Why Employees Forget Most of Their Training (and How to Fix It)

Employees forget training because nobody asks them to recall it. The learning science behind workplace forgetting, and the fixes that make knowledge stick.

Employees forget most of their training for one boring reason: after the session ends, nobody ever asks them to recall it. Memory is built by retrieval, not by exposure. A course they watched once, a slide deck they skimmed, an onboarding week packed into five days. The brain treats anything it is never asked to bring back as safe to drop, so it drops it.

TL;DR. Forgetting after training is normal, not a sign of bad employees or bad content. People forget because most training is built for the moment of delivery instead of for later recall: it is passive, crammed into one sitting, and never tested again. Three findings from cognitive psychology reverse this, and all three are decades old and well replicated: retrieval practice (testing yourself beats rereading), spacing (the same study time spread out beats one block), and desirable difficulties (a little effort at recall is what makes memory durable). The practical takeaway for anyone who builds training: reinforcement is part of the design, not an optional add-on.

I build a learning tool for a living, so I read a lot of L&D content. Almost all of it opens with the same line: "employees forget 90% of training within a week." It is a great stat. It is also mostly wrong. Let me start there, because the real story is more useful than the myth.

The "employees forget 90% of training" stat is mostly wrong

You have seen the claim, usually drawn as a steep curve: people forget 70% of new information within a day and 90% within a week. It gets attributed to Hermann Ebbinghaus and his 1885 forgetting curve, and it gets pasted into pitch decks as if it measured corporate training.

It did not. Ebbinghaus ran his experiments on a single subject, himself, between 1880 and 1885. His "material" was lists of nonsense syllables like WID and ZOF, chosen precisely because they were meaningless and carried no prior associations. He timed how long it took to relearn them after various delays and published the results in his monograph Über das Gedächtnis (a 2015 replication walks through his method in detail). The famous smooth curve everyone reproduces? He did not even draw it that way; the now-iconic chart came later.

Here is the part the cliché leaves out. When Ebbinghaus tested himself on meaningful material instead of gibberish, learning a stanza of Byron's Don Juan, he found it took roughly a tenth of the effort of the nonsense syllables. Meaning changes everything. Your onboarding content is not a list of random syllables, so a forgetting rate measured on random syllables tells you very little about it.

So drop the fake precision. The honest version is simpler and harder to argue with: without reinforcement, people forget a large and unpredictable share of what they were taught, and they forget it fast. You do not need a bogus percentage to act on that. You need to understand the mechanism.

So why does training actually not stick?

Forgetting is not random decay. It tracks how the material was learned. Three design choices, baked into most corporate training, are doing the damage.

Nobody ever asks them to recall it

This is the big one. Memory is strengthened by the act of pulling information back out, not by putting it in again. When a new hire watches a compliance video and clicks "next," information went in once and was never retrieved. The brain has no reason to keep it. There is no signal that this matters for later, because later never asks.

Most training is built entirely around input. Watch, read, attend, complete. The retrieval side, the part that actually lays down durable memory, is missing.

Everything happens at once

Onboarding is the clearest offender. Five days, a firehose of systems, policies, names, and tools, and then nothing. Cramming a topic into a single block is called massed practice, and it produces the worst long-term retention of any common study pattern. It feels efficient because performance during the session looks great. That in-the-moment fluency is exactly what fools us, and it collapses within days.

Passive consumption feels like learning

Reading a slide, watching a polished video, nodding along in a workshop. All of it produces a strong sense of "I know this," because the material is right in front of you and feels familiar. Familiarity is not the same as the ability to recall it cold next month. This gap between how well we think we learned and how much we can actually retrieve is one of the most reliable findings in the field, and it is why "the training was great" and "nobody remembers the training" are both true at the same time.

The cost of all this is not abstract. Companies spent more than $400 billion on training worldwide in 2024 (The Business Research Company), and Korn Ferry projects the global skills gap will cost $8.5 trillion in unrealized annual revenue by 2030 (Korn Ferry, "The Talent Crunch"). A lot of that spend is buying content that gets delivered once and forgotten on schedule.

The learning science that actually fixes forgetting

The good news is that the fixes are not speculative. They come from some of the most replicated work in cognitive psychology, and they all point the same direction: make people work to retrieve, space that work out, and keep it just hard enough.

Retrieval practice: testing is studying

The single most useful finding for anyone who builds training: being tested on material does more than measure learning. It causes it.

In a landmark 2006 study, Henry Roediger and Jeffrey Karpicke had students study prose passages, then either reread them or take a short recall test with no feedback (Psychological Science, 2006). On a test five minutes later, the rereaders did slightly better, about 81% recall versus 75%. That is the trap: in the moment, rereading looks like the winner. A week later it flipped hard. The students who had been tested recalled about 56% of the material; the rereaders, who had felt so much more fluent, recalled only about 42%.

