IronFuentesEspañol
Cases

Student retention alerts

You are the person accountable for dropout rates at a university. Today you find out a student has left once they have already left. With this you open it on Monday, see nine names in red and call those nine.

The problem

An institution with thousands of students has no way of knowing who to call first. The data exists — grades, attendance, payments — but it lives in separate systems and nobody cross-references it. By the time someone notices a student stopped showing up, a whole semester has gone by.

The flow

  1. Grades, attendance and payments

    Airtable

  2. Runs every Monday

    n8n · Schedule

    A person steps in
  3. Normalises and matches by student

    n8n · Code

  4. Scores risk from five factors

    n8n · Function

  5. Applies intervention rules

    n8n · Switch

  6. Stores the history

    Supabase

  7. Alerts whoever is responsible

    Email · WhatsApp

  8. Follow-up dashboard

    Airtable Interface

8 steps. The ones marked in gold do not run on their own: they require someone to decide, which is what separates responsible automation from the kind that causes trouble quickly.

Tools

  • Airtable
  • n8n
  • Google Sheets
  • Email and WhatsApp
  • Supabase

What happens when something breaks

A flow that only works with perfect data is no use. These are the cases that were anticipated and what the system does in each one.

  • A student has no attendance record for that term

    The factor is left out of the average instead of counting as zero. Missing data is not bad data, and treating it as zero would invent risk where there is none.

  • The source returns an empty file or the connection fails

    The flow writes nothing, keeps the previous week’s calculation and notifies the technical owner. Alerts are never sent on incomplete data.

  • A student appears twice under different ID numbers

    It is caught by name and programme, flagged for human review and held back from the send until resolved. Records are never merged automatically.

  • The same student would trigger an alert four weeks running

    There is a two-week quiet period per student and per rule. Without it, whoever receives the alerts stops reading them within a month.

Measurable result

2 indicators are still being measured: at-risk students identified before dropping out, hours of manual review avoided per semester. They will be published with their source and method. I would rather leave the gap visible than put up a number I could not defend in an interview.

Demo

A cohort of 180 students, loaded and assessed. The system works out each one’s dropout risk from five declared factors, sorts it into four bands and applies 4 intervention rules.

  • Students assessed

    180

    the whole cohort

  • Require action

    34

    19 % of the cohort

  • Critical risk

    9

    immediate intervention

Cohort map

Each dot is a student. The lower and further left, the higher the risk. Hover with the mouse, or enter the map with the tab key and move through the dots with the arrows.

0%25%50%75%100%0.01.02.03.04.05.0Grade average
Vertical axis: attendance.LowMediumHighCritical

Point at a dot to see the detail.

Distribution by risk band

  • Low136 · 76 %
  • Medium10 · 6 %
  • High25 · 14 %
  • Critical9 · 5 %

Low up to 29, medium up to 54, high up to 74, critical from 75.

Students requiring action, by programme

  • Business Administration12 of 51
  • Accounting10 of 49
  • Systems Engineering6 of 43
  • Psychology6 of 37

The programme name always accompanies the colour: the data never depends on telling shades apart.

Interventions the system proposes

  • Immediate action by the retention committee9 students

    Retention committee · triggers if riesgo ≥ 75.

  • Direct contact and a plan to catch up on classes24 students

    Programme management · triggers if asistencia ≤ 0.6.

  • Assign academic tutoring in the failed subjects22 students

    Academic coordination · triggers if reprobadas ≥ 3.

  • Refer to financial aid and review the payment plan25 students

    Student welfare · triggers if mora = yes and riesgo ≥ 50.

Priority list

The twenty highest-risk cases. Pick one to see how it was worked out.

Students ordered from highest to lowest risk, with their band and programme
StudentProgrammeRiskBand
Accounting91Critical
Business Administration91Critical
Business Administration90Critical
Accounting86Critical
Business Administration85Critical
Business Administration83Critical
Business Administration77Critical
Accounting76Critical
Accounting75Critical
Psychology73High
Business Administration73High
Accounting72High
Business Administration71High
Accounting70High
Psychology70High
Psychology69High
Accounting69High
Business Administration69High
Psychology68High
Accounting68High

How it was worked out

Santiago Naranjo E-1008

91out of 100 · critical risk

  • Academic performance (1.84)+30 of 30
  • Attendance (52 %)+23.6 of 25
  • Failed subjects (3)+20 of 20
  • Overdue balance (yes)+15 of 15
  • Credit progress (51 %)+2.7 of 10

Sum of contributions = 91. No remainder, no hidden adjustment.

Interventions triggered
  • Immediate action by the retention committee

    riesgo ≥ 75 · has 91

  • Direct contact and a plan to catch up on classes

    asistencia ≤ 0.6 · has 0.52

  • Assign academic tutoring in the failed subjects

    reprobadas ≥ 3 · has 3

  • Refer to financial aid and review the payment plan

    mora = true · has true · riesgo ≥ 50 · has 91