Why Singapore & Estonia's EdTech Works, but America's Doesn't?
America has spent billions on school technology. The missing ingredient isn't money; it's good old implementation and state capacity
American school districts access an average of 2,982 distinct edtech tools per year. Across the Pacific pond, Singapore has a single Student Learning Space platform with a curated list of third party apps that follows national standards. Estonia, famous for its digital government including ed tech, ranks 1st in Europe according to the 2022 PISA assessment and consistently ranks among Europe’s top performers.
That’s not a typo. Nearly three thousand tools per district. Individual teachers navigate about 42 different platforms annually. Two-thirds of purchased software licenses go unused. Specifically, 67 percent of software licenses were found to be unused, with some cases reaching as high as 90 percent.
One of the (predictably) standard explanations is teacher training. Teachers don’t know how to use what they’re given. And that’s true, but it’s the wrong frame. You can’t train anyone effectively when you’re managing three thousand tools.
The underlying problem is implementation capacity: can your institution pick something, train people on it, and stick with it? This isn’t about funding or technical expertise, though both matter. It’s about institutional follow-through. The ability to make a decision and then actually execute it across an organization, for years.
Here’s what I keep coming back to: Singapore and Estonia didn’t develop implementation capacity by building good edtech systems. They built good edtech systems because they already had it. Both countries achieved exceptional educational results long before their current platforms existed. The technology reveals their institutional competence. It didn’t create it.
This capacity can exist in American institutions. The Department of Defense runs 161 schools worldwide for military families, all under centralized civilian control. They can standardize. They can train systematically. They can sustain decisions over time. Most American states and districts, with vanishingly few exceptions, cannot.
Which raises an uncomfortable question: if the problem isn’t funding, and it isn’t training, and it isn’t technology, if it’s the basic ability to implement anything coherently, then what exactly is our theory of change?
The Three Models
Singapore’s Ministry of Education didn’t purchase software from vendors. It built the Student Learning Space from the ground up, starting in 2016, working with a domestic technology company and GovTech Singapore. The government owns the platform. It controls the roadmap, the data architecture, the integration decisions. When MOE wanted AI capabilities for lesson planning, it embedded them directly into SLS rather than buying separate tools. The platform now serves about 500,000 users with curriculum-aligned resources from primary school through pre-university.
MOE has whitelisted external tools for integration, but on its terms, not vendors’. Every teacher receives 100 hours of professional development annually, and that training can be platform-specific because there’s only one platform to master.
Estonia took a different path to the same outcome. Two private platforms, eKool and Stuudium, cover 95% of Estonian schools. The government doesn’t own them. But it controls something more important: the standards. All educational systems connect through X-Road, Estonia’s national data exchange layer (ie the quick explanation it is a set of standards, systems, and software to share data between government and approved third parties and it is not blockchain) . Every student has a secure auth account providing access to all educational resources. Private platforms must comply with interoperability requirements to participate.
This solves a problem Americans rightly worry about: vendor lock-in. If eKool raised prices dramatically or degraded service, schools could migrate to Stuudium without losing their data or retraining on entirely foreign systems. The government controls the standards layer; vendors compete within those standards.
Both countries achieved exceptional educational results long before these systems existed. Singapore ranked among the world’s best on international assessments for decades before SLS launched. Estonia has performed at or near the top of European PISA rankings since 2006, when eKool was four years old and X-Road was still being developed. The National Institute of Education has trained Singaporean teachers through rigorous, centralized programs since 1991. Estonia’s ProgeTiger program has trained over 4,000 teachers, and nearly all schools participate. These countries built effective educational technology because they already knew how to build effective anything.
Now look at the United States.
We are aware that districts accessed an average of 2,982 distinct edtech tools in 2024-25. That’s up from 895 in 2018-19. A 3x increase in eight years. The market includes over 10,000 tracked products.
The learning management system market alone fragments among Canvas (about 28% market share), Google Classroom (24%), Schoology (22%), Moodle (7% and declining), and a long tail of others. Student information systems are rarely integrated with instructional tools. The EdTech Top 40 includes products from dozens of vendors across every category, and the list changes annually as tools gain and lose favor.
Here’s where it gets grim. Glimpse K12 analyzed $2 billion in school spending and found that 67% of educational software licenses go unused. In some districts, up to 90% sit idle. The Overdeck Family Foundation estimates only 5% of edtech tools are used at the dosage required to produce measurable impact.
We have ten times more tools than we did eight years ago. Two-thirds of them collect dust. One in twenty gets used enough to matter.
