Personalised learning is the most oversold concept in education technology. The phrase has been in EdTech vendor decks since at least 2012, usually as a description of what happens when a student gets a pre-test, is sorted into one of three content tracks, and proceeds through a linear sequence at a self-selected pace. That is not personalised learning in any meaningful sense. It's tracked instruction with a lower content-switching cost than traditional classroom delivery. The misuse of the term has created a generation of EdTech buyers who are appropriately sceptical about personalisation claims — and who sometimes reject genuinely sophisticated adaptive systems because the marketing vocabulary has been contaminated by a decade of underperforming products.
Genuine adaptive learning, in the technical sense used by cognitive science researchers and the better engineering teams in this space, is a closed-loop instructional system. The system presents a task, receives a student response, updates a probabilistic model of the student's knowledge state, and selects the next task based on that updated model. The knowledge state model is the critical component. In a well-built system, it's typically grounded in an Item Response Theory framework or a Bayesian Knowledge Tracing model, both of which allow the system to maintain a continuous posterior estimate of each student's proficiency across each concept node in the curriculum graph. The instructional strategy — what to present next, when to consolidate versus when to advance, when to branch to a prerequisite — is derived from that model, not from a fixed difficulty ladder.
The distinction between these two approaches becomes practically significant when you consider the behaviour of a student who is struggling. In a track-based system, a student who fails enough questions gets routed to the easier track, where they proceed through lower-difficulty content that may or may not address the specific misconception causing the failure. In a proper adaptive system using knowledge tracing, the system identifies the specific concept node where the student's probability of mastery falls below the mastery threshold, investigates whether the prerequisite nodes are also below threshold, and selects an intervention — re-exposure with a different item type, a scaffolded worked example, a prerequisite drill — that is targeted at the actual knowledge gap rather than the surface difficulty level. This is not a subtle distinction. It's the difference between telling a student she needs to work harder on fractions and telling her that her error pattern indicates she has a fragile understanding of part-whole relationships that is causing systematic failure on fraction-multiplication tasks.
We do need to acknowledge what adaptive systems cannot do, because the hype cycle tends to obscure this. Adaptive systems are very good at managing the practice and retrieval phases of learning, where the task space can be parameterised and student responses can be scored against a model. They are not good at the initial conceptual introduction phase, where a skilled teacher's explanation, analogy, and real-time responsiveness to student confusion is irreplaceable. They are also not good at the metacognitive and motivational dimensions of learning — helping a student understand why something matters, or helping them develop productive struggle habits when they encounter difficulty. The best adaptive systems we've seen are designed to occupy the retrieval and practice space explicitly, with clear handoffs to teacher-led instruction for the phases where adaptive automation doesn't have an advantage. Products that claim to replace the full instructional cycle with adaptive content delivery are either lying or building something pedagogically impoverished.
The Education Perfect platform, which we backed in 2023, is an example of a team that has thought carefully about this boundary. Their adaptive layer operates on the practice and vocabulary retrieval components of K-12 learning across multiple subjects, with teacher-controlled lesson structures setting the conceptual introduction. The adaptive engine tracks mastery on discrete skill nodes and adjusts practice frequency using a spaced-repetition schedule calibrated against individual student forgetting curves. That's a technically rigorous implementation of what adaptive learning can actually do — and it's useful precisely because it doesn't pretend to be something it isn't. The teachers we've spoken with who use it well treat it as a practice engine that frees up lesson time for the higher-order discussion and application work where their expertise is indispensable.