Is Self-Improvement Becoming One of the Biggest Consumer Businesses?
Paid learning has produced large businesses, but their growth tells different stories. We examine what Indonesian buyers want to improve, what progress costs to deliver and why a successful customer may eventually need less help.
N
Neverlater
20 October 2026
In this article
An imagined speaking practice session shows a learner receiving feedback from a peer. Photographic editorial illustration; it does not depict an actual course, provider or measured learning outcome.
Self improvement already supports some large, fast growing businesses. Whether it is becoming one of the biggest consumer markets is harder to establish. We use it as a lens on purchases made to gain a capability, change a habit or reach a personal goal. Education, fitness, beauty and wellness overlap, and each also sells enjoyment or belonging. The useful common question is what progress the buyer expects and what the business must deliver to earn the next payment.
Learning gives a concrete illustration of commercial scale. Duolingo recorded US$1.038 billion of revenue in 2025, up 39%, including US$873.4 million from subscriptions, up 44%. It ended the year with 12.2 million paid subscribers, excluding free trials and nonpaying members of family plans. These are global company figures, not the size of Indonesia’s opportunity. Duolingo annual report
Go deeper
What does this mean for your business?
Discuss this article privately with Neverlater. Its title and link stay attached, so you can start with the question that matters to you.
The growth is uneven. In their separate 2025 reports, Coursera recorded US$502.2 million of Consumer revenue, up 10%, while Udemy recorded US$265.8 million, down 9%. Udemy attributed the expected decline to its shift from individual course transactions toward recurring offers. Those consumer segments differ from Duolingo’s subscription revenue; they are examples of paid learning, not numbers to add into a self improvement market. Coursera and Udemy subsequently combined in May 2026.
Our bet is on companies that identify where a customer gets stuck, deliver useful help and can afford to let that customer graduate. In Indonesia, the offer must fit the money and time buyers can commit, with progress they can recognise. Learning gives us a close view of those economics; fitness shows how the same tension appears elsewhere.
1. The buyer has two budgets
2025 global disclosures show different paths: Duolingo subscription revenue rose 44% to US$873.4 million; Coursera Consumer revenue rose 10% to US$502.2 million; Udemy Consumer revenue fell 9% to US$265.8 million during a planned shift toward recurring offers. The reporting scopes differ. This comparison is not a market total or evidence of learning progress.
A course can be affordable at checkout and impossible to complete. Its price is charged in rupiah; its delivery depends on evenings, attention and the willingness to practise something awkward. A gym membership makes a similar demand. The purchase reserves access. The customer still has to find a place for it in an ordinary week.
PwC's 2026 Indonesian consumer panel captures the appetite: 41% put personal growth and self improvement among their top three wellness goals. In separate questions, 21% reported buying health or fitness apps/subscriptions in the previous year, while 35% said they frequently budgeted for health and wellness. These are responses from 503 panel participants, not national population rates or stages of a conversion funnel. PwC Indonesia survey
That leaves an operator with a more demanding task than finding people who want a better life. It must find a goal urgent enough to win both budgets. Preparing for a client presentation next month gives a buyer a deadline, a reason to practise and a way to judge whether the purchase helped. “Become your best self” gives the seller much more room to disappoint.
Price deserves the same specificity. BPS reported an average monthly wage of Rp3.39 million for employees in May 2026. A Rp600,000 programme would equal about 18% of that monthly figure. This national employee average is neither a median nor the income of our prospective customers, but it is a useful warning against casually calling such a fee small. Actual affordability depends on the chosen buyers and their other commitments. BPS wage release
We would begin with a narrow adult audience and one consequential task: for example, Indonesian professionals who already know basic English but struggle to give a concise work update. Research would need to test where their week breaks down. If the answer is unpredictable working hours, adding more scheduled lessons may make the product harder to use.
2. The scarce product is useful attention
Cakap advertises private tutor sessions alongside Fluency AI for independent speaking practice. This diagram summarises the published design, not measured outcomes or provider costs. Source checked 12 October 2026.
