Rethinking Corporate Learning: Target Gaps, Boost Performance

For decades, companies have treated learning as something that happens at the margins of work.

You work and, every once in a while, you learn.

You sell, you run a machine, you attend to a patient, you supervise a shift. Then, somewhere outside that reality, a course appears — meant to improve whatever happens when the person returns to work.

But consider why a salesperson sells less.

Maybe they know the product poorly. Or they get worse leads. Or the branch has less traffic. Or the system is slow. Or the incentive rewards something else. Or the territory is structurally harder.

A manager has a weak team because they can't develop people — or because the organization demands they sell, administer, report, and resolve exceptions until no time is left to lead.

In every case there is a performance gap.

In only some is there a learnable gap.

There lies the great anomaly of corporate learning: we spend enormous resources trying to modify people before demonstrating that the problem lies in something people can learn.

The consequence is not educational. It is economic.

We Measure the Container

Training architecture begins with populations. Four thousand salespeople. Eight hundred supervisors. Then we decide what content each needs.

But work rarely follows the categories through which HR organizes a company.

A nurse and a mechanic sit in different industries yet share the decisive trait — both must recognize patterns under pressure, where errors are costly. A collections agent and a field seller sit in different functions, yet both work in motion, in four-minute intervals rather than protected hours.

What matters is not who the worker is, but the properties of the work: mobility, time pressure, variability, autonomy, observability, cost of error, dependence on judgment.

The question stops being how to train populations better. It becomes: which part of an outcome can be modified through learning, inside this specific system of work?

That shift looks academic until you follow the money.

The industry sells courses. Companies buy performance.

No company needs learning opportunities. It needs fewer accidents. More sales. Less rework. Lower turnover. Less time to competence.

A person can complete a course and not change behavior. Change behavior and improve no outcome. Improve an outcome that vanishes six weeks later. Sustain an improvement worth less than it cost to produce.

The real chain is long: exposure, acquisition, transfer, performance, persistence, value. Organizations observe its beginning with ease. Who enrolled. Who finished. Who passed. Who said they were satisfied.

Far less often do we know whether a different behavior appeared in real work — and survived once the course and the novelty disappeared.

The metrics easiest to obtain are the ones furthest from the value that justified the investment. It resembles counting how many people walked into a gym when the question was whether anyone got stronger.

We measure the container because it is easier than measuring the transformation.

Transfer Is the Hidden Market

Watch how an organization actually learns and a different infrastructure appears — one that never shows up in the budget.

The employee asks the colleague who knows. The technician sends a photograph. The new hire imitates the veteran. A nurse asks another nurse.

Learning happens embedded in production.

We call this informal learning, as if it were an incomplete version of the formal kind. It may be exactly the reverse: people build their own access to knowledge and feedback wherever the formal system fails to arrive with enough speed or context.

A platform does not only compete against another platform. It competes against the colleague who answers in thirty seconds. Against the reliable supervisor. Against an AI that responds immediately.

If the informal alternative solves the situated problem better, the pedagogical superiority of the course is irrelevant. The worker is not choosing between two educational experiences. They are trying to finish their work.

What stays scarce is not information. It is the ability to insert the right help into the exact moment a person must act differently. A scarcity of context.

AI Makes Mediocre Learning Nearly Free

Explaining, generating exercises, simulating conversations — the cognitive production behind learning is collapsing in cost.

Abundance rarely increases the value of what it makes abundant. It displaces it.

When everyone can generate explanations, value moves to knowing which explanation this person needs now. When everyone can deploy assistants, value moves to understanding the system of work the assistant is meant to help.

AI will make it extraordinarily cheap to build something that looks like learning.

That is not a criticism of AI. It is a description of its economic power.

Precisely for that reason, proving that something changed performance becomes a scarce asset. Value migrates from production to causality.

Companies Also Manufacture Expertise

There is a second cost. Much of professional judgment is built performing tasks that, on an income statement, look inefficient: the junior drafting an analysis the senior could do faster, the apprentice repairing under supervision.

Judged on immediate output, these are candidates for automation. But what happens when we automate not only a task, but the experience through which someone learned to handle harder ones?

A company can raise productivity while quietly degrading capability. Not all friction is learning — much of it is waste. But some tasks are schools disguised as work, and eliminating both by the same criterion looks like efficiency for several years.

Until the company needs the experts it stopped manufacturing.

The Average Is a Hiding Place

Suppose an intervention lifts average performance 8%. It sounds useful.

But the average can hide three worlds: 25% in some units, zero in others, negative in the rest.

The administrative temptation is to summarize. The strategic opportunity is to explain the dispersion. Where it worked, what else was present? More practice? Better supervision? Real chances to apply it?

The average tells you whether something happened. The variation tells you why — and the why is the asset. Once you know the conditions that produce change, you no longer own content. You own a capability to intervene, and that is far harder to copy.

Learnable Performance

We may need a different category. Not training. Not even learning and development.

Learnable Performance: the portion of human performance that can be modified reliably, observably, and persistently enough through learning.

Every word is uncomfortable on purpose.

Portion, because not everything depends on learning. Reliably, because an occasional improvement is not a capability. Observably, because intention is not behavior. Persistently, because a behavior that dies with the intervention has little value. Through learning, because if the bottleneck was the system, the incentive, or the tool, calling the solution educational hides the problem instead of solving it.

The objection is obvious: work is too complex, causality is hard, some capabilities take years to surface. All true. But the current alternative is not neutral either.

When we cannot demonstrate causality, we substitute activity. When we cannot measure performance, we measure participation. When we cannot observe transfer, we measure satisfaction.

The difficulty of measuring something does not make the metric we happen to have valuable.

The next frontier of corporate learning may not be inventing better ways to teach. It may be learning to recognize, with far greater precision, when teaching is actually the answer.

In a plant, someone finishes a course and returns to a machine. In an ambulance, a person recognizes a pattern they would have missed.

From a distance, all of it can be called learning.

Up close, only one thing matters.

That someone does something they could not do before, at the moment it actually counts.

The rest was training.