In 1984, educational psychologist Benjamin Bloom found that students tutored one-to-one scored two standard deviations above students taught the same content in a conventional classroom โ€” the average tutored student outperformed roughly 98% of the class-taught group. Nobody in education seriously disputes that one-to-one tutoring works. The open question has always been why almost nobody gets it, and whether that's about to change.

The number that should stop you

Two standard deviations is a big effect. In plain terms: take a Year 8 student who'd normally sit somewhere in the middle of their class. Give them the same material, but paced one-to-one instead of taught to 28 other people at the same time, and they now perform better than nearly everyone who took the class version.

That's not a small "nice tutor, bit more confident" bump. It's the difference between a solid pass and a genuinely strong outcome, on the same content, taught by teachers of comparable skill. Bloom didn't find that one-to-one tutors were smarter or more inspiring. He found that pacing โ€” matching the next piece of work to what the student has just mastered, not to a fixed schedule โ€” did almost all of the work.

Why your child's school can't do this

This isn't a criticism of teachers. It's arithmetic.

A UK secondary maths class runs at roughly 25-30 students to one teacher. Inside that room on any given Tuesday, some students have already mastered expanding double brackets and are bored solving (x+2)(x+3) for the fourth time. Others are still shaky on collecting like terms and are quietly falling further behind with every new topic layered on top. The teacher has to choose one pace for the room. Slow down for the strugglers and the confident students disengage. Speed up for the confident students and the strugglers lose the thread entirely.

There is no lesson plan that fixes this. It's a structural limit: one adult, running one pace, for thirty different starting points. A school timetable multiplies the problem โ€” the same trade-off repeats in every subject, every lesson, all year.

Why paid tutoring clubs often don't fix it either

Parents who feel this gap often reach for a tutoring club or a group revision class. Worth knowing before you pay for one: many run the same structural problem in miniature. A group of six to twelve students working through one shared worksheet, at one pace, set by whoever is loudest about being stuck (or whoever the tutor assumes is average), is a smaller classroom โ€” not a private tutor.

Genuine one-to-one human tutoring does fix it. It's also the reason it costs ยฃ30-ยฃ60 an hour in the UK, and why most families can afford it for an hour a week, not for daily practice on the topic their child is stuck on right now.

What actually changes when AI can do this affordably

The honest claim isn't that AI is a better teacher than a human. It's that AI is the first realistic way to make Bloom's pacing โ€” checking what a student has just mastered, then choosing the next question accordingly โ€” affordable for daily use rather than one paid hour a week.

That only works if the tool genuinely re-paces itself to the individual student, question by question. A lot of products marketed as "AI personalised learning" don't actually do this โ€” they run a fixed curriculum sequence with a chatbot wrapped around it, which reproduces the classroom problem in software instead of solving it. Step size is the specific mechanism that separates the two, and it's worth understanding before you take any "personalised" claim at face value.

What this looks like day to day

At aitutors.me, Professor Pi checks whether the last problem was solved correctly โ€” and how much hinting it took to get there โ€” before choosing what comes next. Get 5(x + 3) = 25 right on the first attempt and the next problem steps up slightly. Need the second or third hint on the ladder, and the next problem holds steady or steps back, so the gap never opens the way it does in a room of thirty.

That's Bloom's mechanism, not a marketing description of it: match the next piece of work to what the student has just shown they can do.

FAQ

What is Bloom's two sigma problem?

Benjamin Bloom's 1984 study found students tutored one-to-one, at a mastery-based pace, scored two standard deviations above students taught the same content in a conventional class โ€” the average tutored student outperformed about 98% of the class group. The "problem" is that this result has rarely been affordable to deliver at scale.

Why can't schools just do one-to-one tutoring for every child?

Cost and structure. A UK secondary teacher works with 25-30 students at once. Pacing every lesson to one child's mastery level abandons the other 29 for that lesson โ€” no amount of teacher skill or effort removes that constraint.

Does AI tutoring actually solve the two sigma problem?

AI tutoring is the first realistic attempt to make mastery-paced instruction affordable daily rather than for one paid hour a week. Whether a specific tool delivers on it depends on whether it genuinely re-paces to the individual student โ€” see step size explained.

Duke Harewood ยท founder, aitutors.me ยท Updated 11 Jul 2026.