In 1984, educational psychologist Benjamin Bloom published a finding so striking that it has shaped education research for four decades — and so impractical that most schools simply ignored it.

Bloom discovered that students who received one-to-one tutoring using mastery learning techniques outperformed classroom-taught students by two full standard deviations. In practical terms, this meant that the average tutored student performed better than 98% of students in a conventional classroom. Ninety percent of tutored students reached the level of mastery that only the top 20% of classroom students achieved.

He called this the 2 Sigma Problem — not because the result was a problem, but because the question it raised was one that education had never answered: if we know tutoring is this effective, why can't every student have it? The answer has always been the same. Cost. A classroom teacher instructing thirty students is affordable. A dedicated tutor for each of those students is not.

For forty years, that remained the state of the art: extraordinary evidence, unresolvable economics. That is beginning to change.

What the Evidence Actually Shows

Bloom's two sigma claim has been debated and qualified in the decades since. A rigorous analysis published in Education Next concluded that the original two sigma figure was likely an overstatement — the conditions of Bloom's original studies were unusually favourable, and real-world implementations rarely replicate laboratory results perfectly.

But the honest version of the finding is still remarkable. A comprehensive meta-analysis by the National Bureau of Economic Research, covering decades of experimental evidence on tutoring interventions, found that tutoring consistently produces the largest learning gains of any school-based educational intervention studied. The effect is not as large as Bloom's original claim in every context. It is still, by a significant margin, the most reliably effective tool we have.

High-dosage tutoring — defined as frequent, intensive sessions, typically three or more times per week in small groups of no more than three students — has been the subject of some of the most rigorous randomised controlled trials in education research. Studies by the University of Chicago Education Lab in partnership with Saga Education found that high-dosage maths tutoring delivered during the school day could double or even triple student learning gains in a single academic year. A 2024 national study across multiple US school districts found that in-school tutoring programmes were successfully reversing pandemic-era learning loss in a way that most other interventions had failed to achieve.

These are not incremental improvements. They are among the largest effect sizes ever documented in education research for a scalable intervention.

The Pandemic Made the Gap Impossible to Ignore

COVID-19 closed schools across the world, disrupted the education of hundreds of millions of children, and produced a learning loss crisis that the existing education system has struggled to address. The students who fell furthest behind were, with grim predictability, those who were already behind — children from lower-income families, children with less access to technology, children whose schools were least equipped to maintain meaningful instruction during the closure.

By the time schools reopened, the research community had arrived at a fairly clear consensus on what the evidence said was most likely to help. Of the available interventions, tutoring had the strongest evidence base. The Biden administration's American Rescue Plan, which directed approximately $170 billion to public schools, saw over $4 billion allocated to tutoring and accelerated learning programmes.

The scale of that investment was significant. The results were encouraging. A 2024 national report from the University of Chicago Education Lab found that in-school high-dosage tutoring was successfully accelerating student learning across diverse academic settings and demographics — with results holding across districts in Illinois, Georgia, New Mexico, California, Florida, North Carolina, and South Carolina.

But the investment also exposed the central constraint that has always limited tutoring: even with unprecedented federal funding, tutoring programmes at current costs could reach only a fraction of the students who needed them. In-person, small-group tutoring typically runs between $1,800 and $4,000 per student per year. A nationwide rollout capable of reaching every child who had fallen behind would cost tens of billions of dollars annually — a figure no realistic policy scenario could sustain.

The question that the pandemic made urgent — how do you scale the most effective educational intervention in history to every child who needs it? — had no easy answer. Then AI entered the picture.

The AI Tutor and a Forty-Year-Old Problem

In 2024 and 2025, a series of controlled trials produced results that would have seemed implausible a decade ago. Researchers at Harvard University compared students in an introductory physics course who received traditional active-learning classroom instruction with students who used an AI-powered tutoring system built on GPT-4, equipped with pedagogical scaffolds designed by the course faculty and strict guardrails to reduce hallucination.

The AI tutor group learned more than twice as much as the classroom group — and did so in less time. Median test scores for AI tutor students jumped from 2.75 to 4.5 on a 6-point scale, compared to 2.75 to 3.5 for the classroom group. Students also reported feeling more comfortable asking questions and making mistakes with the AI than in a live classroom setting, where social anxiety about appearing ignorant in front of peers often suppresses exactly the kind of active questioning that drives deep learning.

Similar results have emerged from other trials. A study examining AI tutoring in UK classrooms found that well-designed AI tutors could safely and effectively support student learning. Research in Ukraine — where wartime disruption had collapsed conventional schooling for millions of children — found that small-group online tutoring programmes delivered learning gains of 0.49 standard deviations in maths and 0.40 in language, even amidst displacement, power outages, and air raid sirens.

