Course Completion Rates: Why Learners Drop Off

By Brian Tr · Updated Oct 7, 2026

A course completion rate is the share of learners who finish a course. That sounds like one number. It's at least four, and they can differ by an order of magnitude for the same course, depending on who is counted in the denominator and what counts as "finished."

That's why the figures people quote don't line up. A free online course that "only" completes 6% of enrollees may be doing fine. A paid cohort that completes 90% may be hiding a weak program behind a selected audience. A corporate course with 98% completion might just show that managers chased people. This article covers what the term means, what the best public research found, why learners drop off, which fixes have evidence behind them, and when completion is the wrong thing to track.

Key Facts

• Across 39 MOOCs with public data, the median completion rate was 6.5% of enrollees, with a range of 0.9% to 36.1% (Jordan, IRRODL 2014). Counted against active students only, the median was 9.8%.
• In the first year of HarvardX and MITx courses, 43,196 of 841,687 registrations earned a certificate, while 292,852 registrants never engaged with the content (HarvardX and MITx first-year report).
• Among 79,535 pre-course survey respondents in nine HarvardX courses, 19.5% of those who intended to complete earned a certificate, against 5.4% of those who didn't intend to (Reich, 2014).
• In 565 Harvard and MIT course runs on edX (5.63 million learners), 52% of registrants never entered the courseware, and completion rates had not improved over six years (Reich and Ruipérez-Valiente, Science 2019).
• A commitment device tested in a Stanford MOOC (657 students) raised completion likelihood by 40%, while reminders alone showed no measurable effect (Patterson, Cornell CHERI).

What "completion rate" actually measures

The same label covers four different calculations. Before comparing any two numbers, check which one you're looking at.

Jordan's analysis shows how much the denominator matters. About 54% of enrollees became active students. Using total enrollment, the median completion rate was 6.5%. Using active students, the median rose to 9.8% and the range widened to 1.4% to 50.1% (Jordan, 2014). Same courses, different denominator, different headline.

Paid and cohort programs use yet another population. Learners applied, paid and were expected on a start date, so they've already passed a filter that a free course never applies. Comparing their completion to a MOOC's compares two different groups of people. The sibling article on cohort-based vs self-paced learning covers how to separate the effect of the format from the effect of selection.

Corporate assigned training is the odd one out. When a manager or a compliance deadline forces the finish, the completion rate mostly measures enforcement. We found no public benchmark for it that defines its population well, so this article doesn't quote one.

What the research shows

The most careful public evidence comes from MOOCs. They aren't corporate programs or paid bootcamps, so read the numbers as what happens with free, open-enrollment learners, not as a forecast for your product.

Early benchmarks: single digits

Katy Jordan collected public enrollment and completion figures from MOOCs listed as of July 2013. She defined completion as the percentage of enrollees who met the course's criteria for a certificate, because that was the figure most often published. Completion data was available for 39 courses. The median was 6.5%, the range 0.9% to 36.1%, and a rate around 5% was typical. Completion was negatively correlated with course length and didn't vary significantly with date, university ranking or total enrollment (Jordan, 2014). Jordan also cautioned that completing assessments isn't the only way to benefit from a course.

A year of HarvardX and MITx

The first-year report from HarvardX and MITx covered 17 courses from fall 2012 to summer 2013. It counted 841,687 registrations from 597,692 unique users. Of those registrations, 43,196 earned certificates, which is about 5.1% by our arithmetic. Another 35,937 explored half or more of the content without certification, 469,702 viewed less than half, and 292,852 never engaged with the content (HarvardX and MITx report).

Read the four groups together and a pattern shows up. A certificate is a narrow slice of the people who showed up, and a large group never started. The "drop-off" often happens before the course begins in any meaningful sense.

Intent changes the picture

Justin Reich analyzed pre-course surveys from nine HarvardX courses in 2013 and 2014. Of 79,535 respondents, 56% said they intended to complete. Of those, 19.5% earned a certificate. Among the 44% who didn't intend to complete, 5.4% did. The overall rate for respondents was 13.3% (Reich, 2014).

There are two ways to read that. Intent matters: people who meant to finish were nearly four times as likely to as those who didn't. And intent isn't enough: about four in five people who meant to finish still didn't. Survey respondents also differ from non-respondents, so these rates describe respondents, not every registrant.

Six years later, nothing moved

Reich and Ruipérez-Valiente looked at all HarvardX and MITx courses on edX through May 2018: 565 course runs, 261 courses, 12.67 million registrations from 5.63 million learners. They found that 52% of registrants never entered the courseware. They also tracked completion for three overlapping groups: all participants, those who said they intended to complete, and those who paid for a verified track. None showed meaningful improvement over six years. And the largest cohort, 1.1 million new learners in 2015-2016, saw only 12% take another HarvardX or MITx course the following year, with second-year retention falling from 38% in the first cohort to 7% in the 2016-2017 cohort (Reich and Ruipérez-Valiente, 2019).

The sample is two elite universities on one platform, so it won't represent every course. But it's one of the largest open datasets on learner behavior, and the finding is blunt: better video, better design and six years of research didn't shift completion much. Even learners who paid finished at rates well short of 100%.

Not one kind of drop-off

Kizilcec, Piech and Schneider classified learners in three computer science MOOCs by how they interacted with videos and assessments. The method consistently found four trajectories: completing, auditing, disengaging and sampling. The most notable group was learners who stayed engaged through the course without taking assessments. The authors argued that the focus on completion reflects a monolithic view of disengagement that doesn't help designers target different learners (Kizilcec et al., LAK 2013).

Why learners drop off

Research doesn't hand us one tidy list of reasons, so treat the following as four themes that the evidence supports in part, not a ranking.

