In short
- Ericsson's four conditions are a goal, motivation, feedback and repetition. A normal working week reliably supplies only the first two.
- Across 88 studies, deliberate practice explained 26% of performance variance in games, 21% in music, 18% in sports, 4% in education and less than 1% in professions.
- The same analysis found the effect was strong in predictable activities and weak in unpredictable ones. Revenue and automation work is unpredictable.
- So the lever is not hours. It is rep design: shorten the loop, repeat the same judgement, and log error types instead of time.
What Ericsson actually claimed
The 1993 paper by K. Anders Ericsson, Ralf Krampe and Clemens Tesch-Römer in Psychological Review is the origin of the phrase. It studied violinists at different levels of accomplishment and argued, in the paper's own abstract.pdf), that individual differences even among elite performers are closely related to assessed amounts of deliberate practice, and that many characteristics once believed to reflect innate talent are the result of intense practice sustained over at least a decade.
Note what the paper did not say. There is no 10,000-hour rule in it. That number was an average for one group of violinists at age 20, later turned into a slogan by popular writing.
Fifteen years on, Ericsson wrote a short overview for Academic Emergency Medicine that is far more useful to an operator than the original. It opens with a claim that should be uncomfortable for anyone selling seniority: professional expertise has traditionally been judged by length of experience, reputation and perceived mastery, and research shows only a weak relationship between those indicators and actual observed performance. Experience, by itself, is not evidence. That is the research backing for MitHub's position on proof of work versus credentials.
The same paper gives the four conditions under which practice has been reliably associated with improvement. Significant gains appeared when people were:
- Given a task with a well-defined goal.
- Motivated to improve.
- Provided with feedback.
- Given ample opportunity for repetition and gradual refinement.
Read that list against your last working week. Most knowledge work supplies conditions 1 and 2 and almost never supplies 3 and 4.
The finding nobody quotes
Then the evidence got tested, and it got smaller.
In 2014, Brooke Macnamara, David Hambrick and Frederick Oswald published a meta-analysis in Psychological Science covering 88 studies and 11,135 participants. Deliberate practice explained 26% of the variance in performance for games, 21% for music, 18% for sports, 4% for education and less than 1% for professions — a result that was not statistically significant. A 2018 corrigendum to that paper, hosted with it, revises the overall figure slightly upward (14% of variance overall) and the professions figure to 1%, still not significant.
In 2019, Macnamara and Megha Maitra went back and replicated the original violin study in Royal Society Open Science, with a double-blind procedure. They did not reproduce the core finding that accumulated deliberate practice tracked each skill level. Practice explained 26% of the variance — substantial, but roughly half the 48% reported in 1993.
Two moderators in the 2014 analysis matter more to your career than the headline:
| Moderator | Variance explained |
|---|---|
| Activities high in predictability (e.g. running) | 24% |
| Activities moderate in predictability | 12% |
| Activities low in predictability (e.g. handling an aviation emergency) | 4% |
| Studies measuring practice with retrospective questionnaires | 12% |
| Studies measuring practice with a log | 5% |
Two things fall out of that table. First, the less predictable the work, the less raw practice predicts performance — and diagnosing a broken revenue process is much closer to the aviation emergency than to running. Second, the better the measurement, the smaller the effect. When people kept logs rather than estimating from memory, the relationship dropped by more than half. A lot of the original story was people remembering their own effort generously.
The authors were also honest about why the professions number was so low: deliberate practice is poorly defined in those domains, and the professions category rested on only seven samples covering computer programming, military piloting, soccer refereeing and insurance selling. That is not proof that practice fails at work. It is evidence that nobody has managed to define what practice even means at work — which is the problem worth solving.
Why office work resists practice
Four structural reasons, and each has a fix.
There is no scoreboard. A chess player has a rating. You have a quarterly review. Without a number that moves, you cannot tell improvement from confidence.
Feedback arrives too late to be feedback. You design a lead-routing rule in March and learn in June that it was misrouting a segment. By then you cannot connect the outcome to the decision. Feedback that arrives after you have forgotten your reasoning teaches nothing.
Nothing repeats identically. Musicians play the same bar fifty times. You run a campaign once, and the next one has different data, a different market and a different budget.
Competence turns into automaticity. This is Ericsson's sharpest point. After limited training, performance adapts to typical demands, becomes automated, and conscious control drops away — and at that point, additional experience stops improving accuracy. He calls the result arrested development, and says aspiring experts have to actively counteract it by setting new goals and higher standards that exceed their current reliable performance. This is the mechanism behind "ten years of experience" that is really one year repeated ten times.
