The dashboard reached the tutor meeting with a reassuring look of precision: a red line beside Maya's name, sixteen assignments submitted after midnight, and three weeks of irregular logins. Its label read “disengagement risk.” No one had asked what was happening. Her father was sleeping at the hospital with her grandmother, and Maya was making dinner, replying to relatives, and waiting for the apartment to go quiet before opening her laptop. The indicator could see hours, clicks, and lateness. It could not see care work, exhaustion, or the choice to keep studying when there was almost no energy left.
That scene does not prove school analytics are useless. A pattern can tell adults to look more closely. The trouble begins when a clue becomes a verdict: low commitment, poor habits, limited ability. A system that assigns a private cause to platform traces mistakes an administrative signal for a description of a person. When that mistake shapes calls home, sanctions, or teacher expectations, the school adds pressure exactly where it should be offering support.
What a screen records, and what it cannot conclude#
Platforms collect data in order to work: login time, opened exercises, submissions, messages, a camera switched off in class. Using it to notice change and offer help can be sensible. But every data point sits at a distance from the question that matters. A late submission does not explain why; opening a resource briefly does not measure understanding; working at night does not reveal a preference. The same pattern might describe a student with a job, a young person without a private room, someone with unstable internet, or a child caring for an ill relative.
Platforms collect data in order to work: login time, opened exercises, submissions, messages, a camera switched off in class.
That gap matters because schools have authority. When an alert is presented as objective, it can replace a conversation with an automatic response: removal from an activity, a discipline referral, lower expectations, or demands for explanations that intrude on family life. Maya did not need to prove that her circumstances were serious enough to deserve compassion. She needed an incomplete observation not to become guilt before she had a chance to speak.
It is equally important to separate performance from availability. A teenager can keep high marks while losing sleep, and a temporary decline can be an understandable response to a crisis. Analytics often rewards regularity because regularity is easy to count. Education should protect the possibility of learning under unequal conditions. Turning regularity into a moral virtue punishes students who do not control their time, space, or household care responsibilities.
Replace the alert with a safe invitation#
The first design decision is both linguistic and practical: an alert should say, “Check whether support would help,” not, “You are disengaged.” That wording requires the notice to reach a named adult who is trained and has time to listen, rather than a chain of automations. The conversation must be possible without requiring a student to disclose diagnoses, family conflict, or details they do not wish to share. It is enough to ask which obstacle is interfering and which adjustment would make it possible to continue.
For Maya, a tutor could offer three choices without asking for a confession: an agreed extension, downloadable material, and a reviewable weekly plan. If she wants to explain the care work, she can. If she does not, the school can act on what it needs to know: a temporary burden makes the usual timetable unworkable. Sensitive information should never be the admission fee for a reasonable adjustment.
A safe conversation also includes a right to challenge the record. Maya should be able to say that the label does not describe her situation, know who can see it, and request correction or removal of a mistaken interpretation. This is not a minor administrative point. If a school file preserves “low motivation” where there was exhaustion from care, that phrase can travel into future meetings and change how other adults read every later difficulty.
Support without turning help into surveillance#
An agreed support should address a specific barrier, not open a surveillance case. An extension should not require nightly screenshots. A device loan should not become permission to observe all domestic activity. A tutor can agree learning priorities without asking a family to expose its private routines. Proportionality always asks: what is the minimum information needed to offer this support, and when is it no longer legitimate to keep it?
It helps to prepare a visible menu of adjustments before any alert appears: flexible deadlines, divided tasks, asynchronous materials, one contact person, connectivity loans, voluntary coordination with wellbeing staff, and a route back to the ordinary timetable. Visible alternatives prevent every teacher from improvising and prevent access to support from depending on how eloquently a family can explain itself under pressure.
Support should not isolate a student either. Some students want only the tutor to know; others need coordination across teachers so they do not repeat the same explanation five times. As far as possible, that choice belongs to them. The school may record the adjustment, review date, and responsible person, but it does not need an exhaustive narrative of the hardship. Documenting a decision is not the same as taking ownership of someone’s story.
Review the system with people who live its effects#
A school can learn by examining past alerts in aggregate. How many led to useful support? How many rested on mistaken interpretations? Which groups receive labels most often? Those questions reveal bias that an overall average hides. If alerts cluster around students who share devices, care for siblings, or rely on long journeys, the pattern is not an individual deficit. It is evidence that the model is confusing unequal conditions with disengagement.
Students and families must be part of that review, not merely the vendor and technical staff. They may explain that a notification sent home raises tension, that a task calendar ignores care-heavy weekends, or that a compulsory video call is not private. Their experience is not an anecdote to append to a report. It is evidence about what the data cannot capture.
Clear boundaries matter too. A tool should not trigger sanctions, decide progression, or send a reproach on its own. Every contact needs a human route and a reasonable timeframe. If staff cannot explain in ordinary language what the system observed, what it does not know, and what support is being offered, they are not ready to use that alert with a real person.
A more honest measure of success#
Success is not a lower count of red marks. A school could obtain that result by lowering thresholds or encouraging students to leave the platform open. The better question is whether people facing difficulty could continue learning without paying for it through extra guilt, exposure, or withdrawal. Answering it requires simple measures—recovered deadlines, sustained participation, resolved requests—alongside voluntary conversations about whether support arrived in time and whether the process felt respectful.
After several weeks, Maya may find a different routine; she may not. What matters is that the school has not used her exhaustion as evidence of incapacity. A follow-up note can say, “materials and deadlines agreed; review on 15 May,” rather than “unmotivated student.” The first phrase leaves room for learning and change. The second fixes an identity no graph was qualified to grant.
The immediate action is small and specific: review the next alert before it is sent, remove any label that assigns intention, and prepare one support question alongside two possible alternatives. Analytics can direct attention, but it must not close the conversation. When a school acknowledges what it does not know, data stops acting like a sentence and can finally become an opening for care.
