Children rarely leave school suddenly. They leave gradually, and the gradual part is legible in data a school already holds — a term of thinning attendance, a fee that stops being paid, marks sliding in two subjects, a phone number that no longer connects. Every one of those signals sits in a different screen, owned by a different person, noticed by nobody in combination.
The dropout early warning pack combines them once a week and puts the result in front of the principal.
What the Risk Score Is Built From
The scoring service reads five families of signal for each enrolled student in the current academic year:
Attendance patterns. Not just the headline percentage but the shape of it — how the rate has moved across the year, and whether absences cluster.
Fee payment history. Repeated overdue entries are among the most honest indicators a school has, because a family under financial pressure usually stops paying before it stops sending the child.
Academic performance. Results recorded in report cards, read as a trend rather than a snapshot.
Behaviour incidents. Records logged against the student.
Guardian contact data. Whether the school can actually reach the family. A missing or dead contact number is itself a risk factor, and one that no amount of automated messaging can route around.
Each factor contributes to a 0-100 score, and each contributing factor is returned alongside it. That last detail is what makes the output usable: a principal receives not “Student A: 71” but the reasons that produced 71.
Rule-Based Scoring, Not a Black Box Model
It is worth being precise about this, because the industry vocabulary is loose.
The calculator that the weekly scan reads is a rule-based weighted scoring engine. Defined rules, defined weights, defined bands. It is not a trained statistical model, it has not learned weights from historical outcomes at your school, and it does not claim predictive accuracy that has been measured against real dropouts.
That is a deliberate position rather than an interim one. A rule-based score is auditable — a teacher who disagrees with a flag can see exactly which rule fired and argue with it. A trained model’s score cannot be argued with in a staff meeting, which in a setting where the output affects a child’s treatment is a genuine drawback, not a technicality.
The scoring architecture allows an alternative algorithm to be selected later. What ships and runs today is the heuristic one.
Risk Bands and What Each One Should Trigger
Scores map to four bands:
| Score | Band | Reasonable response |
|---|---|---|
| 0–24 | Low | No action; the band exists to make the distribution readable |
| 25–49 | Medium | Class teacher notes it; watch the next month’s attendance |
| 50–74 | High | A phone call to the family this week, by a named person |
| 75–100 | Critical | Counsellor or principal involvement; treat as a live case |
The bands are worth agreeing on before the first scan runs. A school that has not decided in advance what “high” obliges anyone to do will receive its first list, feel briefly alarmed, and then file it — which is the usual fate of early-warning systems everywhere.
The Sunday Night Scan and Who Receives It
The pack runs a weekly scan, defaulting to Sunday at 20:00 in the school’s own timezone. The day and hour are settings. Sunday evening is chosen so the list is waiting at the start of the working week rather than arriving mid-Wednesday.
The scan applies a score cut-off and a maximum count — both configurable — and produces a digest for the principal listing the flagged students with their contributing factors and, unless suppressed, suggested interventions.
In agent mode the pack goes further. Deterministic SQL aggregates each candidate’s attendance trend over the last thirty days, fee default count, and marks trend across the last two exam terms. A single structured call assigns a risk level and a short reasoning line per student, and the pack dispatches an in-app notification to each flagged student’s class teacher plus a summary to the principal. The model reads a prepared summary; it does not query student records directly.
Turning a Flag Into an Intervention
The gap between a flagged list and a retained student is entirely human, and the system is honest about that.
What it can do is make the follow-up cheap. Because the flag arrives with its reasons, the conversation with the family starts from something specific — “Rohan has missed eleven days since the term began, is everything all right at home?” is a different call from “we noticed he’s on a list.”
What it deliberately does not do is message the parent. There is no automated at-risk notification to families, and that is the right default. A message telling a parent their child has been flagged as a dropout risk, arriving without a human voice attached, would do more harm than the flag does good. The pack notifies staff. Staff decide what the family hears.
Where a flag is driven by fee arrears, the fee follow-up pack is already handling that conversation on its own schedule and its own tone ladder. Where it is driven by absence, the attendance packs are already alerting the guardian. The dropout scan is the layer that notices when several of those are happening to the same child at once.
Accuracy Limits and the Alert-Only Setting
Two honest limitations.
The score is only as good as your data. A school that marks attendance irregularly, or records behaviour incidents only for serious cases, will get a score weighted towards whatever it does record consistently. In practice this usually means fee history dominates in the first term of use, which makes the list look like a defaulter list. That resolves as attendance and academic data accumulate.
Interventions are suggestions. The pack ships an alert-only setting precisely for this. With it on, the digest carries the flag and the reasons and stops there. Schools still calibrating their thresholds, or those with a counsellor who would rather form their own view, should run in alert-only mode and turn suggestions on later, if at all.
Both settings live with the rest of the pack’s thresholds, alongside the guardrails described in school autopilot: operations that run without you.