The MitHub Rep Designer
Here is the framework MitHub uses to manufacture the missing conditions. Five steps, run on one skill at a time.
1. Name the judgement, not the task. "Build a Clay table" is a task; you will get faster at it and stop. "Predict which enrichment provider will cover this segment" is a judgement, and judgements can be wrong in interesting ways. Practise judgements.
2. Find or fake the scoreboard. Every judgement has a checkable answer somewhere: a coverage percentage, a reply rate, a QA score, a stage-conversion number, a peer's independent verdict. If no scoreboard exists, build the cheapest possible proxy and write down the rule before you look.
3. Predict before you look. This is the single highest-leverage habit in the whole method, because it converts ordinary work into a rep at zero cost. Before running the query, write the number you expect. Before opening the transcript, write the outcome you expect. The gap between prediction and reality is the feedback signal, and it arrives in minutes rather than months.
4. Compress the loop. Deliberately choose the version of the task with the fastest verdict. Reviewing ten past call transcripts blind gives you ten reps in an hour; waiting for ten new calls gives you ten reps in a month. Historical data is a practice simulator, and most teams are sitting on one.
5. Log the error type, not the hours. Hours logged predicts nothing, as the research above shows. What you want is a growing list of the kinds of mistake you make — "I trust a field that is 40% empty", "I assume the CRM stage was set by a human". Patterns in that list tell you what to practise next.
A rep library for revenue and automation work
Concrete reps you can run this week, each with a built-in verdict.
| Skill | The rep | The scoreboard |
|---|---|---|
| Data judgement | Before pulling a report, write down the row count and the share of empty phone numbers you expect | The actual query result |
| Diagnosis | Read a pipeline report for 5 minutes, write the one number you think is leaking, then go and check | The stage-conversion math |
| Prompt and agent design | Write the failure you expect from a prompt, then run it 10 times | How many of the 10 failed the way you predicted |
| Workflow design | Write the error path before the happy path, then break the workflow on purpose | Whether your error handling caught it |
| QA and judgement of AI output | Score 10 archived call transcripts blind against a rubric, then compare with someone else's scores | Agreement rate |
| Commercial explanation | Write a 5-sentence summary of a system for a non-technical owner; ask them what they would do next | Whether they name the action you intended |
Notice that none of these require a new job, a budget or permission. They require archived data and the discipline to write the prediction down first. The same loop underpins how to learn AI automation and the scientific-method cycle in the faculty's Operate chapter: observe, hypothesise, build, measure, learn, adjust.
A six-week example (hypothetical)
Imagine Andrés, an operations analyst who wants to become a revenue engineer.
- Weeks 1–2. Judgement: "Can I predict which leads never get contacted?" Rep: each morning he writes his prediction for yesterday's no-contact count and reason, then checks. By day 10 his error log says he consistently underestimates the effect of records with a second phone number.
- Weeks 3–4. Judgement: "Can I predict where a workflow will break?" Rep: for five existing automations he writes the three failures he expects, then tries to cause them. Three of fifteen predictions were right. His error log says he thinks about bad data and never about timeouts.
- Weeks 5–6. Judgement: "Can I explain a leak so an owner acts?" Rep: five one-page summaries, each read by a different colleague, each scored on whether they named the right next action.
Six weeks, roughly thirty reps, one error log with two clear patterns. Compare that with six weeks of doing the job well and learning nothing specific about your own failure modes.
What deliberate practice is not
- Not volume. The log-method finding is the warning: measured practice predicts less than remembered practice.
- Not flow. If it feels smooth, you are performing, not practising. Practice happens just beyond reliable performance.
- Not a guarantee. In unpredictable domains the ceiling on what practice explains is real — starting point, environment, opportunity and luck take the rest. Treat practice as the part you control, not the whole story.
- Not the same as improvement to the business. A rep improves you. Turning that into value still requires shipping something, which is where kaizen's keep-what-works loop and building systems rather than doing tasks come in.
Key takeaways
- Ericsson's conditions — clear goal, motivation, feedback, repetition — are the real content of deliberate practice, and work supplies only half of them.
- The evidence is weakest exactly where most readers live: professions, and low-predictability tasks.
- Experience is a poor indicator of performance, and automaticity causes people to plateau without noticing.
- Predict before you look, use historical data as a simulator, and keep an error-type log instead of an hours log.
- Practice makes you better; only shipped, documented work makes that visible to a market.