What the forgetting curve actually shows
Hermann Ebbinghaus mapped how memory fades in the 1880s by testing his own recall of lists of nonsense syllables. The curve he produced showed rapid early forgetting followed by a slower decline. A modern replication confirmed the basic shape holds.
Ebbinghaus noted that meaningful material was far easier to retain. His estimate: roughly one tenth the effort of nonsense material. The curve he drew was the worst-case scenario for memory, not the baseline for professional training.
The more accurate reading is this: memory decays without reinforcement. The rate depends on how the material was encoded. Content that is abstract, passive, and disconnected from real work conditions decays faster. Content that is meaningful, contextual, and actively applied decays more slowly.
That distinction matters because it changes the diagnosis. If the problem were simply that people forget, the solution would be repetition. More modules. More reminders. More microlearning. If the problem is that training was encoded in the wrong form, repetition does not fix it. It repeats the mistake.
The conditions problem
The standard response to the forgetting curve is to repeat the training. Spaced repetition, refresher modules, microlearning reinforcement. The logic is straightforward: if memory decays, reactivate it.
The problem is that repetition treats forgetting as a quantity problem. Not enough exposure. Not enough review. Add more, and retention improves.
What the evidence points to instead is a conditions problem. Memory is not stored as a context-free record. The environment, the pressure, the type of decision required: these are all part of the memory trace. When the conditions at retrieval match the conditions at encoding, recall is stronger. When they do not match, retrieval fails even when the information was learned.
This is encoding specificity, established by Tulving and Thomson in 1973. The retrieval cue has to connect to the way the memory was stored. If it does not, the memory is not necessarily gone. It is simply inaccessible under the current conditions.
The clearest demonstration of this in practice came from a study by Godden and Baddeley with deep-sea divers. Divers learned word lists either on land or underwater. They recalled them better in whichever environment they had learned in. The environment was part of the encoding, and removing it reduced retrieval performance. That is not a memory failure. It is a conditions mismatch.
For workplace training, the implication is direct. A skill learned in a low-pressure classroom is encoded under low-pressure conditions. When that person faces a high-stakes client situation or a compliance decision under time pressure, they are retrieving under entirely different conditions. The encoding and retrieval environments do not match.
Across 100+ Finsimco simulation cohorts, we see that the real difference is not between people who “remember” and people who “forget.” It is between people who received information and people who had to use it. When participants make decisions under time pressure, with incomplete information and visible consequences, the learning is encoded as a response, not just as content. The widely repeated L&D claim that “90% of training is forgotten in 72 hours” is therefore useful, but incomplete. It describes what happens when learning is passive, abstract, and disconnected from the conditions of the job. It does not describe what happens when people are required to apply knowledge in a realistic environment, make decisions, see consequences, and receive feedback. Under those conditions, the goal is not simply to remember more. It is to build a response that can be retrieved when the job demands it.
What the research shows about encoding
What changes when training is encoded under the right conditions?
Three lines of research give a clear answer.
1. Active processing Freeman and colleagues analysed 225 studies across STEM education. Active learning outperformed traditional lectures on examinations. Failure rates were 1.5 times higher in lecture-only conditions. Active processing forces the brain to build a representation of the material. Receiving it passively does not produce the same result. That is what makes retrieval possible later.
2. Emotional engagement Tyng and colleagues found that emotional arousal during learning strongly influences memory encoding. Events with emotional weight are stored more deeply and recalled more reliably. In financial services, high stakes are not distractions from learning. The pressure of a client conversation matters. So does the consequence of a compliance failure. These are encoding conditions. Training that removes those stakes removes the signal that makes memory durable.
3. Stress and context Vogel and Schwabe found that stress impairs flexible recall. Under pressure, performance shifts toward automatic, habitual responses. A professional does not retrieve what they studied. They revert to what they practised under similar conditions. If training never created that pressure, there is no practised response to draw on.
