The most difficult thing in education is not transfer, although it comes close. Nor is it trying to get rid of learning styles, although I’m still doing my very best. No, the most difficult thing in education is scaling up. Many doomed stories in education usually start with someone who looks at conventional schools and thinks, “Surely we can do this differently.” Better, more personalised, more efficient, more often than not with technology. And, importantly, a person who thinks it can be scalable.

Sometimes it works. At least for a while, before it blows up like any vehicle that gets hit in The Simpsons. I was reminded of this pattern by a fascinating new New York Times article about Guidepost Montessori and its parent company, Higher Ground Education. Well, fascinating may be the wrong word here, because the story becomes pretty grim.

Don’t get me wrong, the original idea certainly wasn’t ridiculous. Ray Girn wanted to bring Montessori education to many more children through a chain of for-profit schools. There have been some posts about the effectiveness of Montessori lately on this blog. Sometimes with promising news, sometimes with more sobering news. But Ray Girn didn’t just want to build some good schools. He wanted to grow them at the pace of a technology company, yes, scaling them up as Silicon Valley likes. He talked about doing for preschool what ride-sharing apps and Airbnb had done elsewhere.

And investors liked this kind of idea. Higher Ground raised $335 million and eventually operated around 150 Guidepost Montessori schools. And as said in my introduction, for a while, the model seemed successful. Some of the early schools appear to have been thoughtfully designed.  The New York Times quotes parents who were extremely happy with them, something that is also not uncommon to the examples that I will talk about further down this blog.

But the company also accumulated an astonishing $440 million in losses. Higher Ground filed for bankruptcy in June 2025, and around 60 schools closed. In some cases, parents discovered almost overnight that the school they relied on no longer existed.

The financial story is complicated, and I won’t try to reproduce it all here. You can read all about it in the article in the NY Times. I want to discuss something else as this is an education blog, not a business blog. What interests me most is what happened inside the schools as the organisation expanded. Girn found out the hard way running schools turns out to be rather different from adding users to an app.

You need buildings, school leaders, teachers, or rather, trained teachers and other staff (never forget about them, they make your school work). If you work with very young children, you need lots of adults because there are legal limits on how many babies and toddlers one adult can supervise. And if you promise Montessori education, ideally those adults also know something about Montessori education.

Higher Ground developed its own training programme as Montessori’s concept is not something that is under strict supervision. This allowed the organisation to recruit people without conventional Montessori training and train them- often online – while they worked. Former employees told the New York Times that finding enough time to complete that training programme could be difficult, as they were e.g. often understaffed. One former assistant teacher recalled being placed in a classroom on his first day, despite having no experience caring for young children.

Former employees described understaffing, combining groups to reduce payroll, and some genuinely disturbing incidents involving children. Girn acknowledges serious mistakes but argues that it is unfair to use isolated cases across 150 schools to suggest systemic negligence.

And, you may be surprised, but I think he has a point worth taking seriously. If you operate 150 schools, individual incidents will occur. You shouldn’t automatically generalise from them to the entire organisation. Quality control is almost impossible to achieve with such rapid growth, and it’s one of the main issues one faces in scaling up. In 2023, according to company documents, around 15 per cent of Guidepost families were leaving within 90 days. Rebecca Girn, Ray’s wife, visited struggling schools herself and described some of what she encountered as, well, “garbage”.

So, reading all this, I had a strong feeling of déjà vu.  In the Netherlands, we had the Steve Jobs Schools, also known as the iPad schools. It’s a story that even led to a paper I wrote. They started in 2013 with big ambitions around personalised learning, flexible organisation and, unsurprisingly, iPads, which were the thing at that moment. At one point, there were plans for even perhaps a hundred schools. By 2017, more than half of the schools that had adopted or seriously considered the model had reportedly abandoned it. Costs and disappointing results were mentioned as reasons. The O4NT organisation behind the model eventually went bankrupt.

