Why Spaced Repetition Crushes One-Time Training (and How Omie Uses FSRS)
Imagine hiring a contractor to build a wall, paying them in full, and then watching them tear down 80% of it three weeks later. That's what happens every time you send employees to a one-day training.
Not metaphorically. Literally. The cognitive science is brutal, and the training industry has largely looked the other way for forty years.
This post is about why that happens, what the math actually looks like, and how a modern algorithm called FSRS finally gives L&D practitioners a weapon against forgetting.
Callout: The training budget isn't the problem. The timing is.
Ebbinghaus and the Forgetting Curve: The Numbers Nobody Wants to Cite
Hermann Ebbinghaus spent the 1880s memorizing thousands of nonsense syllables and tracking how fast he forgot them. The results, published in Über das Gedächtnis (1885), produced what we now call the Forgetting Curve — the exponential decay function of memory retention over time.
The key data points:
- After 20 minutes: ~58% retention
- After 1 hour: ~44% retention
- After 1 day: ~34% retention
- After 1 week: ~23% retention
- After 1 month: ~21% retention (asymptotic plateau for well-encoded material)
For training content that is presented once and not reinforced? Research by Murre and Dros (2015), who replicated Ebbinghaus at scale, found 79% of new information is forgotten within 30 days of a single exposure. That's the number. The one that should end every argument about whether one-time workshops work.
The $200 billion corporate training market is built on a delivery method that loses four-fifths of its investment by default.
Why Your Brain Works This Way
Memory consolidation requires repetition spaced over time. The hippocampus — which handles initial encoding — needs repeated retrieval signals to tell the neocortex "this information is worth keeping." Without those signals, the neural pathways weaken and the memory degrades.
This isn't a bug in human cognition. It's a feature. Your brain is constantly pruning information it doesn't use. From an evolutionary standpoint, that's efficient. For a training manager who paid €40,000 for a leadership workshop, it's devastating.
The solution isn't repetition per se — it's spaced repetition. The timing of review matters as much as the review itself. Review too soon (before forgetting begins) and you waste effort reinforcing what's already stable. Review too late (after full forgetting) and you're re-learning from scratch. The optimal window is the moment just before forgetting would occur.
This is the crux of spaced repetition, and it took about 100 years for the training industry to act on it.
From Pimsleur to SuperMemo: The Algorithm History
The first practical application of spaced repetition was Paul Pimsleur's language learning system in 1967, which used a fixed interval schedule: review at 5 seconds, 25 seconds, 2 minutes, 10 minutes, 1 hour, 5 hours, 1 day, and so on. Rigid, but far better than nothing.
Sebastian Leitner's 1972 flashcard system (the "Leitner Box") operationalized this with physical cards in five boxes — items you knew well moved to boxes reviewed less frequently, items you missed moved back to frequently-reviewed boxes.
The real computational breakthrough came in 1987 with SuperMemo, developed by Polish researcher Piotr Wozniak. His SM-2 algorithm (still widely used in Anki today) introduced the concept of an ease factor — a per-item multiplier that adjusts review intervals based on how hard you found each recall.
The SM-2 formula:
I(1) = 1 day
I(2) = 6 days
I(n) = I(n-1) × EF
EF = EF + (0.1 - (5 - q) × (0.08 + (5 - q) × 0.02))
Where q is a quality rating from 0-5 for each recall attempt. If you recall something easily, the ease factor increases and future intervals grow longer. If you struggle, the ease factor decreases and you'll see it sooner.
SM-2 was a genuine leap forward. But it had limitations: it modeled all learners with the same forgetting curve, it didn't account for the interaction between items, and it used a simplified stability model that doesn't match how human memory actually works under real-world conditions.
FSRS: Machine Learning Meets Forgetting
FSRS (Free Spaced Repetition Scheduler) is the current state of the art. Developed by Jarrett Ye and the open-source Anki community starting in 2022, it represents a fundamental redesign of the scheduling problem.
Where SM-2 uses a fixed exponential model for forgetting, FSRS uses a three-component model derived from memory research:
- Stability (S): How long it takes for retrievability to drop from 100% to 90% — a measure of how "solid" a memory is
- Difficulty (D): An intrinsic property of the item — some concepts are harder to consolidate regardless of how many times you review them
- Retrievability (R): The current probability that you can successfully recall the item right now
The review is scheduled for the moment when R drops to a threshold — typically 90% — which is the sweet spot where retrieval effort is high enough to strengthen the memory trace without the interval being so long that recall fails.
FSRS is trained on millions of real review outcomes from the Anki community, which means its forgetting curve parameters are empirically derived from actual human recall data, not laboratory nonsense-syllable experiments or theoretical models.
The key innovation in FSRS over SM-2: it learns your personal forgetting rate. Every time you review an item, the algorithm updates its estimate of your individual memory stability for that item class. Two people can review the same content and get completely different scheduling — because their forgetting curves are different.