The useful comparison between learning companies is how they distribute help. A recorded lesson can serve another learner without booking another teacher. A live correction, a speaking turn in a small group or an individual review consumes capacity. For a provider, increasing the amount of help can change the economics of the entire offer.
Duolingo shows how large a learning product people use independently can become. In Q2 2026, it reported 58.7 million daily active users on average and 12.7 million paid subscribers at quarter end. Management said it expected better learning to support engagement, retention and word of mouth. Those are different measures and a strategic belief, rather than proof that every returning learner is making useful progress. Q2 2026 shareholder letter
Enjoyable practice may earn renewals. The business still needs to test whether the activity that brings a learner back removes the obstacle they paid to solve.
Cakap makes the allocation of teacher attention more explicit. Its English course page offers private tuition and group classes of 10 to 15 students, alongside a speaking placement session and performance report. Lessons on business presentations are advertised from Rp74,500 per session; the page says English classes last 25 minutes. Eight sessions at that starting rate would total Rp596,000 and 200 scheduled minutes, although this is not a verified bookable package. Cakap English courses
The same page pairs private tutor sessions offering explanations and live correction with Fluency AI, an artificial intelligence tool for independent role play and pronunciation practice. This is a concrete version of the allocation problem: use repeatable practice between sessions and reserve a person’s attention for useful correction. The advertised design alone does not establish better learning outcomes or lower delivery costs. Cakap tutor and practice model
For a buyer, those minutes need unpacking. How many will they spend speaking? What will a teacher correct? What happens between sessions? For the operator, the corresponding question is where a tutor changes the result enough to justify scarce time. Repeating a basic explanation and diagnosing why someone freezes during a presentation are different jobs.
The same separation appears in fitness. FIT HUB's membership FAQ includes club and class access, while personal training is a separate package available to members, with options from four to 96 sessions. Access and concentrated human attention have different economics. Neither membership nor a coaching package, by itself, establishes a fitness result. FIT HUB membership FAQ
Our preferred design would make routine practice easy to repeat and reserve expensive help for a specific obstacle. Whether software, peers or a professional can remove that obstacle is an empirical question. “Unlimited support” is a dangerous promise to make before knowing what support people will need.
3. A useful promise has a finish line
An imagined adult online lesson illustrates live teaching. These fictional participants are not a recorded Cakap class, product screenshot or endorsement. Editorial illustration.
Consider the buyer preparing a work update. An eight week offer could aim at one task: communicate the key point and next step within three minutes, explain a delay and answer two new questions clearly enough for a listener to act. Use comparable tasks before and after the programme.
That task would not establish general fluency, a promotion or a higher salary. The Council of Europe provides useful discipline here: assess the actions a task requires, and prioritise intelligibility over resembling a native speaker. Our checklist is designed for this essay; it is not a CEFR level or recognised qualification. Assessment guidance · Pronunciation guidance
A Duolingo speaking study tested 258 eligible learners who had completed the beginner section of its Spanish or French course, using Pearson's Versant assessment. Some 66% of the Spanish group and 53% of the French group reached elementary (A2) speaking proficiency or higher. That tells us something about selected completers. It does not tell a new Indonesian English learner their chance of reaching the same point. Study description
That is also a useful buying rule: ask the seller to describe a successful customer without using the words “engaged”, “active” or “completed”. If the offer is meant to change a capability, the answer should describe that capability outside the product.
4. Better results can leave less money in the business
Neverlater’s invented cohorts retain all 100 original participants. Standard support has 40 documented successes, 30 assessed below the threshold and 30 missing outcomes; extra coaching has 60, 20 and 20. Allowing missing outcomes to be either success or failure produces overlapping ranges of 40–70% and 60–80%, not statistical confidence intervals.
Illustrative operating scenario: all programme prices, costs, participants and outcomes in this section and the next are invented to test a business decision. They are unrelated to any named provider's economics and do not establish that extra coaching causes improvement.
Take 100 adults paying Rp600,000 each for the programme of eight weeks for work updates. Everyone starts below a predefined task threshold. Ten groups of ten receive eight classes lasting one hour: 80 instructor hours. At Rp200,000 per hour, that costs Rp16 million. Allow another Rp4 million for preparation and assessment, Rp12 million for customer acquisition, and Rp8 million for payments, platform, support and expected refunds.
