Everyone measures the day. Nobody measures day 90.
Training is almost always evaluated by how the room felt at 5pm. The number that matters is what a person can still do a quarter later, and it is decided less by the content than by what happens in the weeks afterwards. Build a programme below and watch what survives.
An illustrative model, not a measurement of your programme. Ebbinghaus ran his experiments on himself using nonsense syllables, so the famous percentages do not transfer directly to meaningful, job-relevant material — but the shape of the curve, and the effect of spacing and testing on it, replicate consistently.
Four problems training will not fix
Training only solves one thing: people who would do it correctly if they knew how. Everything else is a system problem wearing a course as a disguise, and the course will be delivered, rated highly, and change nothing.
Nobody said what "good" is
People cannot hit a standard that has never been written down. Publishing the expectation costs nothing and frequently resolves the whole problem before a trainer is booked.
The tool makes it hard
If the correct action takes eleven clicks and the wrong one takes two, training people to prefer eleven clicks is a losing strategy. Fix the path, not the person.
The incentives point elsewhere
When quality is taught and speed is measured, speed wins every time. People are behaving rationally inside the system you built, and a workshop will not outrank a bonus.
They know, and cannot
No time, no authority, no staffing. This is the most common cause of all, and the most commonly answered with a course — which then adds another half-day to a workload that was already the problem.
"Could they do it if their life depended on it?"
If the honest answer is yes, it is not a training problem. Training closes the gap between not knowing and knowing. It has no effect whatsoever on the gap between knowing and doing — and most workplace performance gaps are the second kind.
When training genuinely is the answer
New hires, new systems, genuinely novel skills, and anything where the consequence of an error is severe and rare — surgery, aviation, high-voltage work, firefighting. In those cases the training must be practised, not attended.
Eight kinds of training, and what each is really for
Five things that make learning stay
All five share an uncomfortable property: they make the training feel harder and score worse on the feedback form, while producing far more of what you actually paid for.
Test, do not review
Pulling an answer out of your own head strengthens the memory far more than reading it again. Re-reading feels productive and is nearly worthless; a short quiz nobody grades is one of the highest-return interventions available.
Four hours across four weeks beats four hours in a day
Same content, same total time, dramatically different outcome. Every review after a partial forget resets the curve higher and flattens the next decline — that is the sawtooth in the chart above.
Mix the problem types
Practising A A A then B B B produces confident performance in the room and poor performance later. Mixing them forces the learner to identify which method applies — which is the actual job.
Specific, immediate, on the action
"Good job" teaches nothing. "You skipped the confirmation step, do it again from the top" teaches everything. Feedback delayed by a week is barely feedback at all.
Struggle is the point
Practice that feels smooth is usually below the edge of ability, and that is where learning stops. It should be uncomfortable, with errors made and corrected in the room rather than in production.
Repetition is not practice
Doing something ten thousand times makes you fast at your current level. Deliberate practice means a specific sub-skill, chosen because you are bad at it, worked just past the edge of your ability, with immediate correction, repeatedly. It is exhausting, which is why almost nobody sustains more than a few hours a day of it.
Simulation, and the fidelity trap
Aviation, medicine, firefighting and the military all train for events too dangerous or too infrequent to learn on the job. The instinct is to make the simulator as realistic as possible. That instinct is often wrong.
Realism is not the objective
A full-motion simulator costing millions is worse than a paper walkthrough if the skill being trained is a decision sequence. Match the fidelity to the skill — physical fidelity for physical skills, situational fidelity for judgement.
The debrief is the training
In aviation and medical simulation the accepted view is that the scenario exists to generate material for the debrief. Run a short scenario and spend twice as long discussing it, or you have staged theatre.
Fail on purpose
Scenarios should be designed so people make the error in a room where it costs nothing. A drill everyone passes has confirmed nothing except that it was too easy.
Rare skills decay fastest
Anything used once a year — evacuation, resuscitation, emergency shutdown — is gone within months without refreshers. This is exactly why regulated industries mandate repetition on a calendar rather than trusting competence once demonstrated.
