How to Hire the Right People

The problem: your gut is a bad hiring manager

Every hiring funnel has the same hole in it. People come in at the top — an application, a conversation, a good feeling — and the company discovers who it actually hired about three months later, when the money is already spent.

The failure rate is not small. Leadership IQ, a commercial research firm that tracked more than 20,000 new hires through 5,247 hiring managers, reports that 46% of new hires fail within 18 months — and that 89% of the failures are attitudinal, not technical: coachability, emotional intelligence, motivation, temperament. Only 11% fail for lack of skill. Those are managers reporting on their own hires, not laboratory measurements, and should be read that way. But the direction is hard to argue with: people rarely fail because they cannot do the tasks. They fail because of who they are when nobody is watching.

The bill is measurable too. In CareerBuilder and Harris Poll's 2017 survey of 2,257 hiring managers, employers put the average loss from a single bad hire at $14,900, and 74% admitted they had hired the wrong person for a position. A widely repeated rule of thumb claims a bad hire costs at least 30% of the employee's first-year earnings, usually attributed to the US Department of Labor. Nobody has ever traced that figure to an actual Department of Labor publication. We print it here flagged for exactly what it is: a rule of thumb wearing a uniform it did not earn. That distinction — between a number that was measured and a number that was merely repeated — is the spine of everything that follows.

So the question is real, and it is expensive. Can a test, taken before the hire, tell you who will still be standing in ninety days? A century of companies, armies, courts and scientists have worked on that question. Here is what they found.

A short history of asking strangers questions

The idea starts with Frank Parsons, who founded Boston's Vocational Bureau in 1908 and set out the method in Choosing a Vocation, published after his death in 1909: know yourself, know the work, reason honestly about the fit between the two. Trait, factor, match. Every assessment sold today descends from that sentence.

The First World War industrialised it. In 1917, Robert Yerkes and a committee of the American Psychological Association built the Army Alpha test for literate recruits and the Army Beta for those who could not read English. Roughly 1.75 million recruits were examined — the first proof that psychological measurement could run at factory scale. In 1919 Robert Woodworth produced the Personal Data Sheet, generally counted as the first personality inventory; in 1921 the test-makers founded the Psychological Corporation. Testing became a business almost as soon as it became a science.

The famous brands came next. William Marston's 1928 book Emotions of Normal People supplied the theory behind DISC, though Marston himself never built a hiring test — the first DISC instrument was developed by Walter Clarke in the 1950s. The Myers-Briggs Type Indicator grew out of Katharine Briggs's and Isabel Briggs Myers's work during the Second World War, sorting people into sixteen types from four either-or preferences. The Big Five — the five broad trait dimensions modern research actually uses — consolidated in the 1980s and 1990s around Costa and McCrae's NEO instruments.

The most honest experiment of the era belonged to AT&T. In 1956 Douglas Bray launched the Management Progress Study: about 422 young managers passed through assessment centres lasting three and a half days — simulations, business games, leaderless discussions, interviews, tests — and then, radically, the results were sealed. Nobody used them to decide anything. The men were simply tracked. Of 55 who reached middle management, 43 — 78% — had been predicted to. Of 73 who never rose above the first level, 69 — 95% — had been predicted not to. Prediction, tested against twenty years of reality, worked.

From the 1970s, Schmidt and Hunter's validity generalisation research showed that test validity was not local luck, and their 1998 meta-analysis became the field's reference table. Then testing moved onto phones: online batteries in the 2000s, algorithmic video interviews in the 2010s, game-based assessment with Pymetrics and Arctic Shores. HireVue dropped its facial-expression analysis in 2021 after sustained criticism. The instruments got faster, prettier and cheaper. Whether they got more honest is the subject of the next sections.

The instruments: DISC

DISC describes behavioural style — how a person prefers to act — in about fifteen to twenty minutes, at roughly $24 to $130 per person by third-party estimates. It is everywhere in corporate training rooms.

