TL;DR. The list price of an online course is the wrong number to judge it by. The number that matters is the cost per completed outcome. In six years of HarvardX and MITx data, only 3.13% of all participants completed a course in 2017-18, while 46.02% of learners who paid for the verified track did (Reich and Ruipérez-Valiente, 2019). Divide a $49 certificate fee by that rate and the real cost is about $106.48 per certificate earned. On a subscription the meter runs on time: at $39 to $79 a month, a Specialization with a 3 to 6 month timeframe costs $117 to $474. Paying is partly a commitment device and partly a filter. In a 2018 edX programme, learners given free access completed less often (50%) than learners who paid (77%), and in the EU 88.5% of participation in job-related non-formal training in 2022 was employer-sponsored. CEO move: price every course by cost per completed, visible outcome, and ask who else should be paying. Student move: audit free first, and pay only at the point where the credential or the commitment is the thing you are missing.
This piece sits in the CEOtudent Skill Strategy cluster. Start with the most in-demand skills of 2026 to choose what to learn, and run the skill audit to see what you already have. What a certificate signals is covered in certificates vs. portfolios vs. degrees, and typical list prices in the digital product price benchmarks. This article adds the missing step: what a course costs once you count the people who never finish.
This piece is general information about learning costs and published research. It is not financial, tax or career advice for your situation.
Why is list price the wrong number?
A course price tells you what you pay to start, not what you pay to finish. A business would not judge a supplier by unit price if most deliveries never arrived; it would look at cost per delivered unit. The same logic applies to learning:
Effective cost per completed outcome = price paid / probability of completing.
A $49 course you have a 46% chance of finishing costs more per finish than its sticker price. A free course you have a 3% chance of finishing costs nothing in cash but a great deal in abandoned hours. Neither number shows up on a checkout page.
There is a second hidden variable. On a per-course price, the cost is fixed and the risk is non-completion. On a subscription, the cost grows every month, so the risk is slowness. Most “is it worth it?” debates mix the two up.
The CEO + Student lens splits the job. The CEO owns the capital decision: expected cost per result, and whether a cheaper payer exists. The student owns execution: what will make you finish, and how fast.
What does the completion data actually show?
The largest open dataset on this question comes from Harvard and MIT. Reich and Ruipérez-Valiente (2019) analysed every HarvardX and MITx course on edX up to May 2018: 565 course iterations from 261 courses, with 12.67 million registrations from 5.63 million learners. Their supplementary Table S3 reports completion for three overlapping groups: all participants, learners who said in a survey that they intended to complete, and learners who paid for the verified certificate track.
| Year | All participants | Said they intended to complete | Paid for verified track |
|---|---|---|---|
| 2013-14 | 4.96% | 11.51% | 44.23% |
| 2014-15 | 5.91% | 17.74% | 50.17% |
| 2015-16 | 4.13% | 16.44% | 49.73% |
| 2016-17 | 3.98% | 16.13% | 56.43% |
| 2017-18 | 3.13% | 15.42% | 46.02% |
Source: Reich and Ruipérez-Valiente (2019), Science, supplementary Table S3 (percentages as printed). Cohort sizes in 2017-18: 980,589 participants, 42,005 intenders, 46,326 verified learners. CEOtudent recomputed each rate from the printed counts; three of the all-participant rates (2013-14 to 2015-16) differ from the printed figure by 0.01 to 0.06 points, so the printed values are used.
Three more facts from the paper sharpen the picture. Of all registrants, 52% never entered the courseware at all (Table S4). Across the 565 course iterations, the median course certified 2.88% of its registrants. And the authors note that completion rates “have not improved over six years” despite heavy investment in course design.
In 2017-18, paying learners completed at 14.7 times the rate of all participants and 3.0 times the rate of learners who said they intended to finish. The second ratio is the fairer comparison because it roughly holds intent constant, but it is still not causal: people who pay differ in money, time and motivation.
The data ends in 2018 and covers two universities on one platform. Kizilcec and colleagues (2020) cite a similar order of magnitude: completion “around 20%” among students who intend to complete in Harvard, MIT and Stanford MOOCs. No platform publishes comparable data today, so these remain the best open benchmarks.
Does paying make people finish, or do finishers pay?
Both, and the split matters for your decision.
Evidence that price filters who enrols. Littenberg-Tobias, Ruipérez-Valiente and Reich (2020) studied edX courses for educators where some cohorts received coupons that made the normally $49 verified certificate free. In courses without coupons, 3% of participants paid for the verified track. In courses with coupons, 13% signed up, about four times as many. Price decides who is in the room.
