How to improve online course completion rates

What do online course completion rates actually look like?

Most people who enroll in an online course don’t finish it, and the best public data shows the rate hasn’t improved with time. The best public data comes from MOOCs, because researchers could see every learner. A 2019 analysis in Science by Justin Reich and José Ruipérez-Valiente covered every Harvard and MIT course on edX from 2012 to 2018: 5.63 million learners in 12.67 million course registrations. The results:

Two lessons for a paid course creator. First, paying helps but doesn’t solve it: more than half of paying edX learners still didn’t finish. Second, published figures are only a rough frame. Katy Jordan’s study found completion fell as courses got longer, so a 12-week course and a 3-week one aren’t comparable. You will see much higher completion figures quoted for cohort-based or premium courses, usually self-reported by the vendor; treat those as marketing until you can see how “completion” and “enrolled” were defined. The benchmark that matters is your own last cohort.

Why do learners drop out of online courses?

“No time” is the easy explanation. Look at where learners actually stop, though, and six causes come up again and again:

• No plan. Learners intend to finish, then never decide when. Yeomans and Reich call this the intention–action gap: among MOOC learners who intend to complete, the majority don’t.
• No accountability. Nobody notices if they miss a week, so missing a week costs nothing, and then it’s two.
• Passive, long video. Watching feels like progress but asks nothing of the learner, and long videos lose people partway through.
• Stuck on one concept. In a Spanish course, this is often the module on ser vs estar. The learner can’t get past it, the forum answer comes two days later, and by then they have stopped opening the app.
• Modules too long to finish in one sitting. A 45-minute module started at 10 pm is a module half-finished, and half-finished modules are where streaks end.
• Heavy work before the habit forms. Jordan found that courses graded only automatically had higher completion rates. One reading: a big peer-reviewed project in week one asks for commitment before the learner has any.

Only the first two are about motivation. The other four are design problems you can fix this month.

Which metrics should you track to improve course completion?

Final completion is a lagging number: by the time it moves, the cohort is over. Track the leading ones instead. Jordan’s analysis of MOOC activity found the first and second weeks are critical, after which the share of active students levels out. So most of what you can change happens early.

• Activation: enrollees who start lesson one within 7 days ÷ all enrollees.
• Unit-one completion: learners who finish the first unit ÷ all enrollees.
• Week-two retention: learners active in days 8–14 ÷ learners who activated.
• Stall points: for each lesson, the share of learners who reach it but never finish it. The biggest numbers are your to-do list.
• Time to an answer: how long a question posted by a learner waits for a reply.

Here is what the funnel looks like for an illustrative 8-module Spanish course (made-up numbers, not a benchmark):

Read it from the top. 360 people never started, which is an onboarding problem. Week-two retention is 330 ÷ 640, about 52%. Module 4 loses 90 of the 240 who reach it, a stall point of 37.5%, far worse than the modules after it. That is where a better explanation, practice questions or live help would pay off first.

How do you diagnose a stall point?

A drop in the funnel tells you where learners leave, not why. The pattern around it usually does. Match what you see to the likely cause before you change anything:

In the Spanish example, module 4 fits the third row. Ser vs estar is one idea with a lot of exceptions, and one video rarely lands it for everyone. A short contrast table, ten quick “ser or estar?” questions with instant feedback, and somewhere to ask when a sentence doesn’t seem to fit the rule will do more than re-recording the video in better lighting.

Which fixes does the research support?

Cut videos into short segments. Philip Guo, Juho Kim and Rob Rubin analyzed 6.9 million video sessions across four edX courses in 2014. Median engagement was at most 6 minutes regardless of how long the video was, and students often made it less than halfway through videos longer than 9 minutes. The share of sessions followed by an attempt at the next problem fell from 56% for the shortest videos to 31% for the longest. Their recommendation: plan lessons in chunks under 6 minutes.

Put a question after every segment. In a 2013 PNAS study by Szpunar, Khan and Schacter, Harvard students watched a 21-minute statistics lecture split into four segments. Those quizzed after each segment reported mind-wandering on 19% of probes, against 39% for students shown the same questions with the answers filled in and 41% for students who did arithmetic instead. On the final test they scored 90%, against 76% and 68%. It was a lab study, but the mechanism is simple: a learner who knows a question is coming pays attention. Our guide to checking for understanding has twelve question formats to use.

