Learning Intelligence Platform
A weekly risk model replaces the termly review meeting, giving advisors across an 80,000-student group a ranked intervention list before the gap becomes unrecoverable.
The challenge
Across 14 campuses, the group already collected plenty of student data. It just couldn't act on it in time:
- Attendance, assessment, and engagement data sat in five separate systems with no shared student key
- Risk was reviewed once per term, by which point half the term was already lost
- Advisors relied on gut feel and whichever spreadsheet was most recently updated
- No group-wide definition of 'at risk' — each of 14 campuses scored differently
- High-risk students were typically identified after a second failed assessment, not before
Manually reconciling the five source systems into a single review took roughly three weeks each term.
How it works
A weekly model instead of a termly meeting
The fix wasn't a better spreadsheet — it was a standing pipeline that scores every student every week:
- 01
Established a canonical student ID reconciling records across the SIS, LMS, attendance, and assessment systems
- 02
Trained a gradient-boosted risk model on three years of historical outcomes: attendance decay, assessment trend, engagement drop-off
- 03
Standardised a single group-wide risk definition and threshold across all 14 campuses
- 04
Automated a Sunday-night scoring run that refreshes every active student's risk band
- 05
Delivered a ranked Monday-morning intervention list to each campus's advising team
- 06
Closed the loop by logging which interventions were taken and feeding outcomes back into the model
What we built
Key capabilities
Weekly, not termly
Every student's risk band refreshes overnight, every week — advisors act inside the term instead of after it.
One risk language, 14 campuses
A single group-wide threshold replaces 14 informal, incompatible ways of calling a student 'at risk'.
Ranked, not raw
Advisors get a prioritised list of who to reach first, not a data dump to interpret.
Learns from what worked
Logged interventions and their outcomes feed back into the model, so the ranking improves term over term.
Before vs after
What changed for advisors
- Review cadence
- Termly → Weekly
- Data assembly
- ~3 wks manual → Automated overnight
- At-risk definition
- 14 campus variants → 1 group standard
- Flagged-to-on-track rate
- 24% → 34%
- Intervention timing
- After 2nd failed assessment → Before the first
Business impact
What it changed
42% relative improvement in recovery
The share of flagged students who moved to on-track by term end rose from 24% to 34% — a 42% relative gain, sustained across two consecutive terms.
Advisors act inside the term
A Monday-morning list, refreshed weekly, replaced a termly meeting that arrived after the window to help had mostly closed.
One definition, 14 campuses
The group can now compare risk and outcomes across campuses for the first time, instead of reconciling 14 local conventions.
Technology stack
“The data existed all along. What changed was the cadence — weekly instead of termly turned a lagging report into a working intervention system.”
Keep reading