A manager sends their team on a two-day course. Everyone enjoys it, the feedback forms glow, and a fortnight later almost nothing has changed. This is the central, slightly embarrassing fact of corporate learning and development: the event is easy, the change is hard, and most programmes quietly optimise for the event. Getting people genuinely better at their jobs is a different discipline, and it starts by admitting that a course is the smallest part of it.

The quick version

  • Learning & development (L&D) is the organised effort to grow people's skills and capability, through courses, but far more through the work itself, coaching, and feedback.
  • Most learning happens on the job, not in training rooms. The popular shorthand for this is the 70-20-10 model (experience, others, formal learning), a useful reminder, even though the exact numbers are not strong science.
  • People forget fast. Without spacing and practice, the bulk of new material fades within days, Ebbinghaus's "forgetting curve." The fix is repeated retrieval over time, not longer courses.
  • Judge L&D by behaviour change and results, not by attendance and happy sheets. Kirkpatrick's four levels is the standard way to keep that honest.

The idea in depth

Strip away the jargon and L&D answers one question: how do people in an organisation get better at what they do? The unhelpful default answer is "we send them on a course." The more honest answer is that a course is one small input, and usually not the one that does the work.

The most widely cited expression of this is the 70-20-10 model, which holds that roughly 70% of development comes from challenging on-the-job experience, 20% from other people (managers, mentors, coaching, feedback), and 10% from formal training. It grew out of research at the Center for Creative Leadership in the 1980s by Morgan McCall, Michael Lombardo and Robert Eichinger, who asked successful executives to recall the experiences that shaped them; it was popularised in Lombardo and Eichinger's The Career Architect Development Planner (1996). The executives pointed overwhelmingly at hard assignments and key relationships, not classrooms.

So the move is: stop treating "development" as a synonym for "training budget." If most growth comes from the work, then the development decision that matters most is which assignment a manager hands someone, a stretch project just past their current ability, with enough support that they can succeed. The course is the garnish; the stretch assignment is the meal.

An honest limitation. The precise 70-20-10 split is not robust evidence. As the Association for Talent Development has noted, the ratios trace to executives' retrospective self-reports, a small, senior sample, recalling years later how they thought they learned ("70:20:10: Where Is the Evidence?", ATD). The suspiciously round numbers should be read as a rule of thumb, "experience and people matter far more than courses", not as a measured ratio to allocate a budget against.

Why training evaporates, and what to do about it

Even when formal learning is the right tool, most of it leaks away. The oldest evidence here is also among the most durable: in 1885 Hermann Ebbinghaus, testing his own recall of nonsense syllables, mapped the forgetting curve, memory of new material drops sharply within hours and days unless it is reinforced. A single training event sits right on the steepest part of that curve.

The antidote is not more content; it is spacing. The benefit of spreading practice out over time, rather than cramming it, is one of the best-replicated findings in learning science. Cepeda, Pashler and colleagues' meta-analysis in Psychological Bulletin (2006) pulled together 317 experiments and found distributed practice reliably beat massed practice for long-term retention (Cepeda et al., 2006). Pulling a memory back out, retrieval practice, strengthens it far more than re-reading does.

More learning sticks from three short, spaced practice sessions than from one long course twice the length.

So the practical move is to work with the curve, not against it. Replace the one-off two-day course with a shorter session followed by spaced reinforcement, a five-minute quiz a week later, a peer discussion a fortnight after that, a real task that forces people to use the skill within the month. The goal is to make people retrieve the material several times across weeks, which is exactly what the forgetting curve says you must do.

flowchart LR
  A(["Learn it once
(course / session)"]) --> B(["Forgetting curve
memory drops fast"])
  B --> C(["Space it
retrieve days later"])
  C --> D(["Space again
retrieve weeks later"])
  D --> E(["Use it for real
on a live task"])
  E --> F(["Sticks
now it's a skill"])
Spacing beats cramming: each retrieval flattens the forgetting curve a little more. Leaders Loop

Proving it worked: Kirkpatrick's four levels

If experience does most of the work and training mostly evaporates, how do you know any of your L&D effort is paying off? The standard answer, in use since 1959, is Donald Kirkpatrick's four levels of evaluation: Reaction (did people like it?), Learning (did they actually acquire the knowledge or skill?), Behaviour (are they doing things differently back at work?) and Results (did the business outcome move?). It was refreshed by James and Wendy Kirkpatrick as the "New World" model in 2016, but the four rungs are unchanged.