Ask for a plan on day one. In three HarvardX courses, Michael Yeomans and Justin Reich added open-ended planning prompts at the start of the course. Completion rose 29% and certificate purchases 40%, most for students also enrolled in traditional schools. One odd result is worth knowing: learners whose plans focused on specific times were less likely to finish. Ask about obstacles and strategies, not just a calendar slot.

Do deadlines, cohorts and reminders help?

These are the fixes creators reach for first, and the evidence for them in self-paced courses is thinner than the research above. Treat them as good practice to test on your own cohorts, not as proven effects.

• Cohort start dates. A fixed start date turns “someday” into a date, and a group moving together gives learners someone to notice their absence. The cost is flexibility: people who enroll mid-cycle wait.
• Soft deadlines per module. A suggested finish date per module with a weekly “you’re on track” or “you’re one module behind” message. Keep it forgiving: a missed hard deadline hands the learner a reason to stop.
• Reminders that resume, not nag. “You stopped 3 minutes into lesson 5: ser for descriptions. Pick up where you left off” beats “We miss you.” Send it from what the learner did last, not on a fixed schedule.
• A small number that belongs to the learner. Lessons done this week, or a streak of study days. Keep it tied to learning, not to logins.

Where does live help fit?

The stall point in the funnel above is rarely a motivation problem. It is a learner who watched the ser vs estar video twice, still can’t tell which to use, and has nobody to ask at the moment they are stuck. There are four ways to fix that, and each has a trade-off:

• Live office hours with the creator. The best help there is, and the one that does not scale: your evenings go first.
• A community forum or chat group. Cheap, and good for motivation, but answers arrive hours or days later, if at all.
• Teaching assistants. Fast and personal, at a cost per hour that grows with enrollment.
• An AI tutor on the hard chapters. Available at 11 pm and priced per use. It helps only if it teaches from your course, asks the learner questions and checks the answers; a chatbot that only answers questions can end up doing the learner’s thinking for them.

Size the job before you pick. In the illustrative funnel, 240 learners reach module 4. If each needs about 15 minutes of one-to-one help to get past it, that is 3,600 minutes, or 60 hours of a person’s time for one module in one cohort. The same minutes from an AI tutor priced per learner-minute cost $108 at $0.03 a minute (Tutezy’s rate). Neither number is large; the difference is that the person has to be available at the moment each learner gets stuck.

Whatever you choose, put it where your stall points are, not everywhere. We compare the recorded-course and live-tutor models in Tutezy vs recorded courses, and the three ways to add an AI tutor to an existing course in how to add an AI tutor to your online course.

A 30-day plan to improve online course completion

Thirty days is too short to see final completion change, but long enough to move activation, unit-one completion and week-two retention for a new cohort. The plan:

1. 1

   Days 1–3: instrument the funnel

   Log enrolled, started lesson, finished lesson and last active for every learner. Rebuild the funnel table above for your last cohort.
2. 2

   Days 4–7: find the stall points

   Rank lessons by the share of learners who reach them but don’t finish. Read every question and comment posted on the top three.
3. 3

   Days 8–11: fix onboarding

   Make lesson 1 short with an early win. Add an open-ended planning prompt to the welcome: how will you fit this in, and what might get in the way?
4. 4

   Days 12–17: split and quiz

   Cut long videos at the stall points into segments under about 6 minutes, each followed by two or three questions.
5. 5

   Days 18–22: put help at the stall points

   Office hours, a same-day answer target, or an AI tutor on those chapters. Measure time to an answer.
6. 6

   Days 23–26: add gentle accountability

   Soft module deadlines, a weekly progress note, and resume reminders that name the exact lesson.
7. 7

   Days 27–30: compare cohorts

   Set the new cohort’s activation, unit-one completion and week-two retention against the last cohort’s. Keep what moved; drop what didn’t.

If you sell to organizations rather than individuals, the same plan applies; the reminders come from a manager instead of an app. For turning existing material into shorter, question-led lessons, see turning PDFs, slides and videos into interactive lessons.