Key takeaways
- Length predicts completion more than anything else we measured. Completion falls at every step as courses get longer, from 85.5% under 5 minutes to 70.6% over 15 minutes.
- Completion without assessment measures very little. Courses with no questions completed at 90.7%, and 80% of learners who failed to complete a course with questions had already reached the end of the content.
- The first course is the hardest one to finish. Of learners who completed one course, 57.4% completed a second, and 73.3% of those completed a third.
- Collaboration costs time and buys reach. Courses with three or more authors took 12.8 days to reach first publish against about 3 hours for solo courses, and they reached 3.4 times more learners.
How we measured this
All figures come from Easygenerator product data covering the 12 months from June 2025 to June 2026. The dataset holds 213,660 created courses and 22.9 million learner-course interactions. Completion rates include only courses with at least 5 learner results. Course length and drop-off points cover courses delivered through Easygenerator-hosted links, so courses exported as SCORM packages and played in an external LMS fall outside those two measurements.
These are observational comparisons, not controlled tests. Authors who adopt AI, collaborate, or build short courses may differ in ways this data cannot capture.
Disclosure. Easygenerator is the publisher of this research and the source of the data. Every figure describes activity on the Easygenerator platform rather than the e-learning industry as a whole.
Completion falls as courses get longer
Courses under 5 minutes on the Easygenerator platform reached an average completion rate of 85.5%. Courses between 5 and 15 minutes reached 81.5%, and courses over 15 minutes reached 70.6%. The pattern holds at every step, and the gap between the shortest courses and the longest ones comes to almost 15 percentage points.
85.5%
Courses under 5 minutes on the Easygenerator platform reached an average completion rate of 85.5%, compared with 70.6% for courses running over 15 minutes.
Easygenerator platform data, 3,802 courses and 730,747 learner results, June 2025 to June 2026.
Easygenerator does not ask authors to declare how long a course should take, so we measured length from what learners actually did. Each course carries the median time spent by learners who reached the end, and that becomes its effective length.
That method matters for how much weight the gap can carry. The longest group is not made up of courses that barely crossed the 15 minute line. The typical course in that group runs close to 32 minutes, so the comparison sits between genuinely short courses and genuinely long ones.
| Under 5 minutes | 5 to 15 minutes | Over 15 minutes | |
|---|---|---|---|
| Courses measured | 704 | 1,232 | 1,866 |
| Learner results | 117,414 | 224,870 | 388,463 |
| Typical actual length | 2.6 min | 9.3 min | 31.9 min |
| Completion rate, average per course | 85.5% | 81.5% | 70.6% |
| Completion rate, pooled | 87.0% | 85.0% | 67.6% |
Course length measured as the median time spent by learners who reached the end. Sessions over 8 hours excluded as left-open browser tabs.
The pooled measure widens the gap rather than narrows it. Pooled completion counts every learner rather than every course, so courses with heavy traffic weigh more. On that measure, short courses reached 87.0% and long courses fell to 67.6%. Long courses lose more ground once you weight by learner volume, which suggests the effect grows at scale.
Length is not the only thing at work here. Long courses tend to cover harder subjects, they carry more assessment, and they show up more often in mandatory programs where learners have less say about when they start. The data shows a strong pattern rather than a mechanism, and a 40 minute course on a complex topic is not the same product as a 3 minute refresher.
The practical read for L&D teams is to treat 15 minutes as a working ceiling rather than a target. Where the subject genuinely needs 40 minutes, the option is to split it rather than to cut it, because three short courses give a learner three finish lines instead of one.
That last point connects to something else in this data. Learners who complete one course become much more likely to complete the next one, which means a split course does more than reduce drop-off. It also builds the habit that carries a learner through the rest of the program.
Courses without questions look better and measure less
Courses with no questions on the Easygenerator platform completed at 90.7%, against 72.3% for courses with 1 to 3 questions and 71.6% for courses with 4 or more. The 19 point gap looks like an argument for cutting assessment out of courses. The drop-off data says something close to the opposite.
| 0 questions | 1 to 3 questions | 4 or more questions | |
|---|---|---|---|
| Courses measured | 11,413 | 9,865 | 23,936 |
| Completion rate, average per course | 90.7% | 72.3% | 71.6% |
| Completion rate, pooled | 94.5% | 84.7% | 83.8% |
Notice where the effect stops. Completion falls sharply between 0 questions and 1 question, and then it barely moves. Courses with 4 or more questions completed within a point of courses with 1 to 3. The first question costs almost all of the completion, and every question after that costs very little.