This is what "personalized learning" should mean at the algorithm level, not just "your name in the welcome email."
FSRS vs SM-2 in Practice
In benchmark testing on the Anki dataset (200,000+ reviews), FSRS achieves 10-15% lower forgetting rate at matched review load compared to SM-2. That means the same number of practice events produces substantially better retention.
For learning science practitioners: the efficiency gain is largest for high-difficulty items and for learners with faster-than-average forgetting rates — which tends to describe content encountered in workplace contexts, where distractions are constant and emotional load is high.
The Corporate Training Gap
Here's what makes this frustrating: the technology to implement spaced repetition at scale has existed for decades. And yet the dominant delivery model for corporate training is still the one-day workshop, the two-hour onboarding module, or the annual compliance training dump.
Why? A few reasons:
The event model is easier to budget. You can invoice for a "training day." You can't easily invoice for "17 intelligently spaced micro-interventions over the next 90 days."
LMS platforms are built around completion, not retention. A completion certificate fires at the end of the course. There's no certificate for "still remembers this 30 days later."
Training is still conflated with learning. Showing up to a workshop is measurable. Actually acquiring and retaining a skill is not — or at least, organizations don't systematically measure it.
The result: an industry that optimizes for the wrong signal. Productivity and management skills courses generate completion reports. They do not generate behavioral change, because they weren't designed to.
Callout: Completion rates tell you whether people attended. Retention curves tell you whether they learned. Most organizations measure the former and assume the latter.
How Omie Implements FSRS
Omie's core architectural decision was to apply spaced repetition logic not to flashcards, but to skill-concept surfaces.
The unit of repetition in Omie is not a question-and-answer pair. It's a concept cluster — the combination of a skill domain (e.g., giving feedback) and a specific application context (e.g., giving feedback to a senior peer). Each cluster has its own stability and difficulty parameters, updated via a simplified FSRS-inspired model.
Every day, Omie's scheduling layer computes the retrievability score for each concept cluster in a user's profile. The cluster with the lowest retrievability — meaning it's closest to being forgotten — is the one that surfaces in today's 10-minute nugget.
This means the daily learning content isn't random and it isn't the same for everyone. Two users with identical job titles and identical skill gaps will get different content if their review histories diverge. One who reviewed decision-making frameworks last week won't see them again for another 12 days. One who hasn't seen them since onboarding will see them today.
This is what personalization means when it's built on cognitive science instead of marketing copy.
Why Not Just Flashcards?
Flashcard-based spaced repetition (Anki, Quizlet) works brilliantly for declarative knowledge — vocabulary, facts, definitions. It doesn't map cleanly onto skill development, which requires application, not just recall.
Omie's model spaces concept exposure and application prompts together. You don't just see "what is the SBI feedback model?" — you see a 10-minute case study that requires you to apply SBI to a realistic scenario you're likely to face this week. The spaced component is the scheduling of that application moment, not a multiple-choice quiz.
The distinction matters for communication and leadership skills because those are procedural, not declarative. You can recall the SBI model on a quiz and still deliver terrible feedback in a meeting. Application practice, spaced over time, changes behavior. Recall practice alone does not.
Building a Spaced System Without Omie
If you're an L&D practitioner who wants to implement spacing logic in your existing program without switching platforms, here's the minimum viable version:
- Identify your top 10 concepts from your last major training initiative
- Create a 5-minute retrieval exercise for each (not a re-read — a practice or application prompt)
- Schedule them at: Day 1 after training, Day 3, Day 7, Day 14, Day 30
- Track who completes each and flag anyone who misses two consecutive intervals for follow-up
- Measure retention with a one-question behavioral check at Day 45: "Have you applied X in the past 2 weeks?"
This is not FSRS. It's a fixed interval schedule — closer to Pimsleur than to Wozniak. But it will outperform a one-time training event by a measurable margin, and you can run it in a spreadsheet and a calendar invite.
The ceiling on this approach is that it doesn't personalize to individual forgetting rates. But for an HR team that has never implemented spaced reinforcement before, the floor is infinitely higher than zero.
The ROI Math
If your company spends €100,000 per year on L&D and 79% of that investment is forgotten within 30 days, you're getting €21,000 of actual learning return.
Implement a basic spaced reinforcement protocol — even a fixed-interval one — and the research suggests you can recover 40-60% of that forgetting. That's a potential €40,000-€60,000 swing on the same budget.
Implement FSRS-based personalized spacing and you're targeting 85-90% retention at the same review load, which puts you in a fundamentally different category.
The training budget isn't the constraint. The forgetting curve is. And now you have the tools to fight it.
Start by understanding which skills your team has already been exposed to and where the retention gaps are widest. Run a Learning Scan on your team to find out — it takes 10 minutes and gives you the gap map you need to design your reinforcement schedule.