flowchart TB
  A(["1 · Reaction
did they like it?"]) --> B(["2 · Learning
did they acquire it?"])
  B --> C(["3 · Behaviour
are they doing it differently?"])
  C --> D(["4 · Results
did the business outcome move?"])
  C -.->|"where most
programmes fail"| C
Kirkpatrick's four levels, the value climbs as you go down the list, but most evaluation stops at level 1. Leaders Loop

The reason the model still matters is that almost everyone stops at Level 1. Happy sheets are easy to collect and flatter everyone involved; behaviour change is hard to observe and sometimes unflattering. But Level 1 reaction is a famously weak predictor of whether anyone got better, people routinely rate enjoyable sessions highly and change nothing.

This is where the field's hardest truth lives: the transfer problem. The classic review is Baldwin and Ford's "Transfer of Training" (Personnel Psychology, 1988), which framed transfer as depending on three things: the trainee, the training design, and, crucially, the work environment they return to. Skills don't transfer in a vacuum; they transfer when the job rewards using them.

The move here is to set the bar at Level 3 before you run anything. Decide, in advance, what people should be doing differently and how you'll see it. Then engineer the work environment to pull the new behaviour through, a manager who asks about it in the next one-to-one, a checklist that bakes it into the workflow, a visible moment to apply it. Evaluation isn't an after-the-fact survey; it's the design brief.

An honest limitation. Be wary of one zombie statistic that haunts this topic: the claim that "only 10% of training transfers to the job." It is widely repeated and poorly founded. As Robert Fitzpatrick documented in a piece on Will Thalheimer's learning-research blog, the figure traces back to a 1982 article by David Georgenson that posed it as a rhetorical question, not a finding, which later writers hardened into a fake "fact," sometimes bolted to invented dollar amounts (worklearning.com). The transfer problem is real; that specific number is not. Cite the principle, not the phantom percentage.

A worked example

Illustrative figures, a composite scenario, not a specific company's data.

Priya runs a 30-person customer-support team. Complaint-handling is weak, so the obvious move is a workshop. Version one: she books an all-day "Handling Difficult Customers" course. Everyone attends, the trainer is excellent, feedback averages 4.6 out of 5. Six weeks later, escalation rates are unchanged. She has bought a Level 1 success and a Level 3 failure.

Version two starts from the behaviour she wants: agents should acknowledge the customer's problem in their own words before offering a fix. So she designs backwards. A 90-minute session teaches the one technique. A week later, agents do a five-minute role-play in their team huddle (spaced retrieval). Priya adds a single line to the call-review template, "Did the agent acknowledge before solving?", so the work environment now pulls the behaviour through (Baldwin and Ford's missing third factor). Team leads coach to that one line in their one-to-ones for a month (the "20"). The "course" shrank; the development grew. Because the skill is rehearsed, observed and rewarded on the actual job, it survives the forgetting curve and shows up at Level 3, which is the only level Priya's director will ever notice.

Frequently asked questions

Is classroom training a waste of money, then?

No, but it's a starter, not the meal. Formal learning is good for introducing a model, a vocabulary, or a safety-critical procedure consistently. It fails when it's treated as the whole intervention. Pair every course with spaced practice and a real on-the-job application, or expect most of it to evaporate.

What's the difference between learning and development?

Loosely: learning is acquiring a specific skill or knowledge now; development is the longer arc of growing a person's capability for bigger or different roles over time. A SQL course is learning. Building someone toward a team-lead role across two years of assignments, coaching and feedback is development. L&D as a function covers both.

How is this different from career development and succession planning?

L&D is the engine of capability-building; career development and succession planning is the destination it serves, making sure the right people are ready for the right roles when seats open up. You can't plan a succession into a bench you never developed.

We're small with no training budget. Where do we start?

You already have the two biggest levers and they're free: assignments and feedback (the "70" and "20"). Hand someone a stretch task slightly beyond their comfort, then make time to debrief it well. A manager who deliberately delegates challenging work and coaches on it out-develops most training catalogues, see delegation & empowerment.

How do we measure L&D without a fancy system?

Pick one behaviour per programme and watch for it. Define what "doing it differently" looks like, agree how you'll spot it (a metric, a sample of work, a manager's observation), and check at 30 and 90 days. That's a credible Level 3 evaluation with no software at all.

Related in the Toolkit

Where to go next