The case for replacing the CV with a job simulation is usually made with a stack of borrowed numbers: a cost per bad hire, a retention multiple, a bias reduction, a famous employer that cut its hiring time. This dossier used to make it that way. When we opened the documents those numbers were credited to, most of them were not there. What is there is narrower, older and more useful: a research literature that has ranked hiring methods for decades, a correction to that literature in 2022, and two employer documents that describe what is changing and what is not.
This is the rewritten dossier. It rests only on what those documents print, and it reaches one conclusion that the evidence supports without help: the ranking of hiring methods survived its correction, but the size of every advantage shrank. That makes the way a simulation is scored more important than the fact that it is a simulation.
What the research actually ranked
In 1998 Frank Schmidt and John Hunter published in the Psychological Bulletin a synthesis of research on how well selection methods predict job performance. It became the reference table of the field. It placed general mental ability, work sample tests and structured interviews near the top, and it placed measures such as years of experience and years of education near the bottom. Years of job experience, in that table, carries a validity of .18.
That coefficient is the one most often quoted as the validity of the CV, and this dossier quoted it that way. It is not. The paper measures years of experience, not the act of reading a résumé. A CV is a bundle of signals, some of which the paper measured separately and most of which it did not measure at all. What the paper does support is the ordering: methods that observe or structure the work sit above methods that summarise a history.
The ordering is the finding worth keeping. It says that a hiring decision gets better the closer it comes to watching the candidate do something the job requires, under conditions the evaluator controls. That is the entire theoretical case for the job simulation, and it predates the word.
What the correction changed
In November 2022 Paul Sackett, Charlene Zhang, Christopher Berry and Filip Lievens published in the Journal of Applied Psychology a reanalysis of the same literature. Their target was a technical step: the correction applied to raw validity estimates to account for the fact that studies only observe people who were hired. They concluded that the usual corrections were too large, so the validity of many selection procedures had been substantially overestimated.
Their abstract states the result in two sentences that matter here. Most of the procedures that ranked high in earlier summaries remain high in rank, with mean validity estimates reduced by .10 to .20 points. And structured interviews emerged as the top-ranked procedure.
Read together, those two sentences dissolve the old sales pitch and keep its substance. The ranking held, so the argument that observed and structured methods beat summarised history still stands. The magnitudes fell, so any claim that a work sample is some multiple more predictive than a CV no longer has arithmetic under it. And the method that rose to first place is not a format at all. A structured interview is an ordinary conversation to which three things have been added: the same questions for every candidate, questions drawn from the job, and answers scored against anchors written before the interview began.
That third addition is the whole lesson. What lifts a method up the table is not realism and not technology. It is structure and scoring.
Why a simulation can fail the test it was built to pass
A job simulation borrows its authority from the work sample, and the work sample earned that authority under specific conditions. The task was drawn from an analysis of the job. Every candidate faced the same task. Trained evaluators scored the output against criteria fixed in advance. Remove any of those conditions and what remains looks like a work sample and measures something else.
Three failures follow directly from the research, and each is common in practice.
The first is the unscored exercise. A candidate completes a case, a coding task or a role play, and a hiring manager forms an impression of it. The exercise is realistic, the judgement is not structured, and the result behaves like the unstructured interview the literature ranks lower. An unscored simulation is a CV with extra steps.
The second is the proxy task. A game or a puzzle is used because it is engaging and cheap to administer at scale, and it measures a trait that is only loosely connected to the role. Its validity is whatever that trait's validity is, which may be respectable, but it is not the validity of a work sample and should not be sold as one.
The third is the drifting rubric. A simulation starts with written criteria, and over successive hiring rounds evaluators adjust them to fit the candidates they liked. The scoring becomes a record of preference. The format has not changed, and the measure has quietly stopped being a measure.
In each case the failure is invisible from the outside. The candidate sees a simulation, the employer reports that it uses simulations, and the evidence that made simulations worth using no longer applies.
What employers say is changing
The World Economic Forum's Future of Jobs Report 2025, published in January 2025 from a survey of employers, gives the demand side of the picture. Its headline is about training rather than hiring: if the world's workforce were made up of 100 people, 59 would need training by 2030. Among the practices employers plan to use to improve the availability of talent, the report lists removing degree requirements in favour of skills-based hiring, a plan it records for 19% of employers across all industries.
That share deserves to be read at its real size. It is a stated plan, not an observed practice, and it describes fewer than one employer in five. The shift to skills is real in the survey and modest in scale, which is consistent with the research: a method known to predict better for decades is being adopted slowly, because building a scored assessment costs more than reading a CV.
