Reducing dropout by steering learning paths with data

I structure how learning data is used: learning KPIs, learner-journey dashboards and early detection of learners at risk of dropping out.

Problem
Spot disengaging learners earlier, and give teaching teams a reliable, shared view of progress, without adding to their workload.
Role
Data Project Manager (work-study)
Period
Mar. 2026 – present
Sector
EdTech · Online learning

Fig. 1Castor Education

Learners at risk of dropping out, by alert threshold

At this threshold, the team must contact 46 learners to catch 5 dropouts out of 10.

4 of these alerts concern learners who would have stayed. 29 dropouts slip under the radar.

60
← More alertsFewer alerts →
Who gets flagged at this threshold
  • Dropouts caught
  • Unneeded alerts
  • Dropouts missed
  • Stayed, not flagged
02040020406080100Risk score (0 to 100)
The trade-off, threshold by threshold
  • Dropouts caught
  • Alerts that were right
0%50%100%Caught 59%Right 91%304050607080Alert threshold

Click or drag on the curve to pick a threshold.

View the data
The trade-off, threshold by threshold
ThresholdFlaggedDropouts caughtAlerts that were right
≥ 30225100 %32 %
≥ 3518599 %38 %
≥ 4014999 %47 %
≥ 4511697 %59 %
≥ 508783 %68 %
≥ 556273 %84 %
≥ 604659 %91 %
≥ 653141 %94 %
≥ 702028 %100 %
≥ 751521 %100 %
≥ 801115 %100 %
≥ 8557 %100 %
Who gets flagged at this threshold
Risk scoreDropped outStayed
0–401
5–904
10–1409
15–19019
20–24022
25–29040
30–34139
35–39036
40–44132
45–491019
50–54718
55–59106
60–64132
65–6992
70–7450
75–7940
80–8460
85–8920
90–9420
95–9910
Synthetic data: the method is the project’s, the figures are not Castor Education’s.

01Context

Castor Education trains students, employees and career-changers for digital, data and AI roles. Learning data exists (logins, progress, assessments), but it is scattered across the LMS and rarely used to steer learner support.

02Challenge

Spot disengaging learners earlier, and give teaching teams a reliable, shared view of progress, without adding to their workload.

03Approach

  1. Define the metrics

    Agreeing with teaching teams on the KPIs that matter: attendance, progress, assessment results, early dropout signals.

  2. Make LMS data reliable

    Extracting, cleaning and automating data flows from the learning platform into a single source of truth.

  3. Build the dashboards

    Learner-journey dashboards designed for the team meeting: what is on track, what is stuck, whom to call.

  4. Find the friction points

    Journey analysis to find the modules where learners drop off, and first work on risk scoring.

04Deliverables

  • Learning KPI framework
  • Learner-journey dashboards
  • Automated LMS data flows
  • Friction-point analysis and recommendations

Tools

  • Python
  • Pandas
  • SQL
  • Power BI
  • LMS
  • Excel

05What I took away

A metric is only worth something if it triggers an action. The best dashboards tell you who to call on Monday morning.