70:20:10: the ratio that isn’t research.

Some numbers do work in the world. Others do rhetorical work about the world. The learning field has both kinds: effect sizes that sum up thousands of measured outcomes, and figures that survive because they are useful to say in meetings. Telling them apart is a professional skill — arguably the skill for anyone who buys, builds, or defends training. And no figure offers better practice than the most famous one in the industry.

Almost every L&D strategy deck holds the same pie chart: 70% of learning happens on the job, 20% through other people, 10% in formal training. The ratio has the two marks of perfect corporate folklore. It is precise enough to sound measured. And it is handy enough to justify almost any decision — most often the decision to spend less on the 10.

It is worth being exact about what stands behind it. The answer is: a memory survey. It is worth being just as exact about what the real research on informal learning shows. The honest picture is more interesting than the folklore or the debunking. Informal learning turns out to be real, large, tied to concrete behaviors, and open to management. That is better news for firms than the ratio ever was — and news the ratio’s arithmetic hides.

Where the number came from

Ask a room of learning professionals where 70:20:10 comes from. The confident answers name “decades of research,” “a Princeton study,” or simply “the science.” The real trail is short. It ends in one place.

The source is the Center for Creative Leadership’s work on executive development in the 1980s, published as The Lessons of Experience. Researchers asked 191 successful executives, across six major firms, to recall the key events of their careers and what each event taught them (McCall, Lombardo & Morrison, 1988). The answers were rich, and the book remains a genuinely useful study of executive careers.

But note what the method is. It is self-report, years after the fact, from people picked precisely because they had succeeded, about which remembered events felt formative. Nobody measured learning. Nobody compared conditions. Nobody asked the executives who didn’t make it.

The tidy ratio came later. A planning handbook set it down as a rule of thumb for designing growth assignments for executives (Lombardo & Eichinger, 1996). From there it broke loose from its context entirely. A rule for stretching high-potential managers became a universal law of how all humans gain all skills.

What a memory survey can and cannot carry

None of this makes the original research bad. It makes it research about something else. As a study of which experiences executives credit for their growth, the CCL program was thorough. Its findings — stretch assignments matter, hardship teaches, bosses shape careers — remain plausible and useful. The trouble starts when its answers are recast as measurements of where learning happens. Every inferential step in that move is broken.

The sample is survivors only: executives whose careers stalled — perhaps for want of formal training — were never interviewed. So the design cannot see what the successful path lacked. The tool is recall, decades later, and recall favors the dramatic over the gradual: a botched product launch is memorable; the forty small gains in skill that let you survive it are not. And the output is a count of remembered events, with no unit that could ever yield a percentage of learning. The 70, 20, and 10 are shares of anecdotes, not of skill.

Run the same method on any field and the absurdity shows at once. Interview top surgeons about the experiences that made them, and most will name cases, complications, and mentors — not anatomy lectures. No one would conclude that medical schools should cut anatomy to 10% of the course. The lectures do quiet, load-bearing work that memory cannot see. They made the formative cases survivable, and readable, in the first place.

The searches for the missing evidence

A fair question at this point: does the pedigree matter, if the ratio’s spirit — most learning happens outside the classroom — is broadly right? It matters because the number is not used in spirit. It is used in arithmetic. Budgets are split against it, formal programs are capped by it, and vendors are hired to “enable the 70” on the strength of it. A slogan can survive being roughly right; a budget formula cannot. Once the digits drive decisions, where the digits came from becomes a due-diligence question. This origin story does not survive one.

Because the ratio is cited as “research shows,” people have gone looking for the research. A systematic search set out to trace the model’s empirical base and found none. The trail dead-ends in the memory surveys, and in later surveys asking managers what they believe the proportions to be (Kajewski & Madsen, 2012). The most thorough academic autopsy reaches the same verdict, and gives the genre its best subtitle. The claim that informal learning dominates is a fact in search of evidence: its supporting citations, followed to their roots, cite each other and, in the end, the 1988 memory study (Clardy, 2018). Field research on the framework in use finds it working as a budgeting rule and a way to talk about learning, while its effect on actual learning transfer remains unshown (Johnson, Blackman & Buick, 2018).

How a heuristic became “research shows”

The ratio took forty years to travel from planning handbook to universal truth. The trip is a case study in how ideas launder their origins, and the mechanism deserves naming because it is still running. Each retelling upgraded the claim’s credentials a little. A rule of thumb became “the CCL model.” The model became “the research.” The research became a percentage precise to two digits.

Citations began pointing at other citations — conference decks citing vendor whitepapers citing earlier decks — until the origin was invisible from any single link. The number took on the authority of things everyone knows, the kind no one can test. The academic autopsies that finally walked the full chain found what such checks usually find. Every path ends in the same small memory study, bearing none of the weight stacked on it (Kajewski & Madsen, 2012), (Clardy, 2018).

Two features made this number so contagious. It flatters its audience: “most learning happens at work” tells operating managers that their world is where growth really lives. And it licenses a saving: it hands budget owners a research-flavored reason to cut the one learning line item with a price tag. An idea that both compliments the powerful and cuts costs does not need evidence to spread; it needs only the absence of anyone checking. The field’s slow reckoning with the ratio has been less a scientific correction than a profession growing up. The same discipline that learned to retire learning styles is learning to ask, of its own founding numbers, who measured that, and how?

What the informal-learning evidence actually shows

It is worth spelling out what a study able to support the ratio would look like. The contrast lights up the gap. It would follow workers forward in time, not ask survivors to look back. It would measure skill directly at intervals — tests, work products, trained observation — rather than ask anyone what they remember learning. It would log the learning exposures in between: courses taken, feedback received, problems met, colleagues consulted. And it would tie measured skill change to those exposures, per role and tenure — almost surely finding different mixes for a first-year analyst and a twenty-year executive.

