How Education Works | Dropout Early Warning, Student Re-Engagement & Return-to-Learning — How Systems Act Before Disengagement Becomes Educational Exit

HEW-NODE-0212 · How Education Works · Dropout early warning, student re-engagement and return-to-learning

Most students do not leave education in one clean moment.

They begin arriving late. Attendance becomes irregular. Work is missed. A difficult subject turns into avoidance. A family loses income. Transport becomes unreliable. A learner starts caring for a younger sibling. Pregnancy, illness, disability, bullying, conflict, displacement, discipline, work, housing instability or a growing sense that school has no route to a meaningful future may gradually change the cost of continuing.

By the time a student is formally recorded as having dropped out, the education system may have had weeks or months of earlier signals.

This is the job of dropout early warning, student re-engagement and return-to-learning systems: to notice weakening participation early, understand why it is happening, coordinate proportionate support, and preserve a credible route back when a learner has already left.

This node has a deliberate boundary. School Attendance owns the daily machinery that turns enrolment into presence. Tiered Academic Intervention & Progress Monitoring owns escalating academic support inside school. Flexible Learning Pathways & Second-Chance Education owns the wider architecture of alternative learning routes. Learning Recovery & Acceleration owns repair of unfinished learning. This page owns the system that connects the warning signal to action, follows the individual case, prevents avoidable exit, and manages the return journey when disengagement has already become absence from education.

Quick Answer

Define what counts as disengagement and dropout → identify a small set of useful early signals → make the data timely enough to act → distinguish prediction from proof → assign a responsible adult or team → contact the learner and family → identify the actual barrier → select support that addresses that barrier → record action and response → monitor whether participation improves → escalate when risk persists → coordinate education, social, health and protection services where needed → preserve learning during interruption → maintain a return route → recognise prior learning → create a re-entry plan → monitor the first weeks back → learn from patterns across cases → change the system when the same preventable barrier repeatedly pushes learners out.

The central principle is simple: an early-warning system is not a dropout-prediction machine. It is a decision system whose value depends on what happens after the warning.

Dropout Is an Outcome; Disengagement Is a Process

A learner can remain officially enrolled while participation has already collapsed.

Attendance may be technically high enough to avoid a formal trigger while the student skips particular lessons. Assignments may still be submitted but only after repeated prompting. A learner may be physically present and psychologically absent. Another may want to continue but lack transport or childcare.

Systems therefore need to understand educational exit as a process with several possible states: stable participation, emerging risk, persistent disengagement, interrupted participation, formal dropout, alternative provision, re-entry and sustained return.

Not Every Absence Is a Dropout Signal

A student with influenza may miss a week and return without any long-term risk. Another learner may miss one day every week because of work or caregiving and be on a much more serious trajectory.

Early warning works by combining patterns and context, not by turning every absence into an alarm.

The Most Useful Signals Are Usually Close to Participation

Common signal families include attendance, academic performance, behaviour or disciplinary events, progression history, mobility, school transitions and known vulnerability or support needs.

UNICEF’s longstanding Monitoring Education Participation framework groups major dropout-risk indicators around academic performance, behaviour, chronic absenteeism, disability and entry or progression difficulties. The exact indicators should be validated locally rather than copied mechanically.

Attendance Is Often the Earliest Visible Signal

Attendance data can show chronic absence, increasing absence, consecutive days missed, subject-specific absence or unexplained gaps.

The pattern matters. A learner whose attendance falls from 96% to 83% over six weeks may deserve attention even if the annual figure still looks respectable.

Trend Can Matter More Than Threshold

Fixed thresholds are easy to administer. They can be late.

A system can combine thresholds with change detection: sudden deterioration, repeated Monday absence, rising late arrival, missed assessments or a sequence of small disruptions that is unusual for that learner.

Academic Signals Need Careful Interpretation

Falling grades may indicate disengagement. They may also reflect curriculum difficulty, language barriers, poor assessment design or ineffective instruction.

A low mark should trigger inquiry rather than a label. The early-warning question is not “Is this student weak?” but “Has something changed that increases the risk of leaving, and what support follows from that?”

Behavioural Events Can Be Signals and Causes

Repeated discipline events can signal conflict, unmet support needs, bullying, frustration or weakening belonging. Exclusionary discipline can itself increase separation from school.

This is why the neighbouring School Discipline, Suspension, Exclusion & Reintegration node matters: behaviour management and dropout prevention are connected but not identical jobs.