In the United States, Carpe Diem became an early poster child for blended learning. Its Arizona school attracted considerable attention for combining computer-based learning with a radically different way of staffing and organisational model. There were attempts to reproduce the model elsewhere. The basic idea was again seductive: if technology allows students to work at their own level and pace, perhaps we can redesign the whole organisation around it.

Then there was AltSchool, perhaps one of the purest Silicon Valley versions of the idea. Founded by former Google executive Max Ventilla, it raised enormous amounts of venture capital to create technology-rich microschools built around, again,  personalised learning. Here, the ambition wasn’t merely to run a few exceptional or unusual schools. The technology and model were supposed to… wait for it… scale. Eventually, AltSchool stopped operating its own schools and pivoted towards selling its technology to other schools.

To be clear, these examples aren’t identical. E.g. their pedagogies differed, as did their business models. Their reasons for struggling seem to be different. But if you think that I want to turn this into a morality tale about stupid technology people finally discovering that teachers matter. No. I know from closeby how difficult it can be to get education to work more evidence-informed. Even the spectacular results of intensive early-childhood programmes such as the Perry Preschool Project proved much harder to reproduce at scale.

Still, there does seem to be a recurring mistake. Maybe you’ve noticed it already, but school is not an app. With software, scaling can mean building something once and then giving another thousand, million, or hundred million people access to essentially the same thing. Of course, even though that is more complicated than it sounds, the marginal cost of an additional user can become extraordinarily small. But schools don’t operate like that.

Open another school, and you need another building, with more teachers,  more school leaders and more relationships with parents. This all requires more organisational knowledge and the issue I already mentioned, more quality control.

The problem becomes even more interesting when the success of the original school partly depends on precisely those things that are hardest to copy. I think that, e.g., Michaela is a success in part because of its exceptional founder in combination with unusually committed teachers. But elements such as the student population, strong relationships, or a culture that emerged over time can also play a role in the success that is hard to grasp, let alone copy.

As a scientist, I do like falsification, so I need to ask myself: are there counterexamples? One that immediately came to my mind is École 42. It started in Paris in 2013 as a radically unconventional computer programming school. There are no conventional teachers or classes. Students work through projects and learn largely from each other. It can be regarded as a social-constructivist dream, and that really seems to be the case. Today, the 42 network says it has more than 50 campuses in over 30 countries. Talking about scaling up with +25k students worldwide. Whatever one thinks about the educational model or its claims about outcomes, it has undeniably managed to reproduce itself internationally for more than a decade. I do think it’s past the danger zone. And no, I don’t have any link to École 42 at all.

So why was it possible to scale 42 when other educational innovations struggled? It’s hard to tell, but perhaps in part because, paradoxically, it isn’t trying to scale everything that we normally mean by a school. Bear with me.  It has a very specific purpose: learning computer programming, primarily serving adults. It uses a highly standardised project-based curriculum. There are no conventional teachers to recruit for every new classroom, which helps. Students are selected through an intensive entry process, the famous Piscine. I do think this has got a lot to do with it. And if this rather unusual way of learning doesn’t suit you, you don’t have to attend 42.

Try applying that model to a preschool, or even any kind of real public education.  You can’t select the toddlers who are particularly well suited to your pedagogical model. You can’t have the three-year-olds run the school through peer learning. And don’t think you can replace the adults with a clever platform. Each additional group of small children requires more people to look after them.

Maybe this is one of the main reasons why 42 can genuinely operate as a network while so many attempts to disrupt conventional schooling have found scaling much harder than expected. 42 reduced the number of things that other schools need to replicate. A preschool, primary school, or secondary school – I think even a good university- can’t do that nearly as easily.

This doesn’t mean educational innovation can’t scale. Of course it can, although sometimes it is extremely difficult. Nor does it mean technology has no role in making education more efficient or effective. But perhaps we need to distinguish between scaling a product and scaling an institution. I think education is unusually dependent on people, relationships, expertise, and local context.  At least some of these elements are the thing itself. Maybe we should ask ourselves not ‘how can we scale this, but is what’s making this school work actually scalable?

P.S.: I could add many historic examples to this blog. If you want more, I strongly recommend Larry Cuban’s blog and work.