Intent was thin to begin with. Many registrants never meant to finish. The Reich data shows 44% of survey respondents didn't intend to, and the HarvardX counts show hundreds of thousands who barely engaged. For these learners, leaving isn't a failure of the course.

Time and competing demands. Adult learners fit courses around work and family. A course that needs five hours a week loses people the week a project lands. Jordan's finding that longer courses completed at lower rates is consistent with this, though it doesn't prove why.

Weak self-regulation. In a study of 4,831 learners across six MOOCs, goal setting and strategic planning predicted whether learners reached their own course goals, while help seeking appeared to be counterproductive (Kizilcec et al., Computers and Education 2017). Learners who plan do better. Most learners don't plan.

No one expects them. In a self-paced course nobody notices an absence. Deadlines, a live cohort and a person waiting for your work all change that, which is the logic behind the cohort model. See learning communities for how peer groups create that expectation.

Course design matters too, but the evidence that design alone moves completion is weaker than most vendors imply. Reich and Ruipérez-Valiente note that six years of investment in course development and learning research hadn't produced meaningful gains in completion.

Which levers have evidence

Here's what has been tested, with the population each result describes.

Structure and deadlines

A Finnish study compared two versions of a programming MOOC with over 8,000 students. The fixed-schedule version had 45% overall retention, the self-paced version 13% (Noponen et al., IEEE FIE 2020). That's a large gap, but the summary doesn't state whether learners were randomly assigned, so who chose which version could explain part of it.

A cleaner comparison held the course constant. An MITx supply chain course moved from instructor-paced to self-paced, and a learning-analytics thesis reported that completion fell by 6%, pass rate by 10% and an engagement score by 7%, while satisfaction stayed the same. The effect wasn't uniform across learners (Bilal, MIT 2019). So pacing helps, but by less than the raw cohort numbers suggest.

Commitment devices

Richard Patterson ran a randomized experiment with 657 students in a Stanford MOOC. A tool that let students pre-commit to time limits on distracting websites led to 24% more time on the course, grades 0.29 standard deviations higher and a 40% higher likelihood of completing. Reminder and focusing tools weren't statistically distinguishable from the control (Patterson, Cornell CHERI working paper, 2015). It was a single course with a self-selected sample, but the contrast is useful: a nudge that asks for a commitment beat a nudge that only reminds.

Nudges at scale: smaller than hoped

The strongest warning comes from the largest test. Kizilcec and colleagues ran five behavioral interventions across 247 courses from Harvard, MIT and Stanford, with 269,169 students over two and a half years. The interventions were plan-making (long and short term), value-relevance, mental contrasting with implementation intentions, and social accountability. Effects were roughly an order of magnitude smaller than in earlier, smaller studies. The value-relevance exercise raised completion by about 2.8 percentage points, but only for learners in less-developed countries in courses with a large achievement gap. Plan-making lifted early engagement for a week or two with no lasting effect on completion (Kizilcec et al., PNAS 2020).

The practical lesson: a nudge that worked once probably won't work everywhere, so test on your own learners.

Instructor and social presence

Tomkin and Charlevoix randomly assigned participants in a University of Illinois physical science MOOC to either no instructor interaction or an arm where the professor and teaching assistants answered forum comments and posted weekly feedback. Instructor involvement had no significant effect on overall completion, though it did raise completion of the forum badge the intervention targeted (Tomkin and Charlevoix, L@S 2014). Presence helps with what it targets. It isn't a general completion fix.

Reich and Ruipérez-Valiente draw a harder conclusion: established research suggests human connections through advisors, tutors and peer groups are the most important student supports, and those cost money, which pushes against low prices (Reich and Ruipérez-Valiente, 2019).

A quick summary of the evidence

For design ideas that fit these findings, see spaced repetition and the forgetting curve for scheduling, microlearning for smaller units, and gamification in learning for the mechanics some products use to bring people back.

When completion is the wrong metric

Completion is easy to count, which is why it's everywhere. It's also a poor proxy in several situations.

When learners have different goals. A developer who watches three videos to solve a problem and leaves got what they came for. Counting them as a dropout punishes a course for working. The Kizilcec trajectories and Reich's intent data both make this point.

When the course isn't linear. In reference or just-in-time content, nobody should "finish." Short, lookup-style content is built to be used in pieces.

When assigned training meets completion targets. If the goal is a 100% completion rate, people will click through. The number tells you someone reached the last slide. It doesn't tell you they can do anything new.

When a certificate isn't the outcome. Jordan noted that people can benefit from a course without completing its assessments. The HarvardX report found about 36,000 registrants who explored at least half the content without a certificate. A buyer who cares about job performance doesn't care about the certificate count.

Better companions to completion are the share of active learners who reach a meaningful milestone, retention across the next course or module, and an on-the-job outcome the buyer cares about. Learner engagement metrics lays out how to measure those, and proving training ROI covers how to connect them to business results.

What this means for EdTech companies

Completion rate shows up in sales conversations, board decks and renewal reviews. A few habits keep it useful.

1. Always state the denominator. "Completion rate" without "of enrollees" or "of active learners" invites a bad comparison.
2. Track the funnel before the course. If a large share of registrants never start, the problem is activation, not course design. See the online learning enrollment funnel.
3. Segment by intent and cohort. Blended averages hide the learners you could actually help.
4. Test your own nudges. The 247-course study shows that effects shrink at scale and vary by context, so pilot and measure.
5. Don't borrow benchmarks across populations. A MOOC's 6.5% median doesn't predict a paid corporate program, and a vendor's 90% cohort figure doesn't describe a free course.

Related reading

• Learner Engagement Metrics
• Cohort-Based vs Self-Paced Learning
• Online Learning Enrollment Funnel
• Microlearning
• Proving Training ROI