None of this is to say that AI tutoring is a solved problem. The EdResearch for Action principles for effective tutoring are clear that not all tutoring programmes are equal — quality of design, alignment with curriculum, relationship between tutor and student, and frequency of sessions all matter enormously. AI systems that lack good pedagogical scaffolding, or that are deployed without appropriate teacher oversight, produce far weaker results than well-designed human tutoring. The technology is a tool, not a solution.

But the direction of travel is significant. For the first time, the economics of tutoring are changing. AI-assisted models can dramatically reduce per-student costs while maintaining a meaningful fraction of human tutoring's effectiveness. Hybrid models — in which AI handles practice and retrieval while human tutors focus on the relational and motivational dimensions of learning — may produce results that neither could achieve alone.

What Makes Tutoring So Much More Effective

To understand why tutoring produces such large learning gains, it helps to understand what is actually different about the experience of being tutored rather than taught in a classroom.

The most fundamental difference is responsiveness. A classroom teacher, managing twenty-five to thirty students, cannot know at any given moment exactly what each student understands and what each student is confused about. Instruction is necessarily calibrated to the average, which means it moves too fast for some students and too slowly for others. Students who fall behind rarely catch up — they carry their gaps forward, and each new concept they encounter is built on foundations that are less solid than assumed.

A tutor — human or AI — knows exactly what the student in front of them does and does not understand. Every explanation can be adjusted in real time. Every misconception can be addressed immediately. Progress is not governed by a curriculum calendar but by actual mastery.

The second difference is psychological safety. In a classroom, asking a question requires a student to reveal their confusion in front of peers. Many students — particularly adolescents, particularly those who have already experienced academic failure — will not do this. They sit through lessons they do not follow, pretend to understand, and fall further behind. With a tutor, one-to-one or in a small group, the social stakes are lower. Questions are easier to ask. Failure is less visible and less costly.

The third difference is feedback quality. Research by John Hattie, whose Visible Learning synthesis covers more than 1,200 meta-analyses, consistently finds that timely, specific, actionable feedback is one of the most powerful predictors of learning outcomes. In a classroom, meaningful individual feedback is scarce — a teacher may mark a test, but by the time the marks are returned, the class has moved on to new material. A tutor provides feedback continuously, in the moment, when it can actually change what the student does next.

The Equity Dimension

The tutoring revolution is not just an educational story. It is an equity story.

Private tutoring has always been available to families who can afford it. The global private tutoring market is estimated to exceed $200 billion annually. In competitive university entrance systems in South Korea, Japan, China, and increasingly in the United Kingdom and United States, private tutoring has become a significant advantage purchased by wealthy families for their children — a shadow education system that compounds the inequalities of the formal one.

High-dosage tutoring in schools, delivered at no cost to students, represents a direct challenge to this dynamic. The University of Chicago Education Lab's research has consistently found that the largest gains from high-dosage tutoring accrue to students from lower-income backgrounds — the same students who are least likely to have access to private tutoring outside school. If that intervention can be scaled, the potential to reduce achievement gaps that have proven stubbornly resistant to other interventions is real.

The National Education Association has identified high-impact tutoring as one of the most promising tools available for addressing post-pandemic learning recovery precisely because its benefits are not uniformly distributed — they are disproportionately large for the students who need the most help.

The Road Ahead

The tutoring revolution is real, but it is also fragile. The federal funding that catalysed a significant expansion of tutoring programmes in the United States expired under pandemic relief timelines, leaving districts to sustain programmes from their regular budgets — a significantly harder ask. The research on AI tutoring, while promising, is still maturing; the failures and limitations of early AI educational tools are instructive cautionary tales about deploying technology before it is ready.

The design principles that make tutoring effective are also demanding. Sessions need to be frequent — at least three times per week. Group sizes need to be small — no more than three students per tutor. Sessions need to take place during the school day, not as an after-school add-on that students disengage from. Tutors need training and ongoing support. All of this requires institutional commitment that is difficult to sustain when budgets are under pressure and competing priorities are many.

But the evidence base has never been stronger, and the technological possibilities have never been more compelling. For forty years, Benjamin Bloom's 2 Sigma Problem sat in the education research literature as a rebuke to the status quo — proof that we knew how to help students learn dramatically better, and a reminder that we had chosen not to, because it was too expensive.

The economics are changing. The question is whether the will to act on what the evidence shows will change alongside them.

What has your experience been with tutoring — as a student, a parent, or an educator? Share your thoughts in the comments below.