Every online academy sits on a data source it rarely reads: the attendance record. Not the polished monthly report — the raw pattern of who joined which live class, and when. Across the academies that run on Clasify, the same handful of patterns show up again and again, and each one says something useful about a student weeks before anything dramatic happens. Here is what the attendance data keeps telling us. Attendance is a leading signal, not a lagging one Most academies treat attendance as a record — something you check after the term to see who showed up. Read it that way and it is history. Read it week to week and it is a forecast. The single most consistent pattern we see: students almost never leave suddenly. Their attendance slides first. A learner who was joining every session starts missing one in three, then one in two, and only then disappears from the roll entirely. By the time a student formally leaves, the attendance data usually flagged them weeks earlier. The academies that keep students are the ones that were reading it. The slow fade is more common than the clean break There are two shapes of churn, and they are not equally common. The clean break — a student who attends fully and then stops dead — is rare and usually external: a move, a job, money. Far more common is the slow fade: attendance drifting down over several weeks. That matters because the slow fade is the one you can catch. It leaves a trail, and the trail is visible in the data long before the student has decided anything for certain. A sudden stop is usually a life event you could not have prevented. A gradual slide is usually reversible — if someone notices while it is happening. The difference between the two is only visible if you watch the trend, not the last lesson. Excused absences are a different signal — treat them separately One pattern quietly corrupts everyone's attendance data: counting an excused absence like a no-show. A student who was ill and told you is behaving completely differently from one who simply did not turn up, yet a naive attendance rate scores them identically. We keep excused absences out of the rate for exactly this reason — otherwise a healthy, engaged cohort with a normal amount of illness reads as if it is disengaging, and you end up chasing students who were never at risk. When the rate is honest, the students it flags are the ones who actually need a message. The first weeks of attendance predict the rest Another consistent pattern: early attendance is disproportionately predictive. A student's behaviour in their first few sessions tends to set the tone. Someone who joins their first lessons on time, from the start, usually keeps going. Someone who misses one of their earliest sessions — often before the teaching ever had a chance — is at elevated risk for the whole enrolment. The practical lesson is that the cheapest retention work you will ever do happens in week one, not month three. A dip is not always disengagement Before you read a falling number as a student losing interest, rule out the dull explanations: a forgotten session, a time-zone mix-up, a join that took one tap too many. A meaningful share of absences are logistics, not motivation — which is why the honest move is to design that friction out first, and only then treat a persistent slide as a signal about the student. The attendance number is only trustworthy once the forgettable misses have been removed from it. What to do with your attendance data None of this requires a data team. It requires reading the attendance you already collect and acting on it in order: Watch the trend per student, not just the latest session. Keep excused absences out of the rate so the signal stays clean. Treat a two-week slide as a prompt to reach out, not a number to file. Put your attention on new students' first weeks. Remove the friction — auto-provisioned rooms, reminders, browser joining — before assuming it is motivation. Attendance is the cheapest early-warning system a school owns, and almost nobody uses it as one. The academies that grow are rarely the ones with the loudest marketing. Often they are just the ones who noticed the slide in week two and sent the message a human would have sent — if a human had been watching the number.