No such study backs 70:20:10. The honest guess from nearby fields is that the result would be a family of curves that vary by context — not three tidy universal numbers.

Here is where honest skepticism has to turn and defend the ratio’s kernel. Informal workplace learning is real, large, and consequential. The founding qualitative work mapped its forms: quiet pattern-learning from doing the work, deliberate self-directed inquiry, and learning from the reactions and examples of colleagues (Eraut, 2004). Reviews of the field treat informal learning as a large share of workplace growth, while declining — correctly — to put a universal number on it (Tannenbaum, Beard, McNall & Salas, 2010). And a meta-analysis — a study that pools many studies — finds informal learning behaviors meaningfully linked to performance (Cerasoli et al., 2018). People who ask, experiment, seek feedback, and reflect do better.

Those behaviors respond to conditions firms can shape — psychological safety, manager modeling, slack time. That makes them a far better policy target than a fixed ratio: not “assume 70% happens,” but “raise the frequency of five observable behaviors”. The defensible sentence is: much workplace learning is informal; nobody knows the percentage; it certainly varies by role, tenure, and industry.

Meanwhile, the “10” the ratio is used to shrink has meta-analytic receipts of its own. Designed training works. Its effects on learning and job behavior are among the better-established results in applied psychology (Arthur, Bennett, Edens & Bell, 2003). That holds when the training is built around practice and feedback and followed into the workplace. That is the transfer problem our compliance-training review covers (Blume, Ford, Baldwin & Huang, 2010).

And here is the irony the folklore hides. The original CCL research itself points toward designed experience, not abandoned experience. The executives’ formative assignments were stretch roles with real stakes, visible outcomes, and bosses watching. They were informal in venue, but rich in the challenge, feedback, and accountability that every learning literature names as active ingredients (McCall et al., 1988).

A defensible experience strategy takes that seriously. Choose assignments against a person’s diagnosed gaps. Pair them with reflection and manager debriefs. Measure the skill change that follows. “They’ll pick it up on the job” — the policy most often justified by the 70 — contains none of those ingredients. It is the one version of learning from experience the source material never described.

The real cost of the folklore

The trouble with 70:20:10 is not that it overrates informal learning. The trouble is that it launders unmanaged learning as a strategy. The “70” in most firms is not designed experience of the kind the CCL researchers studied. It is whatever happens to happen. No diagnosis, no sequence, no feedback, no measurement — the exact ingredients every literature in this library names as the difference between exposure and learning. Citing the ratio to cut formal spending, while leaving the 70 unmeasured, is a plan to fund the measured part less and the unmeasured part not at all.

A lopsided burden of proof completes the trap. Formal training is visible and budgeted, so it is asked to prove itself: completion rates, tests, transfer studies. The “70” is asked to prove nothing, ever, because no one owns it and nothing records it. The ratio thus works as a one-way valve: evidence standards apply exactly where evidence can exist, and are waived where it cannot. An honest learning strategy applies one standard everywhere — show me the skill movement — and funds whatever produces it, in whichever venue. Under that standard, the line between formal and informal mostly dissolves; what remains is designed versus undesigned, measured versus assumed.

What the evidence doesn’t show

• It doesn’t show formal training deserves 10% of attention. The ratio’s most common use — as a budget allocation — is exactly what its provenance cannot support (Kajewski & Madsen, 2012), (Clardy, 2018).
• It doesn’t show the proportions are stable. Even sympathetic readings note the mix must vary enormously — a new hire in a regulated role and a twenty-year veteran do not learn in the same proportions from the same sources (Tannenbaum et al., 2010).
• It doesn’t show remembering equals learning. Salient career events are memorable precisely because they were dramatic; the daily accretion of skill through practice — which the memory method systematically undercounts — is where most durable capability comes from.
• The debunking doesn’t license ignoring informal learning. The correlational evidence that informal learning behaviors track performance is solid (Cerasoli et al., 2018). The target of skepticism is the ratio, not the phenomenon.

Where the evidence stops

1. 1It doesn’t show formal training deserves 10% of attention
2. 2It doesn’t show the proportions are stable
3. 3It doesn’t show remembering equals learning
4. 4The debunking doesn’t license ignoring informal learning

© 2026 FUTURE PROOF™

What this means for practice

The advice here is deliberately plain, because the ratio’s whole appeal was glamour — one chart, three numbers, strategy done. What replaces it is ordinary evidence hygiene, applied to learning.

Retire the pie chart, and replace the question it pretends to answer. “What are the right percentages?” has no empirical answer and never did. “Where did learning actually move last quarter, and what caused it?” has one — if the plumbing exists to ask it.

Building that plumbing is less exotic than it sounds. Start with baseline skill checks. Weave review and recall practice into the flow of work, so on-the-job time emits learning events. Re-measure at intervals, and tie movement to what came before it. With those pieces in place, your real ratio becomes a report — different by role, tenure, and function, as it always was — not a slide inherited from someone else’s 1988.

Design the experience you are crediting, because the source material’s real lesson survives its misquotation. Experience develops people when it is stretching, supported, and reflected on. Choose assignments against diagnosed gaps. Attach feedback loops and manager debriefs. Measure what moved.

And when the ratio next appears in a deck as “research shows,” ask the presenter for the study. Not to score a point — but because a learning organization’s first skill ought to be telling evidence from folklore about the thing it exists to do.