serious hit to both its revenue from tuition and its reputation.
The trouble was that everyone found out too late. Course advisers and the dean of students only learned a student was in trouble once they'd failed the end-of-semester exams or handed in a withdrawal form, the point at which it's already too late to help. The signals existed, but they were scattered across separate systems: class attendance, logins to the learning platform, late-submitted assignments, payments held in the finance office. Nothing was reading them together.
One health profile per student, an early-warning model, and a nudge to a human while there's still time.
We connected the academic, operational and financial systems into a single data platform, and built each student a continuously updated "health profile" instead of scattered fragments no one could see at once.
The model watches the quiet early indicators, not the loud late ones (a 30% drop in academic-portal logins alongside a missed assignment), the pattern of a crisis building, weeks before the exams.
Rather than a post-mortem after a failed semester, the system flags a student drifting toward dropout while the semester is still live and the outcome can still be changed.
The moment a student is flagged, an alert goes to the faculty office or the course adviser, with a full picture and a suggested action, not just a red dot.
The recommendation fits the cause: a personal check-in, an offer of academic tutoring, or a payment-plan arrangement, targeted support for the students who actually need it.
The AI surfaces who's at risk and why; a person decides how to reach out. The relationship stays human. The model just makes sure it happens in time.
Retention moved from a post-mortem to a live, proactive process.
The first-year dropout rate fell by 25% in the first year of the rollout, keeping students on their courses, and keeping a large, recurring stream of tuition revenue the institution had been losing.
The dean's office and advisers moved from crisis management (putting out fires after the fact) to focused, proactive outreach to the students who genuinely need it, and students felt the institution actually saw them.
It works because it reads the leading signals instead of the lagging failure, and because a person always makes the outreach, so the intervention is both timely and humane. Prediction wired to a real, caring action is the difference between a dashboard of at-risk students and actually keeping them. The same discipline runs our churn-prediction work in another sector.
A large Israeli content provider was losing thousands of subscribers a month with no warning. Early churn prediction with automated, personalised retention cut monthly churn ~25% and lifted usage 40%.
Read the case → Lead scoring · FintechA global fintech's B2B reps drowned in unsorted leads while enterprise deals waited days and went cold. AI enrichment and predictive scoring lifted conversion 47.6% and cut strategic-lead response to 5 minutes.
Read the case →
Reef TRH