Case Study · Web3Bridge Retention
The Cohort 13 Retention Experiment
Two retention interventions across three cohorts, an honest diagnosis of why the first only half-worked, and the structural fix that beat it.
- Role
- Ran and evaluated the interventions
- Scope
- Cohorts 11–13 retention experiments
- Method
- Training-officer pods vs split online/onsite programs
- Outcome
- Split-program structure adopted for Cohort 13
Problem
Cohort 13 saw a larger intake than previous cohorts, which made an existing weakness impossible to keep working around: online students were completing the program at a meaningfully lower rate than onsite students, and it was dragging down overall cohort completion.
My Role
I ran and evaluated the first intervention across Cohorts 11 and 12. For Cohort 13, I deliberated directly with Web3Bridge's founder, Israel, and we jointly finalized a second, more structural approach as an experiment.
What Happened
For Cohorts 11 and 12, we trialed a training-officer pod model, originally Israel's idea: students grouped into pods of about 10, each with a dedicated training officer outside their subject mentor, giving them one-on-one support and accountability beyond regular class time. I selected the training officers and managed the pods directly. It helped, modestly, but the gap between online and onsite completion persisted.
Looking at why, the real problem came into focus: online students weren't underperforming from a lack of individual attention, they were working against a schedule and pace built around the onsite cohort, with no realistic way to keep up regardless of how much support they got. The training-officer model was treating a symptom, not the cause.
For Cohort 13, Israel and I deliberated and finalized a more structural experiment: fully separating the online and onsite programs, each with its own dedicated mentors and its own schedule, instead of one shared track with extra support layered on top.
The training-officer model was treating a symptom, not the cause.
Outcome
The full separation outperformed the training-officer model by a clear enough margin that we dropped training officers entirely for Cohort 13, in favor of the split-program structure. The split itself is independently verifiable: Cohort 13's public GitHub repository still shows online-program and onsite-program as separate top-level directories, confirming this wasn't just a description after the fact.
What I'd Do Differently
Nothing about the decision itself, this is the version of the process worth pointing to in an interview: a real hypothesis, a real trial, an honest diagnosis of why the first attempt only partially worked, and a structural fix once we understood the actual cause instead of the symptom. If anything, I'd have framed the training-officer trial as an explicit hypothesis with a defined success metric from the start, rather than assessing it informally after the fact.