At the end, 70 people take a fresh assessment and 40 meet the threshold. Another version adds 40 coaching hours, costing Rp8 million. Suppose 80 people are assessed and 60 meet the threshold. The two views of progress look like this:
Result after eight weeks
Standard support
Extra coaching
Original participants
100
100
Final assessments available
70
80
Meet the task threshold
40
60
Success among those assessed
57.1%
75%
Documented success across the original cohort
40%
60%
Missing outcomes limit the comparison. Whole cohort success could range from 40–70% for standard support and 60–80% for extra coaching. The lower end assumes none of the unassessed participants met the threshold; the upper end assumes all did. The difference could run from ten percentage points against coaching to forty in its favour. These are logical bounds, not confidence intervals. More complete assessment alone can improve the headline without improving learning.
Now put the same two cohorts into the accounts:
Cohort economics
Standard support
Extra coaching
Receipts before expected refunds
Rp60 million
Rp60 million
Teaching, preparation and assessment
Rp20 million
Rp28 million
Customer acquisition
Rp12 million
Rp12 million
Other costs, including refund allowance
Rp8 million
Rp8 million
Contribution before fixed overhead
Rp20 million
Rp12 million
Listed costs per documented success
Rp1 million
Rp800,000
In this invented comparison, extra coaching lowers the listed cost per documented success by 20%, from Rp1 million to Rp800,000. It also cuts the operator's contribution by 40%, from Rp20 million to Rp12 million. Dividing the extra Rp8 million by 20 more documented successes gives Rp400,000 each. That is a cost per additional documented result, not an estimated cost per learning gain caused by coaching.
The figures exclude fixed overhead, financing and income tax; receipts exclude collected sales taxes. Refunds appear once, in other costs. Contribution must still pay the rest of the business.
To test learning gains, randomly assign eligible learners where practical, keep the original groups, and use the same assessment and reminder process. Fix the task criteria in advance. Where feasible, keep assessors unaware of the support received and have a second assessor check a sample. A later task can test whether the skill lasts.
5. The next teaching hour is the real product decision
Small group teaching still requires a tutor's time and attention. An imagined learning operation, not a named provider's class or cost structure.
The 40 extra hours above amount to only 24 minutes per initial participant if shared equally. That makes the operating question concrete. Is the best use one short individual review for everyone, or more help for people who encounter a particular difficulty?
We would test targeted help before adding an expensive universal benefit. A learner who understands the task but needs repetition might practise independently. Another may need a tutor to diagnose a recurring error. A third may need a different class time. Giving all three the same additional lesson would be administratively simple, but it may spend the budget on the wrong constraint.
Targeting must not become selecting the easiest successes. Publish admission and support rules before the cohort starts, and include everyone admitted when reporting results.
There is a price experiment too. With 100 buyers, an extra Rp80,000 each would fund the Rp8 million of coaching before any fees that vary with price. That produces a Rp680,000 offer. But suppose only 90 people buy, the teaching and assessment timetable still costs Rp28 million, and acquisition plus other costs scale at Rp200,000 per buyer. Contribution would be Rp15.2 million: Rp61.2 million of receipts less Rp46 million of costs. The higher price still fails to restore the original Rp20 million.
A price increase that covers added cost on a spreadsheet can fail once demand moves. An employer sponsor could fund coaching relevant to work, but introduces another buyer, a sales process and expectations about reporting. Smaller groups could improve speaking time, but raise teaching cost. Peer practice could absorb repetition, but needs a structure that keeps mistakes from circulating unchecked. Each route changes the product; none is a free source of margin.
Test one funding mechanism alongside one support change. The question is whether buyers value the improvement enough to pay for its delivery. Count referrals and later purchases only after they are observed.
6. Graduation can be a good reason to lose a subscriber
A fictional professional uses a practised skill at work. This editorial illustration does not claim that a particular course produced the result.
Some self improvement businesses should keep customers for years. A language hobby can remain enjoyable; a gym can remain the place someone chooses to exercise. Others solve a finite problem. The professional who can now give the update may no longer need the original programme.