Why Teacher Training Cannot Fix (System) Problems
The training explanation is seductive because it offers a tractable solution: spend more on professional development. Districts already do. According to analysis from the Research Partnership for Professional Learning at Brown University’s Annenberg Institute, US districts spent an average of $8,300 per teacher on professional development in 2022, up from $6,250 in 2001. TNTP’s widely cited 2015 “Mirage” study found three large districts spending about $18,000 per teacher annually. National estimates reach $18 billion in total. (Harvard’s Heather Hill has criticized the TNTP methodology for inflating expenditure by including salary increases from credentials, so the true figure is contested. But even lower estimates represent substantial investment.)
What do we get for this?
TNTP found that only 3 in 10 teachers improved over a two-to-three-year period after participating in professional development. Two in 10 actually got worse. The researchers “found no evidence that any particular kind or amount of professional development consistently helps teachers improve.” Teachers spent an average of 19 school days per year, nearly 10% of instructional time, in training that produced no measurable benefit.
Using LLMs (or AI) in teaching is a new field with questionable results, but how countries preparing for it paint a picture. The OECD’s TALIS 2024 survey, covering 280,000 educators across 55 systems, offers a comparison. On artificial intelligence: 76% of Singapore teachers received AI training, and 75% use AI in teaching. In France, 9% received training and 14% use AI.
The OECD found that “teachers are not using artificial intelligence independently. Rather, education systems are providing guidance on how the technology can be used productively and safely.” Singapore’s teachers have 100 hours of annual professional development. Those hours go toward mastering one platform they’ll use throughout their careers.
American teachers get training too. But a teacher who masters Canvas in one district may move to a district using Moodle or another with Google Classroom. Training hours that could build genuine expertise instead become introduction after introduction to tools that teachers will soon abandon.
A Common Sense Media survey found that 31% of educators cannot use technology because they lack training, and 63% said district communication about available edtech is “moderate or non-existent.” Even when training happens, it doesn’t stick. Cambridge University researchers documented what happened when a large district near Washington DC had to abandon a major platform during COVID-19: “Years of professional development were lost as teachers were forced to switch to new or alternative platforms within a matter of days.”
The Shadow IT Question
There’s a defense of the American approach: maybe the proliferation of tools reflects teacher initiative. Educators finding what works for their students rather than accepting top-down mandates. Thousands of tools as healthy experimentation.
There’s something to this. Anyone who’s worked in enterprise software recognizes the pattern. Shadow IT is technology adopted outside official channels, and it emerges in every organization where the official systems fail to meet real needs. Employees use Dropbox when the approved file-sharing system is clunky. Teams adopt Slack (or even Discord!) when the company mandates something worse (like post 2024 Microsoft Teams).
Good CTOs (should) pay attention to shadow IT because it’s signal, but most CTOs still treat it as noise. It tells you what users actually need. But it also tells you something else: top-down standardization doesn’t work when the standard is bad. You can mandate a tool. You can’t mandate that people use it well, or use it at all. The 67% unused license rate is what happens when procurement runs ahead of actual needs. Centralizing decisions doesn’t help if the decisions are wrong.
Teachers who install Kahoot or Blooket on their own are solving problems that centralized procurement hasn’t addressed. There’s real value in teachers having flexibility to find tools suited to their specific contexts.
But the evidence suggests the experimentation defense doesn’t hold for American edtech. Districts aren’t rejecting official tools in favor of better alternatives discovered through grassroots innovation. The proliferation comes from vendor marketing and conference demos, purchasing processes disconnected from actual classroom needs. Teachers aren’t experimenting. Procurement departments are shopping.
Shadow IT is a symptom of implementation failure. When teachers have to work around official systems, the official systems aren’t working. Singapore teachers don’t need shadow IT because SLS meets their needs. The platform evolved from teacher feedback, not despite it. Teachers should have agency. The question is whether the system channels that agency productively or dissipates it across three thousand tools.
State Consolidation: Necessary but Risky
At least 13 states have implemented statewide LMS contracts with Canvas. Several, notably Utah and Wyoming, have achieved K-12 through college continuity.
Utah is the clearest success. After a RFP process in 2010, the Utah Education Network selected Canvas for statewide programs. By 2018, 38 of 41 districts had adopted the platform. The legislature funds Canvas licensing as an ongoing appropriation, with districts paying only a one-time $2,500 implementation fee. UEN uses Canvas to provide online professional development to more than 3,500 Utah teachers annually.
Wyoming’s statewide K-20 contract reduced per-student costs from over $25 in some districts to under $4, with estimated annual savings of $250,000. When COVID-19 struck, the prior investment paid immediate dividends. A technology director reported that Canvas usage “shot up to nearly all staff members” within a week.
