Why Education Is The Most Difficult Business?

Most education companies find something that genuinely helps people in their learning journey, they pour their soul into it, and grow to a few thousand passionate users. But after that, it becomes really hard to scale. The graveyard is full of ventures that solved a real learning problem, then starved while trying to solve the distribution problem. Yet founders continue to be drawn to the space due to the market's vast size and the friction caused by institutions that remain resistant to change and misaligned with modern technology. But, what makes edtech so hard!?

Why Edtech is Hard?¹

The usual responses are: schools are slow, student's don't care, teachers resist change, curriculum is not aligned to impart industry skills, etc. All these are true but incomplete.

Edtech: A Thousand Islands

The biggest issue is that education is actually many markets pretending to be one. When the structure of the market is similar across geographies and requirements, a product can grow easily. For education, each region has wildly different needs: from the subjects taught to cultural expectations, from standardized test preparation strategies to ways to finance studies. As an edtech early-stage startup that has just found PMF in one niche, they are highly unlikely to scale by replicating their playbook. So, when the founders say "edtechs don't scale", what they mean is "we built for one segment, but it didn't work in another".

It is due to this hardness, the niches are still protected and the market remains fragmented. Products like Netflix, Airbnb, or Doordash capitalize on consumer habits that cross borders and cultures, which is why once they achieved PMF, they scaled relatively in an effortless manner. Education is not like that. Inter-market differences are profound, almost unshakable, and extremely granular.

To overcome this friction, startups face three options: (a) dilute the PMF to serve many kinds of people with the same product (b) forgo growth ambitions and serve best in the original market, or (c) go zero-to-one once again in the next market and keep repeating the process. Most companies end up picking either (a) or (b).

However, it's not that this problem is not solvable. In the transportation market, Uber's global presence led to a $176b market cap, despite every city requiring its own licenses, labor rules, and insurance regime. In the HR SaaS market, Rippling's most recent round valued it at $13.5b, while threading dozens of country-specific labor, tax, and data-privacy codes. So, why we haven't seen a generational company in the education space, yet? Because while it was worth it to become Uber or Rippling, the expected returns in the education were just not enough. Maybe the times are changing now as we are now realizing the importance of data in the modern world.

Who Pays vs. Who Benefits

Edtech intrinsically has a hard design problem to solve: who should we optimize the product for? Students benefit, but parents pay. Teachers benefit, but principals decide. Employees benefit, but HR budgets. B2B gets even messier: institutions pay for the product, educators administer the product, students consume the product, and parents passively evaluate the product's performance. What ends up happening is a race to the bottom on either price or quality, which means most K-12 ed-tech products either suck or don’t sell.

The trick is to find an alignment of interest, where the user is desperate and the buyer is motivated. However, such combination is very rare and already over-saturated. Example: Interview prep. Students want jobs, parents want jobs-for-their-kids, colleges want placement stats, and employers want trained candidates. Multiple payers, singular outcome.

What People Say vs. What They Do

Parents say they want holistic development of their child. Yet they pay only for test scores. Schools say they want to improve critical thinking of their students. Yet they buy whatever improves board results or reduces teacher effort. HR says they want lifelong learners. Yet they buy whatever plugs the skill gap for Q4 hiring. In Edtech, beliefs are aspirational; payments are practical. Companies which understand the latter pattern wins. Serve the pressure points, not the vision statements.

Not Everyone Can Delay Gratification

Most people hate to invest time and money in things that may or may not benefit them for months, or years, and they’ll reliably choose smaller immediate rewards over larger, delayed gains (a phenomenon known as hyperbolic discounting). It’s the famous marshmallow test. The majority of consumers will rather have one now than maybe two later. Startups have found novel ways around this cognitive bias, but all of them have shortcomings.

The first way is to bundle users together in small groups, or so-called cohort-based courses (CBCs). These lend themselves to rather successful (and often enjoyable) learning experiences, but they’re basically impossible to scale without severely diluting the product quality. The best experience is to have a few peers to keep you accountable and motivated, but not enough to forgo a personalized dedicated approach from a world-class teacher.

The second way is gamification. A classic example is Duolingo, now a tremendously successful company with an over $24b market cap. My qualm is that Duolinguo is not actually an education business: it's a gaming app. It's in the business of letting smart people play with their phones without making them feel like they're wasting their time. Learning (very much like exercising) must be hard to be effective. For a host of psychological and biological reasons, the more effort you put in, the more effective it will be.