To understand what the gap actually measures, we looked at where learners were last seen when they failed to complete a course.
| Courses with no questions | Courses with questions | |
|---|---|---|
| Non-completing learners measured | 41,698 | 280,033 |
| Share who reached the end of the content | 53% | 80% |
| Average point reached by those who left mid-course | 79.7% through | 60.8% through |
In courses with questions, 80% of the learners who did not complete still reached the end of the content. They worked through the whole course and then missed the pass mark. In courses without questions, only 53% of non-completers got that far, so nearly half of them stopped somewhere in the middle and never came back.
| 80% | In Easygenerator courses that contain assessment questions, 80% of learners who did not complete had already reached the end of the course content.Easygenerator platform data, 280,033 non-completing learners, June 2025 to June 2026. |
One detail runs the other way, and it is worth stating. Among the learners who did leave mid-course, those in question-free courses got further before they stopped, 79.7% of the way through against 60.8%. The learners who quit a question-free course quit later. There are just far more of them.
What the numbers actually say
A course with no questions counts as completed the moment a learner clicks through to the end, because there is no bar to clear. Knowledge checks trade a few points of reported completion for evidence that someone learned something.
For anyone who reports completion rates upward, the practical consequence is that the two numbers are not comparable. A 90% completion rate on a course without assessment and a 72% rate on a course with assessment describe different achievements, and putting them in the same dashboard column makes the weaker course look stronger.
The more useful pair of numbers to track is the share of learners who reach the end of the content, alongside the share who pass. Those two answer separate questions, and only the second one tells you anything about learning.
That raises an obvious follow-up. If most non-completers reach the end and miss the pass mark, what happens to them next?
Learners who finish one course are more likely to finish the next
Of the learners who completed one course on the Easygenerator platform, 57.4% went on to complete a second. Of those who completed a second, 73.3% completed a third. The odds improve at every step, which makes the first course the hardest one to get a learner through.
|
57.4% Complete a 2nd course after their 1st |
73.3% Complete a 3rd course after their 2nd |
Completion compounds. Once a learner finishes a course on the Easygenerator platform, the odds they finish the next one rise from 57.4% to 73.3%.
Easygenerator platform data, drawn from 22.9 million learner-course interactions, June 2025 to June 2026.
Both figures understate the effect slightly. A learner who completed their first course in May 2026 may well complete their second in July, which falls outside the measurement window. The real second-course rate is higher than 57.4%.
We cannot separate the two explanations behind this pattern. One reads that completion builds a habit and each finish makes the next one easier. The other reads that learners who complete a first course were already the ones most likely to complete anything, so the second course simply filters for people who were never going to drop out. Both are probably true to some degree, and this data cannot tell them apart.
What the pattern does change is where effort pays off. A program that gets 100 learners through one course starts its second course with a much better base rate than a program that never got them through the first. Front-loading attention onto the first course in a sequence looks like a better use of design time than spreading it evenly across all six.
Most learners who fail an assessment come back and pass
Among 783,646 learner-course combinations on the Easygenerator platform where a learner failed at least one attempt, 70.1% eventually passed. Another 23.0% failed and never returned. Most learners who hit a failing grade tried again.
| 77% | Of learners who failed an assessment on the Easygenerator platform, 77% returned for another attempt, and 70.1% eventually passed.Easygenerator platform data, 783,646 learner-course combinations with at least one failed attempt, June 2025 to June 2026. |
Read this next to section 2 and the two findings fit together. Most learners who fail to complete a course with questions have reached the end of the content, and most learners who fail an assessment come back to retake it. A failing grade acts as a pause in the process rather than an exit from it.
The 23.0% who never return are the group worth designing for. This data cannot say why they stopped, and the plausible reasons pull in different directions. Some courses sit inside compliance programs where a retake needs a manager to reset it. Some learners never learn that a second attempt is available. Some simply give up.
The useful move for an L&D team is to check which of those applies in their own setup, because the fix differs completely. A retake that needs administrative approval is a process problem. A retake that nobody knows about is a communication problem.
Collaborative courses take longer to build and reach more people
Courses with a single author on the Easygenerator platform reached first publish in about 3 hours. Courses with two authors took 7.1 days, and courses with three or more took 12.8 days. Those same collaborative courses reached about 3.4 times as many learners.
| 1 author | 2 authors | 3 or more authors | |
|---|---|---|---|
| Courses measured, publishing speed | 78,797 | 16,602 | 10,120 |
| Typical time to first publish | 0.13 days, about 3 hours | 7.1 days | 12.8 days |
| Courses measured, completion and reach | 31,805 | 8,086 | 5,323 |
| Completion rate, average per course | 77.9% | 73.7% | 72.9% |
| Completion rate, pooled | 86.3% | 83.4% | 86.9% |
| Average learners per course | 187 | 280 | 635 |
3.4x
Courses built by three or more authors on the Easygenerator platform reached an average of 635 learners, about 3.4 times the 187 learners reached by courses built by a single author.