LinkedIn's Economic Graph Research Institute published on 3 March 2025 a research note, Skills-Based Hiring: Increasing Access to Opportunity, that measures something different and is often confused with the first. It compares two ways of defining who is eligible for a job: people who held the same job title in the past five years, and people whose recent roles share at least half of the target job's top skills. Across the countries analysed, the skills-based pool is a median 6.1x larger. For France the note gives the example of supply chain managers, who share 17 of their top 30 skills with logistics managers and demand planners, and it estimates a 23.4x expansion of the talent pool in construction.
These are measurements of the pool, not of performance. The note says so itself: it does not filter by education or other hiring signals, and actual hiring is more complicated. A larger pool is an opportunity only if the employer has a way to tell the good candidates in it from the rest. A CV cannot do that for people whose job titles do not match. A scored work sample can. That is the point at which the employer document and the research literature meet.
Where the two lines of evidence meet
Put the four documents side by side and a single mechanism appears.
Skills-based hiring widens the pool. The LinkedIn note shows how much wider it becomes when employers stop filtering on job titles, and it shows that workers without degrees gain slightly more from the change than degree holders, by 6% in the global talent pool. A wider pool raises the cost of screening, because more of the applicants have unfamiliar histories. The CV, which was cheap to read and roughly informative for people on conventional paths, becomes less informative exactly where the new applicants are.
The research literature says what replaces it: a method that observes or structures the work. The 2022 correction says how much to expect: less than the old tables promised, but the same ordering. And the correction's top-ranked method says what the replacement must contain: fixed tasks drawn from the job, the same for everyone, scored against anchors written in advance.
Seen this way, the job simulation is not a new technology replacing an old document. It is the price of a wider pool. Employers who widen the pool without paying for a scored assessment end up screening a larger crowd with a less informative signal, which is worse than where they started.
What the evidence does not show
An honest dossier on this subject has to state its limits as clearly as its findings.
None of the documents cited here measures the retention of employees hired through simulations, the reduction of bias attributable to them, or the time saved by any named employer. Those numbers circulate widely and this dossier used to print several of them. We withdrew them because no document we could open prints them in the form they were quoted, and the withdrawal notice on this page lists each one with what the attributed source actually says.
The research literature is largely built on studies from North America, and its estimates are averages across many jobs. The size of the advantage for any single role, in any single French employer, is not known from these documents. The OECD's Skills Outlook treats skills assessment as a labour market signal, which supports the general direction of this dossier, but it does not supply a validity estimate for simulations.
Consulting research on skills-based organisations, including McKinsey's work on building the future workforce around skills, describes employers redesigning roles around assessed capability. It describes intentions and cases, not controlled comparisons, and we cite it for the direction of practice, not for any measure of effect.
What this means for a candidate
For a candidate the practical conclusion is less exciting than the marketing and more reliable. A simulation is worth preparing for when it is scored, and the way to tell is to ask. A serious employer can say what the task is drawn from, whether every candidate receives the same one, and what the scoring criteria are. An employer who cannot answer is running an audition, and an audition rewards confidence more than competence.
Inspectable work is the credential that travels. A scored simulation result, a project an evaluator can check line by line, a task completed under the conditions of the role: these are signals a stranger can trust without knowing the school or the previous employer. For candidates without a degree, or whose job titles do not match the role, they are the only signals that open the wider pool the LinkedIn note describes.
Preparation should follow the scoring rather than the format. Practising the medium, a particular case style or coding platform, helps less than understanding what a competent answer looks like in the role, because a well built rubric scores the answer, not the fluency with the tool.
What this means for an employer
For an employer the conclusion is a discipline, not a purchase. Replacing CV screening with a simulation improves a hiring decision only if the simulation keeps the three conditions the research attaches to it: a task drawn from the job, the same task for everyone, and scoring against criteria written before the first candidate is seen. The 2022 correction also suggests a cheaper starting point than most employers assume. The top-ranked method is a structured interview, which requires no platform at all, only the discipline of fixed questions and anchored scores.
The corrected estimates also change what an employer should expect. Every method predicts less than the old tables said, so no single assessment should carry a hiring decision alone. Combining a structured interview with a scored work sample is the natural consequence of a literature in which the best methods are good, not decisive.
The reading
The CV is not being replaced because it is old or because simulations are new. It is being replaced, slowly and in a minority of employers, because widening the pool to people with unfamiliar histories makes it less informative exactly where it is needed most. The work sample is the method the research has ranked above it for decades. The 2022 correction kept that ranking and shrank every advantage, and it placed a scored, structured method at the top. The work sample is the credential, but only when it is scored against the real task.