The Ebbinghaus forgetting curve is one of the most replicated findings in psychological science. The core finding: without any reinforcement, humans forget approximately 50% of newly learned information within an hour, 70% within 24 hours, and up to 90% within a week. This is not a pathology — it is normal human memory function. The brain actively prunes information it assesses as low-priority based on frequency and recency of access.
The implication for training and education is stark. A one-day training workshop, no matter how well-designed, will have transferred approximately 10% of its content to durable memory within a week for the average attendee. The other 90% is gone. Organizations that invest in training without designing for retention are paying for experiences, not outcomes.
Spaced Repetition: The Most Powerful Retention Tool We Have
The antidote to the forgetting curve is spaced repetition — the practice of reviewing material at increasing intervals calibrated to each individual's forgetting rate. Ebbinghaus himself discovered that spaced review dramatically reduces forgetting and strengthens memory encoding. Subsequent research, culminating in the development of algorithms like SM-2 by Piotr Wozniak in 1987, turned spaced repetition into an engineering problem.
The core mechanism is the spacing effect: reviewing material just before you are about to forget it produces a stronger memory trace than reviewing the same material when it is still fresh. Each review session that successfully recalls a piece of information extends the interval before the next review is needed — from one day, to three days, to one week, to one month, to one year, until eventually the information is in long-term memory and requires minimal maintenance. The total time investment to achieve durable memory through spaced repetition is dramatically lower than equivalent time spent in initial learning without spaced review.
The challenge of implementing spaced repetition at scale has historically been the individualization problem. Different learners have different forgetting rates for different types of content. Optimal review scheduling requires knowing each learner's individual forgetting curve, not a population average. AI enables this individualization: modern deep knowledge tracing models maintain learner-specific forgetting rate estimates for each knowledge element and schedule review events accordingly. This is a fundamentally different approach from scheduling fixed review reminders, which produce review events that are too early for some learners (wasting time) and too late for others (after forgetting has already occurred).
Retrieval Practice: Harder Than Re-Reading, Far More Effective
The testing effect — also called retrieval practice — is perhaps the most robust and underutilized finding in learning science. The finding: attempting to retrieve information from memory strengthens memory far more effectively than re-reading or re-watching the same content. The act of retrieval, not the act of exposure, is what builds durable memory. This has been replicated in hundreds of studies across multiple age groups, knowledge domains, and retrieval formats.
The counterintuitive implication is that the most valuable study activity is not re-reading notes or watching a lecture again — it is closing the notes and attempting to recall the content without them. The difficulty and the feeling of effort that comes with retrieval is not a sign that the method is not working. It is the mechanism through which the method works. Bjork's research on "desirable difficulties" documents how learning activities that feel harder in the moment produce better long-term retention than easier activities that feel more fluent.
For learning platform design, this means that practice questions, recall exercises, and self-testing should not be positioned as optional assessment tools at the end of a module. They are the primary learning activity. Content exposure — video, text, worked examples — provides the initial knowledge representation. Retrieval practice builds the durable memory trace that enables application. Well-designed adaptive learning platforms interleave exposure and retrieval throughout the learning sequence rather than separating them into "learning" and "assessment" phases.
Interleaving: The Counterintuitive Practice Principle
Blocked practice — the conventional approach of practicing one skill type until mastered, then moving to the next — feels efficient because performance improves rapidly during the practice block. Interleaved practice — alternating between different skill types or problem categories — feels less efficient because performance is more variable. The research shows the opposite result: interleaved practice produces substantially better retention and transfer than blocked practice, despite feeling harder and producing worse immediate performance.
The mechanism appears to be related to discrimination learning: when different problem types are interleaved, the learner must identify which skill or strategy to apply rather than simply executing the same procedure repeatedly. This more effortful process produces stronger encoding and more flexible knowledge that transfers to novel contexts. Blocked practice produces brittle knowledge that works well in the exact context it was practiced but fails to generalize.
What This Means for Learning Platform Design
Learning platforms that take retention science seriously make different design choices than those that optimize for completion rates and learner satisfaction scores. They schedule review events based on individual forgetting curves rather than fixed time intervals. They integrate retrieval practice throughout content sequences rather than confining it to post-module quizzes. They interleave content from multiple skill areas rather than organizing learning into sequential, topic-isolated blocks. They measure retention at 30, 60, and 90 days post-learning rather than only at course completion.
At Learpy, our spaced repetition engine and interleaved practice scheduling are not peripheral features — they are the architectural foundation of how our platform is built. Every content decision, every scheduling decision, and every assessment design decision is informed by the question: will this produce durable memory, or just the feeling of learning? The evidence base for what produces durable memory is, at this point, unambiguous. Building platforms that ignore it is a choice.