Other states achieved similar results through regional cooperatives rather than state contracts. Texas’s TEKS Resource System operates through a shared service agreement among all 20 Education Service Centers, with over 90% of districts participating. New York’s Boards of Cooperative Educational Services, created in 1948, operates 37 regional bodies serving nearly all districts outside the five largest cities. The Regional Information Centers within BOCES support more than 95% of public school districts and negotiate contracts collectively.
But consolidation carries risks that Estonia’s model avoids. When Utah standardizes on Canvas, it becomes dependent on Instructure. If Instructure raises prices, degrades service, or gets acquired by a private equity firm with different priorities, Utah has limited options. Years of training, content development, and workflow integration create switching costs that vendors can exploit. The same consolidation that enables coherent training also creates vendor lock-in.
Fragmentation is worse. But this shows what’s missing: interoperability standards. Utah has a statewide platform but no state-mandated data portability requirements. If Utah wanted to switch from Canvas to another LMS, migrating course content, assessment data, and integration configurations would be enormously costly. Estonia’s X-Road architecture makes such transitions feasible. American states have no equivalent.
And consolidation doesn’t guarantee adoption. New Mexico contracted with Desire2Learn as its statewide LMS. But data showed “there wasn’t a strong correlation between the state contract and what districts were actually using.” Many districts adopted Canvas instead.
This is a prime (but certainly not the only) example of shadow IT in education. You can mandate a platform. You can’t mandate that districts use it, especially when they’ve already invested in something else. Contracts without mandates or strong incentives don’t drive adoption. Districts with existing platforms and trained staff resist switching even when alternatives are free. The same dynamic that plays out with individual teachers working around clunky official systems plays out at the organizational level.
Nobody has measured whether consolidated states actually achieve better utilization or outcomes than fragmented ones. The data doesn’t exist.
And platform standardization solves only one dimension of the problem. Utah has one LMS. It doesn’t have the training program or standards as Singapore. There’s no EdTech Masterplan (among other education plans) with explicit strategic priorities and dedicated implementation divisions. Consolidation purchasing power and big tech decisions is necessary. It isn’t sufficient and if treated as the end all be all like New Mexico did, it will backfire.
The Administrator Training Void
We’ve focused on teacher training. But teachers don’t make purchasing decisions. Administrators do.
Singapore explicitly incorporates school leaders into its capacity-building framework. The Leaders in Education Programme is a six-month, full-time program at the National Institute of Education preparing vice-principals and ministry officers for principalship. Participants are removed from their positions for half a year. The program uses the 5R5M framework covering five roles of school leadership, and the Learning Partnership in Educational Technology Branch explicitly builds capacity among “teachers, middle managers and school leaders.”
Teachers get 100 hours of annual training on one platform. Principals train on the same platform. Even if Singapore decided to swap out SLS with another platform, they understand what tools that the Ministry of Education uses and the kind of capabilities they could expect. They can support implementation. They can evaluate whether it’s working. The people making decisions and the people using the tools share a common foundation.
There are virtually no mandatory technology training requirements for US principals and superintendents. The ISTE Standards for Education Leaders exist but aren’t required anywhere. The Consortium for School Networking offers a Certified Education Technology Leader certification, but only about 900 individuals have earned it since the program’s inception. There are roughly 100,000 superintendents and principals nationally.
So untrained administrators make purchasing decisions. Research on edtech procurement found that districts buy products based on vendor marketing and conference demonstrations rather than evidence. The CEO of Teaching Lab observed: “The incentives in the market are kind of messed up, so that you’re not always incentivized to do that type of work” of ensuring teachers actually want and can use products.
This closes the loop on the dysfunction. Teachers get trained on tools they’ll soon abandon. Administrators who’ve never been trained on anything make the buying decisions. Procurement runs on vendor pitches and conference buzz. Districts end up with thousands of all sort of programs, ~two-thirds of which go unused. Teachers route around the official systems that don’t meet their needs. And the cycle continues.
The DoDEA exception
The Department of Defense Education Activity operates 161 schools worldwide serving about 67,000 students. DoDEA’s 2024 NAEP scores exceeded national public school averages by 14 to 25 points across all tested subjects and grades. A 2022 GAO report confirmed that DoDEA’s fourth-grade math and reading scores were higher than 98% and 100% of states, respectively, over the past decade. The New York Times noted that DoDEA campuses “quietly achieve results most educators can only dream of.” It’s the closest thing an “American Singapore” school district as we can get.
DoDEA uses Schoology as its learning management system across all schools, particularly for its Virtual High School program. Additionally, Google Classroom is available alongside Schoology to support virtual learning environments, assignments management, and communication with classroom teachers. Many brick-and-mortar schools use Google Classroom for daily instruction.