Other companies, the most famous being Bloom Tech (once upon a time known as Lambda School), tried to solve the time-horizon issue by postponing payment. They popularized Income Sharing Agreements (ISA), a way for education providers to get paid later using a portion of the student's future income. It was a powerful idea, at the time, but it suffered from the same scaling and monetization issue as any other cohort-based course: scale and quality are at odds. To squeeze more out of it Bloom Tech resorted to aggressive financial engineering, bundling the student loans and reselling them to sophisticated investors while mischaracterizing the loan nature, ending up burning the company's brand.

A few startups have been building a platform where people learn economically valuable skills for free, and all recruiters pay to reach them. If you can't sell content, you can sell access to your customers, that you acquire cheaply before upskilling them. Recruiting and education are two sides of the same coin: it's an ingenious plan! But this too failed to work well.

One reason is that customers who need the product most, small companies and local businesses who can't fill roles, are those with the least amount of money to pay. There's also an extreme variance between small and big companies' recruiting needs, that reminds of the regional fragmentation in the education market. Recruiting businesses have the same scaling issues as education startups: they're geographically siloed and become less profitable and less marginally efficient as they grow.

Another potential angle of attack is to make companies pay for content in service of employee reskilling and retention. So far, this hasn't worked either. Large companies can leverage a higher base comp and better employer branding to attract and retain employees, which is much more effective than pure reskilling. Typically, when large companies invest in educational content for their workforce, they'd rather make their own materials and tools when they can afford to do so; if we consider only those who can't invest in bespoke solutions, the market shrinks considerably.

Edutainment Remains Customer's Top Choice

Most edtech founders I've met believe that the only effective way to easily monetize ed-tech products at scale today is to avoid the education part in the first place, and to sell entertainment instead. Aside from Duolinguo, Masterclass is another great example. Substacks, YouTube channels, Morning Brew, and even magazines like The Economist fall into this category. Edutainment is pleasant and immediately satisfying. It can work spectacularly well, as a business. But while it may effectively convey knowledge and information, its memory decay curve will be steep because of the lack of effort and hardship involved in its (passive) consumption. It doesn’t accrue to actual new skills or long-term retention. It’s not education.

In conclusion, in the education space, you'll either find companies with high PMF that never scale beyond their initial successes, or companies that can scale but can't achieve high PMF because regional players will always build better products, and better products on a smaller scale will inevitably monetize better.

Why One More Edtech Attempt Then?

While all what we said above is truth, we think things may finally be beginning to change due to a new element to the edtech formula: AI. It combines what's already powerful about information technology (nearly perfect memory), with something that has been elusive until recently: scalable genuine personalization. This potentially nullifies the scalability-vs-effectiveness trade off: we can now finally see products that do provide true long-term learning while catering to the individual needs and cultures of its users.

While this is incredibly encouraging, it still doesn't solve some of the monetization issues that are at the heart of the ed-tech problem. Companies are suffering sizable marginal costs to build compelling products, which makes generating profit even harder. There is a competition against do-it-all behemoths like ChatGPT which already is performing well in solving doubts and helping learners learn better. So, surely there are unique challenges to be solved, and perhaps these are easily solvable than going zero-to-one in every single niche in order to scale.

What Works and What Will Work In Edtech

Following are some of the initial thoughts about what works and what will continue to work in the edtech space.

Credentialing System

Universities have maintained their monopoly on credentialing through an accident of history, perhaps not merit. Yet they work because historically they were the only basis on which people were hired. As a result, test prep to get into premier institutes for better credentials survived every hype cycle.

A computer science degree signals something, but this signal became increasingly noisy. Employers started to look for alternative credentials like certifications. As a result, upskilling market boomed. But with too many certificate providers in the market, we have hit a legitimacy crisis of these certifications.

Being yet another certificate provider wouldn't cut. The question is: What's next? Can we create systems where learning and proving are almost inseparable?. The reason GitHub profiles matter for developers isn't that they're credentials, but that they're artifacts. You learned by doing, and the proof is the thing itself. This generalizes. For any skilled profession, you can ask: what would it mean to make learning produce artifacts that are themselves valuable? A medical student learning diagnosis could generate a corpus of case analyses. An aspiring designer's portfolio grows as they learn. A future lawyer learning contract law could analyze real documents and build a database of patterns. Or, can we build the credential directly for the employers?

The startups that win here won't look like education companies. They'll look like infrastructure for professional communities.

Redefining The Unit of Learning

We inherited the 50-minute lecture from an era when books were expensive and professors were information gatekeepers. Then we kept it because institutions are made of concrete and tenure. But the natural unit of learning isn't time-based, it's completion-based, and it varies wildly by person and subject.

Companies understood this and started offering self-paced learning, but they're still thinking in courses. The real insight is that knowledge forms a graph, not a sequence. You can learn multiple things in parallel, and perhaps need not finish the undergraduate work completely before touching the graduate-level material. We need a way where people can navigate the knowledge graph by themselves.