Tabletop exercises
A room, a scenario read aloud, and the team talking through what they would do. Almost free, and it reliably surfaces the same three findings a full drill does: nobody knows who decides, the contact list is out of date, and two teams assumed the other was doing it.
Practise in the real context where you can
Skills are more tied to their context than anyone expects. Training done at the workstation, with real tools and real interruptions, transfers considerably better than the same content in a training room.
Four levels, and almost everyone stops at the first
Kirkpatrick's model has been the standard framework since the 1950s. Its main finding in practice is how rarely anyone climbs past level one.
| Level | Question | How it is measured | Reality |
|---|---|---|---|
| 1 — Reaction | Did they enjoy it? | The feedback form at the end | Almost universal, and barely correlated with whether anyone learned anything. |
| 2 — Learning | Do they now know or can they now do it? | A test or observed demonstration, before and after | Common in regulated fields, rare elsewhere. Cheap to add and the single biggest upgrade available. |
| 3 — Behaviour | Are they doing it differently at work? | Observation or manager review at 60–90 days | Uncommon, because it requires someone to look. This is the level the whole thing was for. |
| 4 — Results | Did the business number move? | Incidents, errors, quality, retention, output | Rare and genuinely hard to attribute — too many other things changed in the same quarter. |
Pick the measure before you book anything
Decide during the design phase what you will look at on day 90, who will look, and what number would mean it worked. Chosen afterwards, the measure is always the one that happens to look best.
One question, ninety days later
Ask the manager: "What is this person doing differently since the course?" A specific answer means it worked; a vague one is your result too, and it cost nothing to obtain.
Completion rates
The most reported training metric in the world measures whether a video finished playing. It is a measure of compliance with an obligation, not of capability, and should never be presented as the latter.
Physical training is running the identical principles
Coaches figured out most of this a century before corporate learning did, and they use clearer words for it.
| In the gym | In the classroom | The shared principle |
|---|---|---|
| Progressive overload | Desirable difficulty | Adaptation only happens just beyond current capacity. Comfortable work maintains; it does not build. |
| Specificity | Transfer of learning | You get better at what you actually practise. Training that resembles the real task transfers; training that resembles a lecture does not. |
| Periodisation | Spaced curriculum | Effort distributed and varied over months beats a single heroic block, in both domains, by a wide margin. |
| Recovery and deload | Sleep and consolidation | The adaptation happens after the session, not during it. Back-to-back intensive days produce fatigue rather than gains. |
| Detraining | The forgetting curve | Unused capacity decays on a predictable schedule. Maintenance costs a fraction of rebuilding. |
| Technique before load | Fundamentals before application | Practising a bad pattern under pressure makes the bad pattern permanent. Slow and correct comes first, always. |
Twelve words from the training brief
- Training needs analysis
- Establishing what the gap actually is before designing anything. The step most often skipped, and the reason most courses answer the wrong question.
- Learning objective
- What a person will be able to do afterwards, written as an observable action. "Understand the policy" is not one; "correctly classify ten incidents" is.
- Retrieval practice
- Recalling from memory rather than reviewing. The highest-return technique in the whole field, and the least popular with learners.
- Spaced repetition
- Reviews scheduled at increasing intervals, timed to just before the point of forgetting.
- Interleaving
- Mixing problem types within a session instead of blocking them. Feels worse, performs better.
- Deliberate practice
- Targeted work on a specific weakness, at the edge of ability, with immediate feedback. Not the same as experience.
- Transfer
- Whether a skill learned in one setting appears in another. The whole point, and the thing least often measured.
- Blended learning
- A mix of self-paced material and live sessions. Effective when the self-paced part is the input and the live part is the practice — and pointless when both are lectures.
- Microlearning
- Short units delivered over time. Powerful because it enforces spacing, weak when used to chop a lecture into fragments.
- Competency framework
- A written description of what capability looks like at each level. Provides the standard everything else is measured against.
- Scaffolding
- Support that is deliberately removed as capability grows. Support that is never removed produces dependence, not skill.
- Refresher
- Scheduled repetition of a rarely used skill. Mandatory in safety-critical work precisely because competence decays invisibly.