All students have DoDEA Google accounts (@student.dodea.edu) for unified identity management, which integrate with ClassLink’s single sign-on solution to access hundreds of digital resources. The organization has a Digital and Virtual Learning Division and a strategic plan that explicitly prioritizes technology modernization and AI integration, including “modernizing its approach to artificial intelligence and technology, ensuring that students and staff can navigate and lead within a digital world.”
But the most telling comparison isn’t test scores. It’s Common Core.
DoDEA schools use Common Core standards. The same standards that generated enormous political controversy across American states. Many states adopted Common Core on paper, then struggled with rollout: inadequate teacher training, misaligned assessments, textbook delays, political backlash. Several states eventually rebranded or abandoned the standards entirely.
DoDEA faced none of these problems. It adopted Common Core, trained teachers coherently, aligned curriculum and assessment, and produced NAEP scores exceeding nearly every state in the nation.
The standards were the same. The implementation was clearly not.
The natural response to the Singapore and Estonia comparisons: why not just adopt their curriculum or license their platforms? Why not have a national curriculum (despite America’s failure with Common Core)? Because adoption isn’t implementation. American education has shown repeatedly that it can adopt reforms while failing to implement them. The machinery that translates policy into classroom practice is broken. Importing Singapore’s curriculum into a system that couldn’t implement Common Core would produce Common Core results: uneven, politicized, disappointing.
Contextual Advantages
DoDEA has advantages beyond consolidation. Military families have employed parents and relatively stable household structures. The schools are well-funded and socioeconomically integrated, even thought military families tend to be lower income and have a history of mental health issues.
But that centralized structure is the point. DoDEA possesses the capacity to standardize, and that capacity extends far beyond technology. It can set curriculum, train teachers coherently, hold schools accountable, and sustain strategic priorities over time. Technology standardization and successful Common Core implementation both flow from broader institutional competence.
DoDEA operates on American soil, employs American educators, and serves American children. The barriers preventing state systems from developing similar capacity are political and institutional, not cultural or intrinsic.
Building the capacity we lack
American states and districts lack the implementation capacity of Singapore or Estonia. They’re unlikely to develop Singapore-style comprehensive state capacity anytime soon. That would require political and institutional transformations far beyond education policy. But implementation capacity isn’t binary. It can be built incrementally, domain by domain.
Educational technology could be a training ground.
State education agencies and large districts already make technology decisions. They negotiate with vendors, write contracts, manage rollouts, coordinate training, evaluate results. They do these things poorly, for the most part. But they do them. The institutional muscles exist in atrophied form.
Building implementation capacity doesn’t require states to develop their own platforms. That would be a mistake. Software development isn’t their comparative advantage, and failed government IT projects litter the landscape. What they need is the capacity to procure well, implement coherently, and hold vendors accountable. These are different skills from building technology. They’re also more achievable.
The states with consolidated LMS contracts have taken a first step. Utah and Virginia showed that statewide coordination is possible. Texas and New York showed that regional intermediaries can aggregate demand and reduce transaction costs. These aren’t Singapore-level achievements, but they represent real institutional learning.
The next step would be building on these foundations: adding interoperability requirements to prevent vendor lock-in, developing serious administrator training programs, creating feedback loops between utilization data and procurement decisions. This isn’t a quick fix. Building institutional capacity takes years, and American political cycles create constant pressure to abandon long-term investments for short-term wins. But it’s more tractable than waiting for comprehensive state capacity to emerge from nowhere.
EdTech is also, in some ways, a forgiving domain for learning. The stakes are lower than infrastructure or healthcare. Failed implementations are costly but not catastrophic. The feedback loops are relatively fast. You can see within a year or two whether teachers are using what you bought.
My Point Being
This isn’t primarily an edtech problem.
The United States has spent decades and billions of dollars on educational technology with remarkably little to show for it. The explanation isn’t bad technology, or bad teachers, or insufficient training budgets. The explanation is that our governance structures lack the capacity to implement coherent systems.
That same implementation deficit appears across American public services. We struggle to build infrastructure, permit housing, reform policing, execute industrial policy. Fragmented authority, inadequate administrative capacity, weak feedback loops. Systems that can’t learn or adapt.
If you care about educational technology working, you have to care about implementation capacity. And if you care about implementation capacity, you have to care about governance. The tools exist. The knowledge exists. The money exists. What doesn’t exist, outside of isolated exceptions like DoDEA, is the institutional machinery to put them together.
But institutional capacity can be built. It has been built, in pockets, even within American education. The question is whether we can learn from those pockets, and from the Singapores and Estonias, to develop the competencies we lack.
EdTech may seem like a small place to start. It might be the right one for a number of districts.