Online course completion rates are famously poor, and the usual response is to add features: badges, streaks, progress bars, certificates. Some of that helps a little. None of it addresses why people stop.
We run our own academy on software we built, so this is observed rather than theorised.
Deadlines beat self-pace
A course someone can take any time is a course they can take later, and later is where good intentions go.
Cohorts with real dates — a start, a weekly session, a submission deadline — consistently outperform the same material offered on demand. The constraint is the feature.
Someone has to notice when a learner goes quiet
The strongest single intervention is a person reaching out when attendance or submissions stop. Not an automated nudge, which is easy to ignore, but a message from someone whose name the learner recognises.
That is an operational cost rather than an engineering one, but the software decides whether it is possible: if nobody can see who is drifting, nobody will reach out.
Feedback speed matters more than feedback depth
An assignment returned in two days with three sentences of comment does more than one returned in two weeks with a page.
This shapes what you build. Auto-marking for anything that can be auto-marked, a queue that makes human marking fast, and a clear view of what is outstanding — those matter more than richer rubrics.
The first session decides a lot
A learner who hits a setup problem in week one and cannot resolve it frequently never returns, and will not tell you why.
- Make the first task small and achievable, not representative of the course's difficulty.
- Front-load environment setup into something you support directly.
- Have a channel where asking a basic question is socially safe.
Build for three users, not one
Learners, instructors and administrators want incompatible things from the same system. Designing entirely around the learner produces a platform instructors avoid, which degrades the learner experience indirectly and more severely.
Instructor tooling — marking queues, attendance, a view of who is struggling — is not back-office overhead. It is what makes the human intervention above possible at all.
Certificates should be verifiable
A certificate nobody can check is decoration. A public verification page turns each one into something a graduate links from a professional profile, which is both useful to them and the cheapest marketing a course has.
More on how we build in this sector on our EdTech page, and on what a platform of this shape costs in the SaaS cost breakdown.
Why are online course completion rates so low?
Mostly because self-paced courses have no deadline, and anything without a deadline can be done later. The strongest lever is structural rather than technical: a cohort with real dates, a weekly session and a submission deadline consistently outperforms the same material offered on demand. Features like badges and streaks move the number far less than a date does.
Should we build our own learning platform or use an LMS?
Use an existing LMS unless your teaching model does not fit one. Established platforms handle course delivery, quizzing and reporting competently for a subscription. Building is justified when your model is unusual — mixed live and asynchronous cohorts, bespoke assessment, credentials that must integrate elsewhere — or when the platform itself is the product you sell.
Enjoyed this? Let's work together.
Start a project