Transitions Are High-Risk Points

Moving from primary to secondary school, lower to upper secondary, school to vocational education, one district to another or one country to another can disrupt identity, friendships, transport, curriculum continuity and support.

A student who was stable in the old institution may become at risk in the transition even without any change in personal motivation.

Early Warning Needs Timely Data, Not Perfect Historical Data

A beautifully cleaned annual dataset that arrives after the school year cannot prevent a learner from leaving in October.

Operational early warning usually needs current attendance and case information, with enough quality to support action. More complex administrative or socio-economic data can improve understanding but should not delay basic intervention.

Simple Rules Can Be Useful

Not every system needs artificial intelligence.

A rule such as “three consecutive unexplained absences, attendance decline of ten percentage points, or two failed core subjects plus attendance decline” may identify a manageable caseload. Its accuracy can be evaluated and refined.

Simplicity has operational value: staff understand why a learner was flagged and what evidence to verify.

Prediction Models Can Improve Prioritisation

Where high-quality longitudinal data exists, statistical or machine-learning models can combine many weak signals into a stronger prediction.

The World Bank’s work in Guatemala and Honduras demonstrates how existing education management information systems can be used to identify learners at greater risk of dropout. Its guidance also stresses that predictions must be communicated carefully, responsibilities assigned, local interventions enabled and models iteratively evaluated.

A Risk Score Is Not a Diagnosis

A model may correctly identify that a learner resembles previous dropouts without knowing why.

The next step is human inquiry: transport? bullying? work? health? curriculum failure? family crisis? school climate? disability support? financial pressure? The intervention should target the mechanism, not the score.

False Positives Are Not Free

If too many students are flagged, counsellors and teachers become overwhelmed. Learners may be stigmatized. Families may receive unnecessary calls. Staff may stop trusting the system.

Thresholds should therefore reflect available intervention capacity, not only predictive accuracy.

False Negatives Matter Too

A system can miss students whose risk is not visible in administrative data: violence at home, undocumented work, sudden mental-health difficulty or discrimination.

Teacher, counsellor, family and student referrals should remain available even when a model says “low risk.”

Fairness Testing Belongs in Predictive Systems

If a model uses disability, poverty, family structure or discipline history, it can encode structural inequality into risk scores.

The question is not only whether the model predicts dropout. It is whether it systematically over-flags or under-flags particular groups, whether staff interpret scores fairly and whether support follows without punitive consequences.

Malaysia Provides a Current National Example

UNICEF reported in June 2026 that Malaysia’s Ministry of Education and UNICEF developed a national early-warning platform using seven indicators, including attendance, academic achievement, disciplinary record, disability or learning-support needs, household income, distance from school and parents’ marital status. The platform is embedded across more than 10,000 schools.

UNICEF reported that in 2025 the system’s targeted interventions enabled more than 9,000 flagged students to return to school. The important design lesson is not merely the use of AI; it is the integration of prediction with educators, training, cross-divisional workflows and an intervention cycle.

Prediction Without Intervention Is Administrative Theatre

A dashboard that turns red does not re-engage a learner.

Every alert should have an operational destination: class teacher, attendance lead, counsellor, social worker, year head, district officer or multidisciplinary team. The responsible role should know what to do next and by when.

Case Ownership Prevents Everyone From Assuming Someone Else Acted

A learner may touch attendance, welfare, special education, health and social-protection teams. Without a named case owner, multiple people can observe risk while nobody coordinates the response.

Case ownership does not mean one person does everything. It means one person is accountable for knowing whether the system is doing anything coherent.

The First Contact Should Seek Understanding, Not Accusation

“Why are you absent again?” can begin as blame.

A better opening is: “We noticed you have missed more school than usual. Is something making it difficult to attend or continue?”

Families and learners often hold information the database cannot see.

Barrier Classification Helps Match the Response

• financial: fees, transport, food, work pressure;
• academic: severe learning gaps, repeated failure, language barriers;
• health: physical or mental-health difficulty;
• care: sibling, child or elder responsibilities;
• safety: bullying, violence, harassment, unsafe travel;
• school climate: weak belonging, discrimination, teacher conflict;
• discipline: suspension, exclusion or repeated sanctions;
• mobility: migration, displacement, homelessness;
• administrative: missing documentation, transfer failure, registration barrier;
• relevance: learner sees no credible pathway from school to future study or work.

A case can contain several barriers at once.