Evidence that payment adds some commitment. In the same paper’s second case study, a 2018 quantum computing programme priced at $3,900 (or $1,700 per course), learners who received free access completed less often (50%) than learners who paid (77%). Part of that gap came from free-track learners who never clicked into the course (26% of free-track non-completers). In the first case study, the engagement gap between verified learners and auditors was 0.45 to 0.51 standard deviations smaller when the certificate was free, which suggests that paying added some effort on top of selection.
Evidence that light-touch commitment tools do not substitute. Kizilcec, Reich and colleagues (2020) tested established behavioural interventions on about a quarter of a million students across 247 online courses from Harvard, MIT and Stanford over 2.5 years. Long-term planning prompts did not improve completion for committed English-fluent students (an estimated 0.19 percentage points in year one, with a 95% confidence interval from -0.69 to 1.07). Self-regulation prompts raised engagement in the first weeks but not final completion.
The working conclusion: paying is a weak but real commitment device and a strong filter. If you would finish only because you paid, payment buys something. If you would finish anyway, it buys a credential. If you would not finish either way, it is a donation.
What does a completed certificate actually cost?
Athey and Palikot (2025), in a paper written in collaboration with Coursera, describe the platform’s model this way: most courses can be audited for free, a certificate for an individual course typically costs $29 to $99, and Specializations and Professional Certificates usually cost $39 to $79 a month. The $49 point is the edX verified fee in the coupon study above. Combining those price points with the verified-track completion range from Table S3 gives an effective cost per completed certificate.
CEOtudent analysis of Reich and Ruipérez-Valiente (2019), Athey and Palikot (2025) and Littenberg-Tobias et al. (2020): effective cost per completed single-course certificate
| Certificate price | At best verified completion (56.43%, 2016-17) | At latest verified completion (46.02%, 2017-18) | At lowest verified completion (44.23%, 2013-14) |
|---|---|---|---|
| $29 | $51.39 | $63.02 | $65.57 |
| $49 | $86.83 | $106.48 | $110.78 |
| $99 | $175.44 | $215.12 | $223.83 |
| Multiplier on list price | 1.77x | 2.17x | 2.26x |
Method: effective cost = price / verified-track completion rate. The completion rates are edX HarvardX and MITx figures from 2013-2018 and the prices are a 2025 description of Coursera; combining them assumes a similar completion rate on today’s paid tracks, which no platform publishes. Treat the table as an order of magnitude, not a quote. Recomputed in calc.py.
The same arithmetic applied to free courses gives a time-based reading. At the 2017-18 rate for all participants (3.13%), there were about 31.9 starts for every finish. Among learners who intended to finish (15.42%), there were about 6.5 starts per finish. Among verified learners (46.02%), about 2.2. A free course is cheap per attempt, but if your history looks like the average participant, you will pay in attempts.
The multiplier on a paid certificate is small and predictable, roughly double. The hidden cost of free learning is measured in abandoned starts rather than dollars.
How does a subscription change the math?
Subscriptions are becoming the default. Udemy reported that 72% of its 2025 revenue came from subscriptions, up from 66% in 2024. After the Coursera and Udemy merger closed on 11 May 2026, the combined company introduced “Paid Subscribers” as a key metric and stopped disclosing registered learners.
On a subscription, the question shifts from completion to pace: how many months will it take? Coursera’s 10-K lists intended completion timeframes for each product and notes that actual time varies by learner.
CEOtudent analysis of Coursera’s 2025 Form 10-K and Athey and Palikot (2025): what the subscription meter can charge for one programme
| Coursera product (intended timeframe in the 10-K) | Fastest finish at $39/month | Slowest finish at $79/month | Spread |
|---|---|---|---|
| Specialization (3 to 6 months) | $117 | $474 | 4.1x |
| Entry-level Professional Certificate (3 to 9 months) | $117 | $711 | 6.1x |
| Non-entry-level Professional Certificate (1 to 10 months) | $39 | $790 | 20.3x |
Method: monthly price range x intended month range. Not a quote of any current price; prices are localized by region and change with promotions. Recomputed in calc.py.
Two features stand out. The spread between a fast learner on a cheap plan and a slow learner on an expensive one can reach 20 times for the same credential. And every extra month costs $39 to $79, so pausing without cancelling is the most expensive habit in online learning.