The honest verdict: in preparing this article, we looked for peer-reviewed evidence that DISC scores predict job performance. We found none. There is no published selection-validity estimate behind it. DISC may be a fine way to start a conversation with a team you already employ. As a filter for deciding who gets hired, it is a horoscope with a pricing page — and that is not an insult to horoscopes, it is a statement about evidence.

The instruments: MBTI

The Myers-Briggs Type Indicator sorts a person into one of sixteen types. Its scientific problems are structural: human traits are continuous, most people score near the middle of each scale, but the MBTI cuts each scale in half and assigns a letter, so a small change of mood flips the type. Pittenger's reviews found that between 39% and 76% of people receive a different type when they retake the test after as little as five weeks. In 1991 a National Academy of Sciences committee concluded there was not sufficient well-designed research to justify using the MBTI even in career counselling.

The decisive witness, though, is the publisher. The Myers-Briggs Company itself states that the MBTI is not designed to predict job performance and should not be used in selection. When the company that sells a test tells you not to hire with it, believe the company that sells the test.

The instruments: the Big Five — and the one that is free

The Big Five is the research standard: five dimensions, of which conscientiousness is the one that predicts performance across every occupation studied — at .22 in Barrick and Mount's 117-study meta-analysis, revised down to .19 by Sackett and colleagues in 2022. Real, general, and modest. The commercial flagship, the NEO-PI-R, runs 240 items and up to an hour — fine for executives, fatal for a phone funnel.

The same science exists in public-domain form: the International Personality Item Pool — IPIP. Its items and scales can be copied, edited, translated or used for any purpose without permission and without a fee; the IPIP-50 takes about seven minutes. One caution: applicants fake. Birkeland's 2006 meta-analysis of 33 studies found applicants score higher than non-applicants on conscientiousness (d = .45) and emotional stability (d = .44), and sales applicants distort most on exactly the traits the job rewards. So a Big Five score in a hiring funnel is a sorting signal and a source of interview questions — not a verdict, and treating it as one is the most common misuse in small-company hiring.

The instruments: Hogan

The Hogan suite is the serious end of the commercial market: the HPI for everyday bright-side behaviour, the HDS for the derailers that appear under stress, the MVPI for motives and values — forty-five to sixty minutes for all three, quote-priced, with practitioner packages reported around $1,150 to $1,400 per person including debriefing. Hogan instruments have a genuine research literature behind them. The headline return-on-investment numbers, however, are published by Hogan and its distributors, and should be read as vendor claims, not independent findings. A real instrument with a marketing department: respect the first, discount the second.

The instruments: Predictive Index

Predictive Index pairs a six-minute behavioural checklist — four drives that resolve into seventeen reference profiles — with a twelve-minute cognitive test of fifty questions. The cognitive half stands on the general evidence for mental-ability testing; the behavioural half's headline claims are the vendor's. The catch for a small company is the business model: an annual subscription by headcount, reported from around $10,000 a year, with mandatory training — an enterprise tool solving a sorting problem that a free stack covers.

The instruments: CliftonStrengths

Gallup's CliftonStrengths ranks a person's 34 talent themes — what they do best — for $24.99 for the top five or $59.99 for the full set. It measures only strengths: no derailers, no weaknesses, nothing about how a person behaves on a bad day in the rain. It was designed for development and is not a selection-validated instrument. Using it to screen hires is using a map of someone's peaks to predict their behaviour in a valley.

The instruments: Caliper

The Caliper Profile measures 22 personality traits plus abstract reasoning in about sixty minutes, at around $265 per person in one dated UK source, interpretation included. It has a long history in sales selection; its validity figures are published by the vendor. An hour of a candidate's life and a manager-grade price buy numbers you cannot independently check. For a high-volume funnel, the arithmetic fails before the psychology does.

The instruments: SHL

SHL's OPQ32r is a serious psychometric instrument: 32 personality scales in an ipsative most-like-me / least-like-me format, fifteen to twenty-five minutes, usually paired with the Verify range of cognitive tests, all quote-priced for enterprises. The format resists simple faking, but ipsative scoring ranks traits within a person, which complicates comparing one candidate against another — the very thing a hiring funnel exists to do.