The Intervention Should Change the Barrier

If transport is the problem, tutoring does not solve it. If bullying is the problem, an attendance warning can make things worse. If the learner cannot read the current curriculum, a motivational speech is unlikely to be enough.

Early-warning systems become useful when they shorten the distance between signal and mechanism-specific support.

Academic Catch-Up Often Matters

Absence creates learning gaps, and learning gaps create more absence. Returning to a classroom that has moved far ahead can make re-entry humiliating.

Targeted tutoring, accelerated learning, adjusted sequencing or temporary additional support can break this loop.

Financial Support Can Be a Retention Intervention

Transport grants, fee waivers, meals, emergency assistance or cash transfers may directly affect attendance when household finances are the barrier.

The neighbouring Student Financial Aid, Grants, Scholarships & Means Testing node owns the finance machinery. Early-warning systems should know how to route an eligible learner into it.

School Health and Dropout Prevention Intersect

Untreated vision problems, chronic illness, mental-health difficulty, reproductive-health needs or disability barriers can weaken participation.

The School Health Services, Screening, Referral & Care Coordination node owns health-service delivery. The dropout system’s job is to recognize when health is affecting education and connect the learner without turning teachers into clinicians.

Pregnancy and Parenthood Require a Continuity Route

Where law and policy permit continued or resumed education, systems need practical arrangements for absence, health appointments, childcare, flexible scheduling and return.

A formal right to re-enter is weak if no school knows how to implement it.

Disability Support Can Be the Difference Between Enrolment and Participation

A learner may appear to be disengaging when the real problem is inaccessible transport, missing assistive technology, sensory overload or an unmet learning support need.

Risk review should therefore ask whether the environment is failing the learner before assuming the learner is failing the environment.

School Belonging Is Operational, Not Sentimental

Students are more likely to remain where they are known, expected and able to imagine success.

Advisory systems, mentoring, clubs, student voice, stable adult relationships and respectful discipline can support belonging. These are not substitutes for rigorous teaching; they help keep learners in contact with it.

Career Guidance Can Restore Relevance

Some adolescents disengage because school feels disconnected from future possibilities.

Good guidance can make vocational, academic, apprenticeship and employment-linked routes visible. The purpose is not to push a struggling learner into a lower-status pathway. It is to reconnect current effort with credible options.

Family Engagement Should Be Specific

Families can help solve transport, routines, communication and expectations. They can also be part of the difficulty when there is conflict, neglect or pressure to work.

Engagement should therefore be respectful and problem-specific rather than assuming every dropout risk can be repaired through stronger parental control.

Escalation Needs Clear Decision Rules

What happens when the first intervention fails?

A system can define escalation based on continued absence, repeated non-response, safeguarding concern, imminent withdrawal or a combination of needs requiring multi-agency support.

Multi-Agency Work Needs Consent and Role Clarity

Health, social protection, child protection, housing, youth services and employment agencies may all be relevant.

Data sharing should follow law and purpose. “Helping the child” is not a licence to circulate every record to every agency.

Case Conferences Should End With Action

A meeting that produces concern but no owner, deadline or next action is not coordination.

Barrier → action → owner → deadline → evidence of completion → review date.

Re-Engagement Begins Before Formal Re-Entry

A learner who has been out of school for months may need contact, trust-building, assessment, documentation support, transport, counselling and a realistic learning plan before returning to a timetable.

“Come back Monday” is not always a re-engagement strategy.

Second-Chance Pathways Prevent One Exit From Becoming Permanent

UNICEF evaluations of flexible pathways show the value of accelerated and second-chance programmes for out-of-school children and adolescents, including compressed curricula, flexible schedules and routes back into formal education.

The important system property is permeability: learners should be able to move from alternative or second-chance provision back into recognised education rather than entering a dead end.

Recognition of Prior Learning Reduces the Cost of Return

A learner should not automatically restart an entire year because participation was interrupted.

Assessment, credit recognition or competency review can establish what has already been learned and what remains.

Re-Entry Needs a Named Start Point

Who receives the returning learner? Which class? Which subjects? Which adult checks in? What learning has been missed? What explanation is given to teachers? What information remains private?

A re-entry plan converts goodwill into operations.

The First Month Back Is a High-Risk Period

A learner may return and leave again if the original barrier remains or the academic gap feels overwhelming.

Return-to-learning plans should include scheduled follow-up rather than assuming re-enrolment means the problem is solved.