All three findings point the same way. Encoding is shaped by what the learner does, what they feel, and what conditions surround them. Change those conditions and you change what gets stored and what can be retrieved.
The retrieval practice finding
There is a second line of evidence that sharpens the conditions argument further.
Karpicke and Roediger tested what actually improves long-term retention. They compared two groups: one that studied material repeatedly, and one that was tested on it repeatedly.
After a week, the repeated-testing group recalled significantly more. The repeated-study group showed almost no improvement in delayed recall. The extra exposure made little difference.
Retrieval practice works because the act of retrieving strengthens the memory trace. Studying again does not produce the same effect. The brain stores what it actively works to recover, not what it passively receives again.
What this means for training design is direct. Reviewing slides or replaying a module is not retrieval practice. The learner is receiving information, not reconstructing it. For retrieval to do its work, the learner has to attempt to recall, apply, or produce something. Passive re-exposure does not qualify.
This is where the conditions argument and the retrieval evidence converge. Retrieval practice improves retention. But retrieval under conditions that match real use improves both retention and transfer.
Picture a professional making a pricing decision under client pressure. The information is incomplete. The clock is running. Compare them to someone who reviewed the same content twice. Both covered the material. The encoding is not the same.
One has a retrievable memory. The other has a practised response.
Bulge bracket investment banks have around 100 analysts starting in the same fall intake in their London offices. For several graduate programmes, Finsimco simulations have been used as the consolidation layer after formal instruction. The aim is not to reteach the technical content from scratch. It is to see whether participants can use what they have just learned when the situation becomes more like the job: incomplete information, time pressure, competing priorities, and decisions that affect the next stage. What we see in the data is a consistent separation between conceptual familiarity and usable performance. Some participants can explain the concept immediately after instruction, but struggle when they must apply it inside a live scenario. Others make mistakes early, receive feedback through the simulation, and improve because the consequence makes the lesson concrete.
That is why our learning science work focuses on transfer rather than content exposure. For new hire onboarding and talent development, we try to prepare participants to apply the material once the job starts, outside the simulation environment and under conditions that feel as close to the real work as possible.
What the right encoding conditions actually look like
The learning environment needs to carry pressure. The decisions made in it need to have consequences. The material needs to be applied, not received. And the context needs to resemble the context where the skill will actually be used.
That is not a description of most training design. It is a description of what most training design leaves out.
Simulation creates those conditions directly. Not as an approximation of real work. As a structured version of it.
Participants face market moves, client demands, time constraints, and incomplete information. They make decisions. Those decisions produce outcomes. The environment carries stakes because the decisions feel real, even when they are not.
What this does to encoding is measurable.
The learner is not receiving information. They are building a response under conditions that resemble the conditions they will face. The emotional weight is present. The decision-making process is active. When a practitioner later faces a comparable situation in the real world, they are not retrieving an abstraction. They are drawing on a practised response, encoded under similar conditions. The practitioner arrives at the real situation with something usable.
Measuring transfer
A post-session test tells you whether the content was absorbed. It does not tell you whether the skill is available three months later under job conditions.
True transfer measurement requires observation after the session, not a questionnaire completed while the room is still warm.
The most practical approach is to agree on two or three specific behavioural indicators with line managers before the programme runs, then check against them at 30 and 60 days.
Not a formal assessment. A short conversation against a specific question:
Are you seeing this skill used differently on the job?
The data Finsimco generates during the simulation, where each participant excelled, where they hesitated, where the gaps are, gives managers a starting point for those conversations. It does not replace the follow-through.
A useful measurement design treats the 72-hour point as an early signal, not the final proof. A team can compare three things:
- what participants remember immediately after training;
- what they can still apply after 72 hours;
- what managers observe after 30 or 60 days when the skill is needed under real work conditions.
That comparison is more useful than quoting a generic forgetting statistic. It separates content recall from transfer. It also shows whether the training created a response that survives beyond the session.