Easygenerator platform data, 45,214 courses, June 2025 to June 2026.
The time difference is large enough that it needs explaining rather than defending. A solo author publishes in an afternoon because nobody has to wait for anybody. Every co-author adds a review cycle, a calendar, and a round of comments.
What the data suggests is that teams do not add co-authors at random. They add them to the courses that matter most, and the reach figures show it. A course that 635 people will take earns a review cycle in a way that a course for a team of 12 does not.
The completion figures hold an oddity that runs against the trend. Courses with three or more authors scored the lowest average completion per course at 72.9%, and the highest pooled completion at 86.9%. Pooled completion weights by learner numbers, so the collaborative courses that pull the heaviest traffic complete very well. The average gets dragged down by smaller collaborative courses that reach few people.
For an L&D team the read is about matching process to stakes. A course for a small team can go out in an afternoon with one author. A course that will reach several hundred people justifies the two weeks, and the data gives a realistic figure to plan against rather than an optimistic one.
Courses built with EasyAI reach first publish about 30% faster
The typical course built with EasyAI on the Easygenerator platform reached first publish in about 16 hours, against about 23 hours for courses built without it. Authors who used EasyAI also published twice as many courses in an active quarter, a median of 2 against 1.
|
16 hours Built with EasyAI |
23 hours Built without EasyAI |
The typical course built with EasyAI reached its first publish in about 16 hours, around 30% faster than the 23 hours taken by courses built without it.
Median time from creation to first publish. Easygenerator platform data, 106,126 courses, June 2025 to June 2026.
| Built with EasyAI | Built without EasyAI | |
|---|---|---|
| Courses measured | 20,329 | 85,797 |
| Typical time to first publish | 0.67 days, about 16 hours | 0.96 days, about 23 hours |
| Average time to first publish | 15.4 days | 16.8 days |
The averages sit far above the medians in both groups, and the reason matters. A small number of courses sit in draft for months before anyone publishes them, and those courses drag the average up. The median describes what a typical course does, so it carries the headline here.
Note also that the two averages sit close together, 15.4 days against 16.8. The speed advantage shows up clearly in the typical case and much less clearly once the long-abandoned drafts enter the calculation.
| EasyAI users | Never used EasyAI | |
|---|---|---|
| Authors measured | 5,368 | 8,700 |
| Typical courses per active quarter | 2 | 1 |
| Average per active quarter | 4.37 | 4.27 |
The same pattern repeats in the publishing figures. The typical EasyAI user publishes twice as many courses per active quarter, while the two averages are almost identical at 4.37 and 4.27. Both groups contain a small number of authors who publish a great deal, and those authors flatten the difference between the averages.
Authors choose whether to use AI, and nobody assigned them to a group. The authors who reach for EasyAI may already work faster, may build simpler courses, or may work in teams with fewer approval steps. A 30% difference in typical publish time is a real pattern in the data.
Courses built with EasyAI reach publication faster than courses built without it, among authors who chose to use it.
Drafts are more likely than published courses to contain EasyAI content. That fits a picture where AI lowers the effort needed to start something, so people start more experiments, and more experiments never finish. It also connects to the largest single finding in this dataset.
The gap between drafting and publishing is where most e-learning stalls
Authors created 213,660 courses on the Easygenerator platform over 12 months and published more than 105,000 of them. The other half never shipped. Once you set aside courses created too recently to have had a fair chance, 48% of courses still sat unpublished after at least 90 days.
105,550
Authors published 105,550 courses on the Easygenerator platform in 12 months. A roughly equal number of courses were created and never reached learners at all.
Easygenerator platform data, 213,660 courses created, June 2025 to June 2026.
Half of all created courses never reach a learner. That number surprises people, and the first instinct is to read it as a problem. The draft data argues against that reading.
| 1 author | 2 authors | 3 or more authors | |
|---|---|---|---|
| Unpublished courses | 94,399 | 9,838 | 3,873 |
| Average days in draft so far | 167 | 163 | 175 |
| Typical days in draft so far | 154 | 148 | 176 |
Draft age barely moves across team sizes. A solo author’s abandoned draft has been sitting for a typical 154 days, and a draft with three or more authors for 176. If review cycles or approval chains caused courses to stall, collaborative drafts would sit far longer than solo ones. They do not.