Retention After Return Is a Better Outcome Than Re-Enrolment Alone

A system that counts every returned learner as success on day one can overstate effectiveness.

Useful outcomes include sustained attendance after four weeks, one term and one year; credit accumulation; learning progress; completion; and whether the learner remains in any recognised education or training pathway.

Costa Rica Shows What a System-Level Early-Warning Approach Can Look Like

UNESCO reported in 2025 on Costa Rica’s work against educational exclusion, where early-warning systems are used to identify risk, learning difficulty and contextual barriers so schools can activate specific responses. UNESCO notes a large long-run reduction in the country’s secondary out-of-school population, while the operational lesson is the integration of warning, school action and wider inclusion policy rather than any claim that one tool caused the whole change.

Guatemala Offers Experimental Evidence on a Low-Cost Model

A World Bank-supported randomized evaluation across roughly 4,000 schools tested training, guidance and risk information for the transition from primary to lower secondary education.

The World Bank reports that the programme reduced dropout by 1.3 percentage points among schools assigned to the intervention and by 3 percentage points among programme compliers, at an estimated cost below US$3 per student. The result is useful because it shows that early warning does not have to begin as an expensive technology programme.

Peru Shows Why Usage Is Not the Same as Impact

World Bank reporting on Peru’s Alerta Escuela notes that reminders increased access and use of the early-warning system but did not necessarily increase preventive action or reduce dropout.

This is one of the most important lessons in the field: adoption metrics such as logins, alerts viewed or reports downloaded are intermediate outcomes. The system succeeds only if useful action reaches learners.

Early Warning Should Minimise Stigma

Labels such as “likely dropout” can change how adults interpret a learner’s behaviour.

Risk information should be restricted to staff who need it, framed as a support signal and removed or updated when risk changes. A student should not carry a permanent identity created by a temporary model prediction.

Students Should Know When Data Is Used to Support Them

Transparency can explain which kinds of information are monitored, why, who can see it and what happens after an alert.

Where models use sensitive household or disability data, governance should be especially clear.

Data Minimisation Matters

More variables do not automatically create a better system.

If attendance and progression data already identify most actionable risk, collecting intrusive family information may add little value. Systems should test the incremental benefit of sensitive variables against privacy and fairness costs.

Alerts Need Expiry

A learner flagged six months ago may now be stable.

Risk status should refresh with new data and close when the case is resolved. Permanent red flags create stale decisions.

Intervention Capacity Determines the Useful Alert Volume

If a counsellor can manage twenty intensive cases, a system that produces two hundred high-risk alerts has not solved prioritisation.

Capacity planning should connect prediction thresholds to staff, referral routes and support budgets.

Caseloads Need Tiers

Some learners need a simple attendance conversation. Others need coordinated social, financial and health support.

• light-touch: automated or teacher contact, reminder, clarification;
• targeted: mentor, tutoring, transport support, family meeting;
• intensive: multidisciplinary case management and sustained follow-up.

Tiering helps scarce specialist capacity reach the cases that need it most.

Staff Need Training in Response, Not Just Software

A sophisticated dashboard in the hands of staff who do not know how to have a supportive conversation can do harm.

Training should cover risk interpretation, bias, confidentiality, interviewing, referral pathways, documentation and when safeguarding obligations override ordinary consent.

Schools Need a Directory of Actual Supports

Staff cannot refer a learner to a service that does not exist or whose eligibility rules are unknown.

An intervention directory can list transport assistance, tutoring, health services, social protection, alternative provision, disability support, counselling, childcare and re-entry programmes with contact and eligibility details.

Referral Completion Should Be Tracked

“Referred to counselling” is not the same as receiving counselling.

Closed-loop referral tracks whether the learner reached the service, whether the service accepted the case and whether the support changed the barrier.

System-Level Analysis Should Look for Repeated Causes

If one learner leaves because the bus route was removed, that is a case. If fifty learners in the same district leave for transport reasons, that is a transport policy problem.

Early-warning data should feed system improvement, not only individual remediation.

Geographic Clusters Can Reveal Access Failures

Mapping disengagement by neighbourhood can reveal travel distance, unsafe routes, seasonal work, school capacity or local economic shocks.

Aggregate analysis should preserve privacy while making structural patterns visible.

School-Level Variation Deserves Investigation

If similar schools serving similar populations have very different dropout rates, the difference may reveal school climate, transition practices, attendance follow-up, curriculum fit or data-quality issues.