One regional price snapshot shows how annual plans shift this. On 15 November 2026, the Coursera Plus page served to the CEOtudent research visitor listed $35 a month after a 7-day free trial, or $239 a year, discounted to $143.40 until 30 November 2026. On those numbers, the annual plan only beats the monthly plan if you would otherwise stay subscribed for more than 6.83 months, or 4.10 months at the promotional price. Coursera localizes prices, so the figures you see may differ; the break-even method is what transfers.
Where is the industry’s own money coming from?
The vendors’ filings show what learners actually pay, on average, and how many pay at all. These are company-reported figures from audited annual reports and quarterly filings.
| Measure (company-reported) | Figure | Source |
|---|---|---|
| Coursera cumulative Registered Learners, end of 2025 | 197.3 million | Coursera 2025 Form 10-K |
| Coursera New Registered Learners in 2025 | 29.1 million | Coursera 2025 Form 10-K |
| Coursera Consumer revenue, 2025 | $502.2 million | Coursera 2025 Form 10-K |
| Share of 2025 cash receipts from courses, Specializations and certificates paid by learners registered before 2025 | over 69% | Coursera 2025 Form 10-K |
| Udemy Consumer segment revenue, 2025 | $265.8 million (2024: $292.1 million) | Udemy 2025 Form 10-K |
| Udemy monthly average buyers, 2025 | 1,275 thousand | Udemy 2025 Form 10-K |
| Udemy paid Consumer subscribers, end of 2025 | 343 thousand (2024: 170 thousand) | Udemy 2025 Form 10-K |
| Coursera and Udemy combined Paid Subscribers, 30 September 2026 | 1,687 thousand (a year earlier: 1,299 thousand) | Coursera Form 10-Q, Q3 2026 |
Sources: Coursera, Inc. and Udemy, Inc. filings on SEC EDGAR, as listed. The combined subscriber figures are conformed to a unified post-merger methodology, per the 10-Q.
Two CEOtudent derivations put these in learner terms. Udemy’s 2025 Consumer revenue over twelve months of average buyers is about $17.37 per buyer per month. Coursera’s 2025 Consumer revenue over its cumulative registered learners is about $2.55 per registered learner, which mainly shows how many accounts pay little or nothing in a year. Neither is a price; both mix one-off purchases, subscriptions and revenue-recognition timing.
The direction is clear: Udemy’s Consumer revenue fell while its paid subscribers grew 102%, and the combined company’s paid subscribers grew 30% in a year. The industry is moving from selling courses to selling time.
Coursera’s own 10-K lists among its risks that learners may hold “a perception that the financial investment may not result in increased earning potential or improved employment opportunities.” That is the question this piece is trying to answer.
Does a paid credential pay off in the job market?
The best recent evidence is more modest than marketing pages suggest, and more specific.
Employment retention, not wages. Castaño-Muñoz (European Commission Joint Research Centre) and Rodrigues (2021) compared learners who took one of six MOOCs in 2015 with a control group who first enrolled in 2017. Using a differences-in-differences design, they estimate that MOOC participation raised the probability of still being employed two years later by 7.4 percentage points (8.3 for a business intelligence course, 7.1 for a communication and marketing course). They found no evidence of a wage effect. The study covers MOOC participation, not paid certificates, and 346 people.
Visibility of the credential, especially for weaker profiles. Athey and Palikot (2025) randomly nudged over 800,000 Coursera certificate earners from developing countries without college degrees to share their certificates on LinkedIn. Among 40,000 tracked for employment, new employment rose 6% (1.0 percentage point), and 9% (1.2 percentage points) for certificate-related jobs. Gains concentrated among learners with weak baseline employability: 12% for the bottom third, negligible for the top third. The paper is a working paper written in collaboration with Coursera.
Put together: a credential earns its price when it fills a visible gap in a thin profile and is shown to employers. For a strong profile, the same certificate adds little.
Should you pay, should your employer pay, or should nobody?
For job-related learning, the default payer in Europe is the employer, not the learner. Eurostat’s 2022 Adult Education Survey shows the split for adults aged 25 to 64 across the EU.
| EU27, adults 25-64, 2022 | Figure |
|---|---|
| Participated in formal or non-formal education and training (last 12 months) | 46.6% |
| Participated in job-related non-formal education and training | 38.1% |
| of which sponsored by the employer | 33.7% |
| of which not sponsored by the employer | 4.4% |
| Among those who wanted to participate but did not: cited costs | 28.7% |
| Among those who wanted to participate but did not: cited schedule | 39.6% |
| Among those who wanted to participate but did not: cited lack of employer support | 13.2% |
Source: Eurostat, Adult Education Survey 2022, datasets trng_aes_100, trng_aes_120 and trng_aes_176 (accessed via the Eurostat API on 15 November 2026). Reasons are not mutually exclusive.
