The evidence

We've known the answer for 40 years. It just didn't scale.

In 1984, Benjamin Bloom showed that students taught one-to-one, to mastery, dramatically outperformed those in ordinary classrooms. The problem he posed, which he called the "2-sigma problem", was how to give every child that benefit affordably. Technology can finally answer it.

2σ
Bloom's tutored students scored about two standard deviations above the classroom average. The average tutored pupil outperformed 98% of conventionally taught peers.
2×
In a 2025 Harvard randomised trial (194 students), learning gains with a well-designed AI tutor were more than double those of an active-learning class, in less time and with more engagement.*
1:1
The mechanism is the same: immediate feedback, work pitched to the individual, and no moving on until it's understood.
Bloom, 1984

Mastery learning works

One-to-one tutoring paired with mastery learning produced one of the largest effects in education research. The barrier was never whether it worked. It was cost and scale, and that is the gap technology can now close.

Kestin et al., Harvard · 2025

But only if it's built right

The Harvard result came from an AI tutor designed with real pedagogical guardrails, one that guides rather than answers. The same study is a warning. A chatbot that simply hands over answers teaches students very little.

What this means for schools

The benefit of an AI tutor is not automatic. It depends on how the tutor is built. That is why we built Saga around mastery and safeguarding rather than raw answer generation.