Read that again. The group that did the thing that felt less productive remembered substantially more, a week out. A single low-stakes recall test beat rereading the whole passage. This is the testing effect, and it is the closest thing L&D has to a free lunch. Quizzes, recall prompts, "explain it back" checks, flashcards. They are not assessment bolted on at the end. They are the learning.

Spacing: the same hours, spread out

Take the exact same amount of study time and spread it across days instead of stacking it into one session, and retention goes up. No extra hours, just better placement.

This is the spacing effect, and the evidence behind it is enormous. A 2006 synthesis by Nicholas Cepeda and colleagues reviewed 317 experiments drawn from 184 articles and found that distributed practice reliably beat massed practice for long-term recall (Psychological Bulletin, 2006). The effect holds across ages, materials, and timescales. The rule of thumb that falls out of it: the longer you need to remember something, the wider you should space the reviews.

For training, this is a direct indictment of the one-and-done model. A great onboarding week with nothing after it is the textbook setup for forgetting. The same content, revisited briefly at day three, day ten, and day thirty, lands far better, and it costs no additional content to produce.

Desirable difficulties: easy learning is forgettable learning

Robert Bjork's research ties the first two together with a counterintuitive idea: the conditions that make learning feel easy and fast often make it fade quickly, while conditions that introduce some struggle make it stick (UCLA Bjork Learning and Forgetting Lab). He calls these "desirable difficulties." Retrieval is hard work, which is the point. Spacing forces you to reconstruct something you have started to forget, which is also the point. Bjork's distinction between learning and performance is the whole game for L&D: performance is how well someone does during the session, learning is what survives to next month, and the two regularly move in opposite directions.

This is why "engagement" and "completion rate" can be misleading. A frictionless, satisfying course with a 95% completion rate can still teach almost nothing durable. A slightly harder one that makes people retrieve and struggle a little will lose some comfort and win on retention.

What this means for how you design training

Put the three findings together and the conclusion is unavoidable: reinforcement is not a nice-to-have you add if there is budget. It is a design requirement. If a training program ends at the moment of delivery, it is engineered to be forgotten, no matter how good the content is.

Concretely, that means building these in from the start:

• Build in retrieval. Every module needs a step where the learner produces the answer instead of reviewing it: a low-stakes quiz, a free-recall prompt, flashcards, a teach-it-back moment.
• A schedule, not an event. Plan the reviews. Touch the key material again days and weeks after the first exposure. People in L&D have started calling this "everboarding," which is just spacing applied to onboarding.
• A little friction on purpose. Resist the urge to make everything maximally smooth. The mild effort of recalling something half-forgotten is what builds the memory.

The honest catch is that doing this by hand is a lot of work. Writing good retrieval questions for every topic, then scheduling spaced reviews per person, is exactly the kind of labor that gets cut first. Which is the problem Uncoursed was built to remove. You give it your real material, an onboarding doc, a product manual, an SOP, a book, and it turns that source into a structured course with quizzes, an exam, and spaced-repetition flashcards generated from your content, plus a learning buddy that answers only from your sources instead of making things up. The learning science stops being a project the L&D team has to staff and becomes the default shape of the course.

If you want to go deeper on the mechanics, I wrote a companion piece on how active recall and spaced repetition work for individual learners, and a practical guide to applying this across an organization with AI.

The reframe I would leave you with: stop asking "was the training good?" and start asking "what can people still do a month later?" Those are different questions, and only the second one is worth spending $400 billion on.

FAQ

How much training do employees actually forget?

There is no honest single number, despite the popular "90% in a week" claim. That figure is a misreading of Ebbinghaus's 1885 experiments, which used meaningless nonsense syllables tested on one person, not workplace training. What is well established is the direction: without retrieval and spaced reinforcement, people forget a large share of new material quickly, and meaningful, well-reinforced content is retained far better.

What is the most effective way to improve training retention?

Retrieval practice. Decades of research, anchored by Roediger and Karpicke's 2006 study, show that testing yourself on material produces much stronger long-term retention than rereading it, even though rereading feels more productive in the moment. Add spacing (reviewing across days and weeks rather than in one block) and you get most of the available gain without adding content.

Why does onboarding in particular fail to stick?

Onboarding usually concentrates a huge amount of information into a few days and then stops. That is massed practice with no reinforcement, the exact pattern shown to produce the weakest long-term recall. Spreading key material across the first weeks and months, with quick recall checks along the way, retains far more than a single intense week.

Isn't more engaging content the answer?

Engagement and retention are not the same thing. Robert Bjork's work shows that conditions which make learning feel smooth and easy often make it fade fast, while a bit of "desirable difficulty," such as having to recall something, is what makes it durable. A highly polished course with no retrieval can have great completion rates and still teach little that lasts.

Read next

New posts, in your inbox.

Memory science, product field notes, and occasional opinions about the state of online learning.