By CEOtudent’s calculation, 88.5% of job-related non-formal participation in the EU was employer-sponsored. Among adults who wanted to learn but did not, schedule was cited more often than cost: time, not money, is the more common constraint.
In the United States, an employer can pay up to $5,250 of an employee’s educational assistance per year tax-free under a Section 127 educational assistance program, according to the IRS and its Publication 15-B for 2026. Publication 15-B also notes that covered expenses generally exclude courses involving sports, games or hobbies, unless the education has a reasonable relationship to the employer’s business or is required as part of a degree program. At $39 a month, $5,250 would cover about 134.6 months of a subscription, so the binding question is whether the employer has a qualifying program. Rules vary; this is not tax advice.
The CEO question is therefore not only “is this worth the money?” but “whose money should it be?”. For a job-related skill, asking the employer first is the highest-return move in this article.
The Pay-or-Don’t-Pay test
CEOtudent editorial framework. Score each question 0 or 1. The framework turns the evidence above into a decision you can make in ten minutes.
| # | Question | Score 1 if… | Evidence it rests on |
|---|---|---|---|
| 1 | Have you audited the free version for at least one week? | Yes, and you still want to continue | 52% of registrants never open the courseware; testing fit is free |
| 2 | Is a visible credential part of what you need? | You will show it to an employer or client, and your profile is thin in this area | Athey and Palikot (2025): gains concentrate among weaker profiles |
| 3 | Would money on the line change your behaviour? | You have abandoned free courses before | Littenberg-Tobias et al. (2020): free-track learners completed less (50% vs 77%) |
| 4 | Can someone else pay? | You have checked and the answer is no | Eurostat 2022: 88.5% of job-related non-formal participation was employer-sponsored |
| 5 | On a subscription, do you have the hours blocked? | You can name the weekly slots and a finish month | Coursera intended timeframes; cost spread up to 20.3x |
| 6 | Is the effective cost (price / your honest completion odds) still worth the outcome? | Yes, even at double the list price | Verified-track multiplier of 1.77x to 2.26x |
Decision rules (CEOtudent editorial framework):
- 5 to 6: pay, and pay in the cheapest format for your pace (per-certificate if you are slow, monthly if you are fast).
- 3 to 4: audit free now, set a date, and pay only at the point the certificate or graded work becomes necessary.
- 0 to 2: do not pay. Either the skill is not the right one yet (go back to the 4-factor decision matrix) or the bottleneck is time, not money.
Free AI tutors can replace paid explanations, but not a credential or an employer’s budget, so questions 2 and 4 stand. Which tools teach rather than just answer is covered in AI tutors in 2026.
A worked example: four ways to learn the same skill
This example uses illustrative assumptions, labelled as such; every number is reproduced in calc.py.
| Option | Cash price | Completion probability | Effective cost per completion |
|---|---|---|---|
| A. Audit free | $0 | 15.42% (illustrative assumption: the 2017-18 rate for learners who intended to finish) | $0 cash, about 6.5 starts per finish |
| B. Single certificate | $49 (illustrative assumption) | 46.02% (illustrative assumption: the 2017-18 verified-track rate) | $106.48 |
| C. Subscription, 5 months | $39 x 5 = $195 (illustrative assumption) | 46.02% (illustrative assumption) | $423.73 |
| C2. Same subscription, 8 months | $39 x 8 = $312 (illustrative assumption) | 46.02% (illustrative assumption) | $677.97 |
| D. Employer pays under Section 127 | $0 to the learner, up to $5,250 a year | Depends on the learner | $0 cash to the learner |
If B and C teach the same skill, B is cheaper, and C wins only if it adds something B lacks, such as graded projects or a recognised credential. Slipping from 5 to 8 months raises C’s effective cost by 60%. Option A costs no cash but implies several starts per finish. Option D beats all of them when available. The dollar figures depend on the assumptions; the ranking logic does not: completion probability and pace move the result more than the sticker price.
How do you lower your cost per skill?
- Audit before you buy, and upgrade at the point of need (graded work or the certificate), not at enrolment.
- Book the hours before the subscription, and cancel between sprints. The 90-day personal curriculum is a template.
- Ask the employer first for job-related skills.
- Make it visible. The employment gains in the best recent experiment came from sharing credentials, not merely holding them.