Variation is a prompt for inquiry, not automatic blame.

Model Performance Should Be Monitored Over Time

A model trained before a pandemic, policy reform or major economic shift may lose accuracy.

Prediction systems need recalibration, drift monitoring and periodic comparison with simpler rules.

The Best Model Is the One That Improves Decisions at Acceptable Cost

A complex model that improves prediction by one percentage point but requires expensive infrastructure and is impossible for staff to interpret may be worse than a simpler model that is trusted and acted upon.

Technical sophistication is a means, not the educational outcome.

Worked Case: Attendance Begins to Slide

A Grade 8 student’s attendance falls from 95% to 78% over six weeks. Grades remain stable, so the annual academic report does not flag concern.

The early-warning system detects the attendance trend. A year head contacts the family and learns that the student is walking a younger sibling to a different school after a transport change.

The intervention is transport coordination and timetable flexibility for two weeks while the family resolves the route. Attendance returns above 90%.

No tutoring was needed because tutoring was never the problem.

Worked Case: The Model Flags the Wrong Reason

A risk model flags a learner because of low grades, previous discipline events and household income.

The case conversation reveals persistent bullying rather than academic disengagement. The school activates its safeguarding and bullying response, changes supervision and establishes a trusted-adult check-in.

The model identified risk correctly but could not diagnose the cause.

Worked Case: A Student Has Already Left

A 16-year-old has been out of school for five months and is working irregularly.

A re-engagement team makes contact through a youth service, reviews prior learning and identifies that a conventional full-day timetable conflicts with family income needs. The learner enters a flexible second-chance programme with evening study and a vocational pathway, then transitions into a recognised apprenticeship route.

Return-to-learning succeeds because the system offers more than one shape of education.

Worked Case: A Returned Learner Leaves Again

A student returns after a long illness, is placed directly into the usual class and receives no academic catch-up. Within three weeks, absence resumes.

The school redesigns re-entry: baseline assessment, a six-week catch-up plan, reduced non-essential workload, peer support and weekly review.

The important change is not “try harder.” It is changing the return environment.

Worked Case: Too Many Alerts

A district launches a model that flags 38% of secondary students as high risk. Counsellors cannot respond meaningfully.

The district recalibrates thresholds, introduces medium-risk automated supports and reserves intensive case management for the top risk tier plus staff referrals. Predictive accuracy falls slightly, but intervention completion rises sharply.

Operational usefulness improves because the warning system is connected to capacity.

Worked Case: The School Solves Cases but Misses the Pattern

Multiple students from one rural area receive individual attendance interventions. Months later, the district notices that all cases cite the same unreliable bus service.

The district changes the route contract. Dozens of future cases never occur.

A strong early-warning system eventually makes some of its own alerts unnecessary.

Failure Mode: The Dashboard Becomes the Intervention

The repair is named case ownership, action deadlines and tracking of whether support actually reached the learner.

Failure Mode: One Threshold Defines Dropout Risk

The repair is to combine pattern, trend and contextual referral rather than wait for one annual percentage.

Failure Mode: Prediction Is Treated as Diagnosis

The repair is a conversation and barrier assessment before selecting support.

Failure Mode: Students Are Permanently Labelled High Risk

The repair is time-limited risk status, restricted access and refresh when circumstances change.

Failure Mode: The Model Is Accurate but No Services Exist

The repair is intervention-capacity planning, service directories and referral funding alongside analytics.

Failure Mode: Staff Receive Software Training but No Case-Management Training

The repair is preparation in supportive contact, confidentiality, referral, bias and follow-up.

Failure Mode: Re-Enrolment Is Counted as Success on Day One

The repair is retention measures after four weeks, one term and longer, plus learning and pathway continuity.

Failure Mode: Everyone Receives the Same Intervention

The repair is barrier-specific response rather than a universal attendance lecture.

Failure Mode: Families Are Treated as the Problem Before Being Heard

The repair is inquiry first, responsibility second, and safeguarding escalation where evidence requires it.

Failure Mode: The System Fixes Individual Cases but Repeats the Structural Cause

The repair is aggregate analysis that turns recurring transport, curriculum, discipline or service barriers into system reform.