Not adaptive learning that adjusts pace: that's just a marginally better horse. We need systems that dynamically generate learning paths based on what you want to do and what you already know. When you want to build a web app, you don't need to take a course on JavaScript. You need these seventeen specific concepts, which can be learned in this order, and will take you four hours.

The business model might be beyond subscriptions, maybe some marketplace dynamics: connecting people who want to learn specific things with the exact atoms of knowledge they need, compensating the creators of those atoms (just a thought as of now). YouTube accidentally built a piece of this infrastructure, but they're optimizing for watch time, not learning outcomes.

Unbundling The Degree

Universities bundle dozens of functions: credentialing, learning, networking, life structure, status signaling, dating markets, research, adult supervision, and more. For historical reasons, you can't buy these separately. Imagine trying to sell cars the way we sell education. "Yes, for $200k you get a vehicle, but also four years of our parking lot, a social club, use of our roads, and a racing team membership. No, you can't just buy the car."

The unbundling has begun. Bootcamps and edtechs took over job training; online courses took information delivery and knowledge transfer; LinkedIn badges, Codeforces rating, Coursera certificates took outcome-based credentialing, but it's not saturated. Most opportunities remain.

The especially interesting ones involve things universities do accidentally, as byproducts. The network effects of attending Stanford aren't a designed product; they're an externality of geographic concentration plus selection effects. But they're possibly more valuable than the education. Could you build that network intentionally?

Build For Agency

The deadliest pattern in edtech is replicating the worst part of school: the passive broadcast model. Video lectures and digital multiple-choice tests are just a shinier, more lonely classroom. Also, we should not pretend that motivation is solved by giving better explanation of concepts. It's solved by social pressure, by integrating learning into activities people already do, or by making the output of learning immediately valuable. Can you build such environment? Can you get your learners immediate feedback?

AI today enables us to create an infinitely patient, always-available Socratic partner. A simulator that teaches you stuff via conversations and debates is amazing. The AI that can spot the flaw in your reasoning as you code, write, or calculate, and ask the perfect question to guide you - not give you the answer - is the holy grail.

The startups that succeed will be criticized for "not focusing on learning outcomes." One can ignore this. If people don't use your product, learning outcomes are zero.

Rethink What's Worth Learning

The real impact of LLMs on education isn't chatbot tutors. Those are, once again, just faster horses. The impact is that AI changes what's worth learning.

When GitHub Copilot can write boilerplate code, the value of knowing syntax decreases; the value of knowing what to build and how to build increases. When AI can explain any concept, the value of memorizing explanations decreases; the value of knowing which questions to ask increases. When AI can solve well-defined problems, the value of problem-solving skills decreases; the value of problem-finding and problem-framing skills increases.

This is disorienting for educators, who are optimized for teaching the old skills. The opportunity is building systems that teach what's now valuable: taste, judgment, strategic thinking, problem identification, knowing what's possible, navigating ambiguity. Can you build such systems?

These are hard to teach because they're hard to formalize. But they're also hard to test, which means credentials for them are even noisier than traditional credentials. The startup that figures out how to teach judgment, and prove someone has it, will capture enormous value.

Balancing the Redesigning of B2B Tools

Institutions move slowly, buy slowly, and optimize for everything except learning outcomes. One's life might get consumed by procurement processes, pilot programs, and serving the needs of administrators rather than learners. However, this trap needs to be balanced as that is the place where a huge amount of money lies. Founders can win serving institutions, but it requires a stomach for enterprise sales and a willingness to optimize for what schools actually want (admissions, standardized testing, admin efficiency, academic credibility, placement, industry partnerships, NAAC/NBA reporting, student satisfaction, alumni engagement, etc.)

Colleges have budgets, pain, urgency, and a forced need to differentiate. In a world where degrees are commoditizing, colleges must upgrade the experience to avoid being replaced. A startup that meaningfully alters outcomes in colleges: learning, employment, reputation, and operations, can scale faster than K12 or B2C market. So, do not underestimate this market but be cautious at the same time.

What's next?

We know that the system can be improved by huge margins. That's why the opportunity is very big. However, many smart people have tried to fix education. Most failed. And somehow the system continues to work. The system in which the learner spends 15,000 hours in K12 and emerges unable to learn on their own; or spends $200k on education combined and yet the value gained remains pretty unclear. The ones who succeeded seemingly did so by ignoring education entirely and building something else that happened to teach (hopefully). Maybe this suggests: If you want to build an edtech company, don't build an education company. Build something people want that makes them better at something valuable. If you do it right, learning happens along the way.