What this does not mean
- Paid tracks do not simply cause completion. The verified-track rate reflects who chooses to pay as much as what paying does.
- The 2013-2018 rates are benchmarks, not a current measurement. Platforms do not publish comparable completion data.
- Free learning is not worthless. For many skills, free material plus deliberate practice is enough.
- A certificate guarantees nothing. The strongest studies find effects on employment retention and on new jobs for weaker profiles, and no evidence of a wage effect in the one study that measured wages.
- Prices are snapshots. Re-run the formula with the price you actually see.
Frequently asked questions
Is it worth paying for an online course certificate?
When you need a visible credential, your profile is thin in that area, and money on the line would help you finish. Budget roughly twice the list price per completed certificate, based on verified-track completion of 44% to 56% in HarvardX and MITx data.
Do people who pay for online courses finish more often?
Yes. In 2017-18, 46.02% of verified-track learners completed HarvardX and MITx courses against 3.13% of all participants and 15.42% of those who intended to finish. Much of that gap likely reflects selection; a coupon study found a smaller but real effect of paying itself.
Is a subscription cheaper than buying single courses?
Only if you are fast. At $39 to $79 a month, a Specialization with an intended timeframe of 3 to 6 months can cost $117 to $474. If you take longer or pause without cancelling, a single-course purchase can be cheaper.
Can an employer pay for online courses tax-free?
In the United States, an employer can provide up to $5,250 a year of educational assistance tax-free under a qualifying Section 127 program, per IRS Publication 15-B for 2026. Eligibility rules apply; check with your employer or a tax professional.
Do free AI tutors make paid courses unnecessary?
They can replace paid explanations and practice, but not a recognised credential or an employer’s training budget, two of the main reasons to pay.
Sources
- Reich, J. and Ruipérez-Valiente, J. A. (2019). The MOOC Pivot. Science, 363(6423), 130-131. Author post-print and supplementary Tables S3 and S4.
- Littenberg-Tobias, J., Ruipérez-Valiente, J. A. and Reich, J. (2020). Studying Learner Behavior in Online Courses With Free-Certificate Coupons: Results From Two Case Studies. International Review of Research in Open and Distributed Learning, 21(1).
- Kizilcec, R. F., Reich, J., Yeomans, M., Dann, C., Brunskill, E., Lopez, G., Turkay, S., Williams, J. J. and Tingley, D. (2020). Scaling up behavioral science interventions in online education. Proceedings of the National Academy of Sciences, 117(26).
- Castaño-Muñoz, J. and Rodrigues, M. (2021). Open to MOOCs? Evidence of their impact on labour market outcomes. Computers and Education, 173, 104289.
- Athey, S. and Palikot, E. (2025). The Value of Non-Traditional Credentials in the Labor Market. arXiv working paper 2405.00247, version 2, 5 November 2025 (conducted in collaboration with Coursera).
- Coursera, Inc. (2026). Annual Report on Form 10-K for the fiscal year ended 31 December 2025. SEC EDGAR.
- Coursera, Inc. (2026). Current Report on Form 8-K, filed 11 May 2026 (completion of the combination with Udemy, Inc.). SEC EDGAR.
- Coursera, Inc. (2026). Quarterly Report on Form 10-Q for the quarter ended 30 September 2026. SEC EDGAR.
- Udemy, Inc. (2026). Annual Report on Form 10-K for the fiscal year ended 31 December 2025. SEC EDGAR.
- Eurostat (2026 update). Adult Education Survey 2022: datasets trng_aes_100, trng_aes_120 and trng_aes_176. Accessed 15 November 2026.
- Internal Revenue Service. Frequently asked questions about educational assistance programs. Accessed 15 November 2026.
- Internal Revenue Service. Publication 15-B (2026), Employer’s Tax Guide to Fringe Benefits. Accessed 15 November 2026.
- Coursera. Coursera Plus pricing page, as served to an unauthenticated visitor on 15 November 2026 (company-published, regional and promotional pricing).
Note on methods: the effective-cost table, the subscription-meter table, the per-buyer and per-learner revenue figures, the 88.5% employer-sponsored share, the break-even months and the worked example are CEOtudent analyses of the sources above, recomputed in a script. The Pay-or-Don’t-Pay test is a CEOtudent editorial framework. Vendor figures are company-reported. We did not use widely repeated vendor learner-outcome survey percentages, because they come from self-selected respondents. We also did not use figures from a 2020 hiring-preference experiment on MOOC credentials, because a 2021 correction to that paper exists and we could not open it.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
