A retention dashboard that groups that person with someone who quit in frustration hides an important difference. We would separate achieved the goal, continued independently, stopped because circumstances changed, and left dissatisfied. Those exits call for different responses. A graduate might welcome a later offer for a genuinely new challenge. A dissatisfied customer needs a better explanation than another renewal discount.
This is where we think the business can become more defensible. If customers trust the company to tell them when they have enough, its next recommendation may carry more weight. That is a thesis to test through referrals and return purchases. Trust has commercial value only if customers act on it.
The strongest counterargument is that consumers also buy identity, encouragement and company. An enjoyable class can be worth paying for even when progress is difficult to quantify. We would happily build around that value if it is what customers knowingly choose. The burden rises when a brand charges a premium for a specific transformation.
For buyers, a useful comparison starts with the next thing they need to do. Ask what practice and feedback the price buys, how it fits their week, and what a successful stopping point looks like. A larger content library may be irrelevant to the obstacle. So may a more expensive coach.
Self improvement is already big enough to deserve serious consumer business attention. Its place among the very biggest markets remains unproven. For Indonesia, our bet is on a company that becomes useful at a moment when someone is ready to change, and earns recommendations after the original job is done. The best outcome is a customer who can explain what they gained without needing the sales page to tell them.
Reference Bank
Original company and product sources checked 11 October 2026. Council of Europe guidance, the additional 2025 company comparison and the May 2026 combination disclosure were checked 12 October 2026. Company disclosures establish reported commercial performance; product pages describe advertised services. The programme is an illustrative model. No private customer cohorts, interviews or clinical outcomes are implied.
Duolingo, 2025 annual report. Reported revenue growth, subscription revenue and subscriber definition. Rounded US dollar figures; growth rates as reported. Companywide revenue includes activities beyond subscriptions.
Udemy, fourth quarter and full year 2025 results, 5 February 2026. Consumer segment revenue and growth, plus management’s explanation of its strategic shift. The decline is not presented as a measured decline in all demand for self improvement.
Coursera, completion of combination with Udemy, 11 May 2026. The comparison uses their separate historical 2025 reporting scopes, not two independent companies as of publication. No revenues, overlapping categories or user counts are summed into a market estimate.
Duolingo, Q2 2026 shareholder letter, 5 August 2026. User and subscriber metrics, definitions and management’s stated connection between learning, engagement and retention.
PwC, Voice of the Consumer 2026 Indonesia, August 2026. Methodology and printed pages 7, 16 and 25. Fieldwork through an external panel in February and March; questions have different meanings and are not combined into a funnel.
BPS, May 2026 employment and wage release, published 5 August 2026. National employee average, used only as a reference for the scale of income. The comparison of the programme price with the wage is author arithmetic.
Cakap, official English course page. Advertised features and starting price checked on the public page. Package eligibility, checkout total and outcomes were not verified.
FIT HUB, membership FAQ. Used for the distinction between membership access and personal training packages, not for outcomes or realised economics.
Duolingo, description of the speaking study, 2021. Company research summary about selected Spanish and French course completers; assessment instrument supplied by another company. It is not a current study of English courses in Indonesia.
Model method: both versions start with 100 participants below the task threshold. Missing assessments stay visible. The example testing sensitivity to price holds the Rp28 million teaching and assessment commitment fixed while acquisition and other costs vary per buyer. The bounds for missing outcomes allow every unassessed participant to have met or not met the task threshold; they do not account for sampling uncertainty or establish causality. Support effectiveness, price response, referral income and later purchases are not estimated from observed data.
The photographic scenes depict fictional situations and do not document classes or results from any named provider. The growth chart compares selected company reporting scopes; it is not a market size estimate or learning measure. The Cakap diagram summarises its advertised product design. The outcome diagram illustrates Neverlater’s invented programme.
What assumptions or evidence should I examine before acting on this article?
A human response, usually within 1-2 business days, not an instant AI answer. Questions remain editable until you press Send. You will be asked to sign in if needed.
No approved comments yet. Be the first to add a thoughtful note.
Discussion
Loading discussion...