Higher Education /Decisioning live field guide · 9 min

At-Risk Early Alert

Pulls LMS, attendance and assessment signals into one view, flags students showing early-warning patterns by week 2, and tells the advisor which signal drove each flag — a person decides who to contact.

Open the live demo

theater/demos/higher-ed_at-risk-early-alert.html · sandbox · read-only

Open

FIG. 1

The live demo, running on fabricated data. Open it to step through the full flow — every output is shown for a person to approve before anything happens.

How it would work

Pulls each student's LMS, attendance, assessment and grade signals into one view, ranks who is showing early-warning patterns, shows the signals behind each rank, and drafts a supportive outreach for an advisor to approve before anyone is contacted.

Input 01

The engagement signals

Per-student signals from the LMS and student system — login frequency and recency, attendance where it is recorded, assessment submissions and due dates, and marks to date — read against the cohort baseline.

Agent 02

Compares, ranks, explains

Compares each student against the cohort, ranks who is furthest from the healthy range, and lists the specific signals that drove the rank — a missed assessment, logins down sharply, attendance falling away.

Output 03

A ranked list, working shown

A ranked watchlist where each flag names its signals and a draft outreach sits ready — for a student success advisor to read, edit and approve before any student is contacted.

Where it works well

It watches every enrolled student, every week, and never quietly drops the quiet ones.

Best for large first-year or fully online cohorts, where there is no tutor in a room to notice the empty seat.
Done by hand it is a fortnightly trawl through LMS reports and the gradebook — the first thing dropped in a busy teaching week.
At cohort scale the recaptured hours go back into the conversation with the student, not into assembling the spreadsheet that found them.

The students who fail are rarely the ones who ask for help — they go quiet. The slow, invisible work is noticing, across a 300-student cohort, who has stopped logging in, missed the first assessment, and slipped below the pass line, before week six when it is already too late to recover the unit.

Where it works badly

It ranks risk from the past behaving like the future — and it inherits whatever bias is in the history.

It ranks risk; it never decides to penalise, fail or report a student. A low rank is not a clearance and a high rank is not proof.
Trained on past cohorts, it can over-flag the groups historically flagged — equity-cohort, part-time, mature-age students — so the flagged list needs reading as a starting point, not a ruling.
A clean professional score hides its own uncertainty; the rank looks more certain than the thin signals behind it.

The honest test

If you would treat a high rank as a reason to contact a student with care, it helps; if you would treat it as evidence to act against them, this is the wrong tool and the wrong use.

A risk score is a base-rate bet, not a verdict. Most flagged students were never going to fail, and some who fail were never flagged. Point it at a cohort where disengagement looks different — shift workers, carers, a placement block with no LMS activity by design — and it confidently ranks the wrong people first.

What it doesn't do — and shouldn't

It ranks and drafts. A person decides who to contact, and how. That boundary is deliberate.

WHAT IT DOES

Surfaces the specific signals behind each rank — the missed assessment, the login drop, the attendance gap

Shows each student against the cohort baseline so the advisor can judge whether the gap is real

Drafts a supportive, non-judgemental outreach the advisor can edit or discard

WHAT IT WON’T

Contact the student, or send anything on its own

Decide an academic-integrity, progression or exclusion outcome

Treat a risk rank as a finding about the person rather than a prompt to check in

Identifying and supporting students at risk is something TEQSA expects providers to do under the Higher Education Standards Framework — but a flag that quietly hardened into an exclusion or a misconduct call, with no person accountable, is exactly the harm a risk model can do. Wellbeing and support obligations sit with named staff, not a score, so a person stays on every decision about a student.

What your data has to look like

Per-student engagement signals, joined across systems, current to this week.

Typical readiness

across orgs we see, before the first job

LMS activity per student, not just per unit

Needs shaping

Attendance recorded somewhere countable

Needs shaping

Assessment due dates and submission status

Usual weak point

A stable student identifier across systems

Usual weak point

A cohort baseline to compare against

Usually ready

The real first job

The signals exist, but scattered — the LMS knows logins, the gradebook knows marks, attendance lives in a sign-in sheet or nowhere. Joining them to one student, kept current to the week, is usually the real first job, larger and more valuable than the ranking on top. Once the signals are clean and joined, every weekly read after that is faster and fairer.

Right fit if…

You run large first-year or online cohorts where at-risk students stay invisible

You have a student success or retention function with the staff to act on flags

Your LMS exposes per-student activity, not just unit-level reports

You want data-informed referrals to replace relying on self-referral alone

Walk away if…

You want a score that decides progression, exclusion or misconduct for you

Your cohorts are small enough that staff already know who is struggling

Engagement is invisible by design — placement blocks, work-integrated learning

Attendance and LMS activity are not captured as data anyone can join

Open questions

The worried-buyer questions, answered straight

It can rank the wrong students first — a risk model is a base-rate bet on past patterns, not a verdict on a person. That is exactly why it only ranks and drafts, and never decides. Each flag names the signals behind it — the missed assessment, the login drop — so a student success advisor judges whether the gap is real before reaching out, and the outreach is a supportive check-in, never a penalty. The tool surfaces who might need a conversation; a person decides who to contact and how.

It only sees what is captured and joined. In Australian higher ed, attendance is often not recorded as data at all, and LMS reporting is built around units, not a per-student timeline — so the first work is usually joining logins, marks and attendance to one student and keeping it current to the week. If a signal is missing for a cohort, the rank leans on the signals that remain and can read a placement block or a quiet learner as disengagement. Getting the signals clean and joined is the piece that pays off across every weekly read after.

No. It removes the fortnightly trawl through LMS reports and the gradebook, so the advisor spends their time on the judgement and the conversation — whether the flag is real, whether the student needs an extension, a referral to counselling, or just a check-in. The decision about who to contact, and the relationship that follows, stays human. The capacity it frees goes back into students, not spreadsheets.

Current to the teaching week. Early-warning signals only help if they surface while there is still time to intervene — a missed first assessment matters in week two, not at the exam-board meeting. If the data is a month stale, the tool ranks risk from a picture that has already moved on, and a student who has since re-engaged still looks high-risk. The honest test: can you say, today, which week’s LMS and attendance data this ranking is built on?

Student engagement and wellbeing data is sensitive — handled under the Australian Privacy Principles for private providers and ANU, and under state or territory privacy law for most public universities. Any deployment runs against your own systems and access controls, not a shared pool; we scope who can see a flag and how long it is kept as part of the build. The demo here runs entirely on fabricated data — Priya Nair is not a real student.

It compares each student’s signals — LMS logins, attendance, assessment submission, marks to date — against the cohort baseline and ranks who is furthest from the healthy range. Every flag shows its working: the specific signals that drove the rank and how they compare to the cohort, so an advisor sees a missed assessment and a login drop, not an unexplained number. A rank with no reasons would be a black box; this one is built to be argued with.

What it takes to build

3–4 weeks · 4 phases

Reused from template~70%

Bespoke to this skin~30%

stack · Claude · LMS/SIS connectors · review dashboard

What it would cost

Fixed scope, fixed price, fixed dates.

Bite-sized first piece

One contained change, low risk

Pilot build

Most builds land here

Embedded support

Scale on proof

Considering this for your cohort?

The honest place to start is a bite-sized first piece — one contained change, low risk. Tell us where students go quiet; we'll play it back, scope it, and show you what's possible.

Book a call How we work