What a Strong Dropout Early-Warning and Re-Engagement System Should Be Able to Answer

• What operational definition of dropout does the system use?
• Which states exist before formal dropout?
• Which indicators are available frequently enough to act?
• Which indicators have been validated locally?
• Are trends visible as well as thresholds?
• How are school transitions monitored?
• Can staff refer a learner even when no model flag exists?
• What false-positive rate is operationally tolerable?
• Which groups are over- or under-flagged?
• Who receives an alert?
• Who owns the case?
• How quickly must first contact occur?
• How is the learner’s perspective captured?
• How is the family’s perspective captured?
• What barrier categories guide response?
• What interventions exist for each barrier?
• How are financial supports accessed?
• How are health needs referred?
• How are disability supports activated?
• How are bullying and safety concerns escalated?
• How is academic catch-up provided?
• When does a case require multidisciplinary support?
• What information may be shared with other agencies?
• How is referral completion confirmed?
• What happens if the first intervention fails?
• When is a learner considered re-engaged?
• What flexible or second-chance routes exist?
• Can prior learning be recognised?
• Who owns the return-to-school plan?
• How is the first month after return monitored?
• How is sustained retention measured?
• Can a learner re-enter more than once without stigma?
• How are risk labels expired?
• How is model drift monitored?
• Does a complex model outperform simple rules enough to justify its cost?
• How are privacy and data minimisation protected?
• Which schools or regions show unusual dropout patterns?
• Which recurring causes require system-level change?
• How does the programme measure intervention completion, not just alert generation?
• Can the system show that its support reaches learners before formal exit?

A Practical Early-Warning Control Loop

Observe participation → detect meaningful change → verify the signal → assign case owner → talk to learner and family → diagnose barrier → select proportionate support → deliver support → confirm referral → monitor response → escalate if needed → preserve learning during interruption → create a return route → monitor re-entry → close case when participation stabilises → analyse repeated causes → redesign the system so preventable risk occurs less often.

How This Node Connects to the Wider Education System

Dropout prevention sits at the intersection of learning, attendance, welfare, access, health, finance, transport, discipline and future pathways. It is therefore one of the clearest examples of why education systems need both data and human judgment: the data can tell the system where to look, but only people and services can change what the learner is living through.

Useful neighbouring routes include the main How Education Works hub; School Attendance; Tiered Academic Intervention & Progress Monitoring; Flexible Learning Pathways & Second-Chance Education; Learning Recovery & Acceleration; Student Financial Aid; and Educational Equity.

Frequently Asked Questions

What is a school dropout early-warning system?

It is a process that uses timely participation, learning or contextual indicators to identify learners whose risk of leaving education may be increasing, then routes that signal into human review and support.

Does an early-warning system need artificial intelligence?

No. Simple attendance and progression rules can be effective when they are timely, understood and connected to action. More complex models can improve prioritisation where data quality and operational capacity justify them.

What is the most important feature of an early-warning system?

The intervention pathway. A highly accurate prediction that produces no useful support has little educational value.

How should schools treat students flagged as high risk?

As learners needing inquiry and possible support, not as inevitable dropouts. Risk scores should not become permanent labels or punitive evidence.

What if the learner has already left?

The system shifts from prevention to re-engagement: contact, barrier assessment, flexible or second-chance routes, recognition of prior learning and a monitored return plan.

Sources and Further Reading

• UNICEF — Future Ready and Resilient: How AI Is Keeping Malaysian Students in School, 1 June 2026. Current national example linking prediction, educators, workflows and targeted intervention.
• UNESCO — Early Warnings in Schools: Costa Rica’s Work Against Educational Exclusion, 26 June 2025. System-level example of early warning and educational inclusion.
• World Bank — Guatemala: Scalable Early Warning Systems for School Dropout Prevention. Randomized evaluation of a large-scale, low-cost intervention.
• World Bank — Early Warning Systems Can Help Prevent Dropouts, but Administrators Need to Consider Critical Aspects to Increase Effectiveness. Implementation lessons from Peru.
• UNICEF Evaluation — A Case for Multiple and Flexible Pathways to Learning. Evidence on second-chance and flexible routes for out-of-school learners.

Final Thought: The Earlier the System Sees the Problem, the More Futures Remain Available

Formal dropout statistics arrive after something has already been lost.

The deeper job of an education system is to notice the weakening connection before the learner becomes a statistic, and to respond with enough precision that support feels like help rather than surveillance.

Sometimes the solution is a bus pass. Sometimes it is tutoring, childcare, a safer classroom, a different pathway, a health referral, a flexible schedule or one adult who notices that the student who was always present has begun disappearing.

And when prevention fails, a strong system does not treat departure as final.

It keeps a route back.