Read those numbers carefully, because the clock is still running on every course in that table. The figures say that the typical unpublished draft has been sitting for about five months so far. They do not say that courses take five months to publish.
Two things about this pattern make it look less like an e-learning problem and more like a content problem in general. Drafts stall at the same rate whoever works on them. And the courses that do ship, ship fast, with a solo author publishing in about 3 hours and a typical course of any kind publishing inside a day.
Our own guess is that most of those drafts represent ideas rather than commitments. Somebody spins up a course to see whether a topic works, the priority shifts, and nobody deletes it.
For an L&D team the useful question is not how to stop people from making drafts. Cheap experiments are a good thing, and a platform where half the ideas die early is a platform where people feel free to try. The question is whether any of the drafts sitting in the account right now are courses that somebody genuinely needs and nobody has chased. A quarterly look at drafts older than 90 days usually finds two or three of those.
What to do with these findings
The findings above describe patterns rather than instructions, and the right response depends on what you are building. The table below maps each situation to what this data suggests.
| If you are building | What the data suggests |
|---|---|
| A compliance course with mandatory sign-off | Keep the questions and plan for completion near 72%. Track how many learners reach the end of the content as a separate figure, because 80% of non-completers already get there. |
| Onboarding for a large intake | Split it into courses under 5 minutes. Getting a learner through their first course lifts the odds they finish the next one from 57.4% to 73.3%. |
| A topic that genuinely needs 40 minutes | Split it rather than cut it. Completion falls to 70.6% above 15 minutes, and three short courses give a learner three finish lines instead of one. |
| A refresher for a team of 12 | One author, published the same day. Solo courses reached first publish in about 3 hours, and a review cycle adds a week or more for little gain at this scale. |
| A course that will reach several hundred learners | Plan for 12 to 13 days with co-authors rather than hours. Collaborative courses with heavy learner traffic reached 86.9% pooled completion, the highest of any group. |
| A course where you need proof of learning | Add at least one question and do not stop at one. The first question costs about 18 points of reported completion, and every question after it costs under a point. |
| An assessment with a high fail rate | Check whether a retake needs administrative approval, and check whether learners know a retake exists. Of learners who failed, 23.0% never returned for a second attempt. |
| A completion report for leadership | Report end-of-content reach and pass rate as two separate numbers. A 90.7% rate on a course without assessment and a 72% rate on a course with assessment do not measure the same thing. |
| A first course in a longer program | Spend disproportionate design time here. The first completion is the hardest to win, and every completion after it comes more easily. |
| An account with a long list of old drafts | Review drafts older than 90 days once a quarter. The typical stalled draft has been sitting for about five months, and a few of them are usually courses somebody still needs. |
What this data cannot tell you
Every figure in this report describes what happened on the Easygenerator platform. None of them describe the e-learning industry, and several carry limits worth stating plainly.
These are observational comparisons, not experiments. Nobody assigned authors to use AI, to work alone, or to build short courses. The authors who chose each of those things may differ from the authors who did not in ways this data cannot measure. Every comparison here shows a pattern rather than a cause.
The sample reflects who uses Easygenerator. Our customers lean toward Employee-generated Learning, where subject-matter experts build their own courses instead of handing material to an instructional design team. Courses built that way tend to run shorter and to cover narrower topics than courses built by a central team. A platform with a different customer base would likely produce different numbers.
Course length and drop-off cover only part of the picture. Both measurements come from courses delivered through Easygenerator-hosted links. Courses exported as SCORM packages and played inside an external LMS sit outside those two figures. Many enterprise programs deliver that way, so the length and drop-off findings rest on a narrower base than the rest of the report. Course length draws on 3,802 courses, against 45,214 for the completion figures.
Drop-off points are estimated. We take the last course section a learner visited and divide its position by the total number of sections. Section order comes from the order in which authors created the sections, which is how most authors build, though not all of them.
Completion is not learning. A learner who finishes a course and passes its assessment has demonstrated something on the day. This data says nothing about what they remembered a month later, or about whether their work changed.
About Easygenerator
Easygenerator is an AI-powered e-learning suite that helps organizations create company-tailored training at scale. Built for internal experts and L&D teams alike, its author-first AI captures the depth of organizational knowledge and transforms insights, documents, and ideas into didactically-sound courses. Teams can create professional training videos with AI avatars and multilingual voiceovers, and help employees practice real workplace conversations to build skills with confidence. With translation into 75+ languages, Easygenerator makes it simple to scale training consistently across the entire organization.