The instruments: Wonderlic and the CCAT

The timed cognitive sprint: fifty questions in twelve minutes for the Wonderlic, fifty in fifteen for Criteria's CCAT, of which fewer than 1% of candidates finish. These tests rest on the genuine evidence base for cognitive ability, which predicts training success and supervisor ratings well. But hold that thought for the sales section below, where the same ability turns out to predict actual sales revenue at .04 — and note that a cognitive gate also carries the heaviest legal exposure of any instrument here. Speed of thought is a real trait. It is not the trait that sells photographs on a sidewalk.

The instruments: integrity tests

Integrity tests ask, directly or through personality items, about attitudes to theft, dishonesty and rule-breaking. This is the quiet heavyweight of the whole field. Ones, Viswesvaran and Schmidt's 1993 meta-analysis — 665 validity coefficients, 576,460 people — found operational validity of .41 for job performance and .47 for counterproductive behaviour: theft, discipline problems, absenteeism. A fifty-year update put the corrected validity against workplace deviance at .43. For any role that hands a stranger cash, card payments and other people's trust, this unglamorous category carries the strongest cheap evidence in the article.

The instruments: situational judgment tests

A situational judgment test shows the candidate a scene from the actual job — a hostile stranger, a teammate cutting a corner, a customer who says no three times — and asks what they would do. Meta-analyses put corrected validity between .26 and .34; video-based versions reach .36 against .25 for paper. The design detail that matters: a test built from a real analysis of the job scores .38, while one built on intuition scores .29. A situational test is only as good as the reality it is copied from — which means a company can build a good one out of its own worst days, and nobody else can sell it to them.

The instruments: gamified assessment

Pymetrics — acquired by Harver in 2022 — replaced questions with about a dozen mini-games measuring attention, memory and risk tolerance, with a model trained on fifty to a hundred of each employer's top performers. Arctic Shores uses interactive tasks instead of questions. The format is genuinely hard to fake and candidates finish it — Harver claims 98% completion, a vendor figure. An independent audit of Pymetrics exists (Wilson and colleagues, 2021), but an audit proves a system can be examined, not that it predicts. Independent evidence that game scores forecast job performance still lags the marketing by a wide margin, and a model trained on yesterday's top performers can reproduce yesterday's biases. Promising, watchable, not yet proven.

The evidence: what a century of studies actually says

Here is the league table, twice. Schmidt and Hunter's 1998 meta-analysis of 85 years of research, and Sackett, Zhang, Berry and Lievens's 2022 re-analysis in the Journal of Applied Psychology, which showed the older numbers had been systematically over-corrected and cut most of them by roughly .10 to .20 — while preserving the ranking almost exactly.

Method	Schmidt & Hunter 1998	Sackett et al. 2022
Structured interview	.51	.42
Work sample	.54	.33
Cognitive ability	.51	.31
Job knowledge	—	.40
Biodata	—	.38
Integrity tests	.41	.31
Conscientiousness	.31	.19
Unstructured interview	.38	.19
Years of experience	.18	.07

Read the table the way an investor reads it. The unstructured interview — the friendly chat on which most small companies hire — sits at .19 in the modern analysis, level with a personality score, and years of experience are nearly worthless at .07. The two champions are embarrassingly unglamorous: ask every candidate the same job-related questions and score the answers against a rubric — the structured interview, .42 — and watch the candidate do a slice of the actual job — the work sample, .33. Neither can be bought in a box. Both are disciplines, not products.

Combinations beat singles. Schmidt and Hunter found cognitive ability plus an integrity test reached a multiple correlation of .65, and ability plus a work sample or a structured interview reached .63. One warning about bigger numbers you will meet in sales brochures: a 2016 paper by Schmidt, Oh and Shaffer reported still higher figures — up to .78 — and vendors quote them constantly. That paper was a working paper. It was never peer-reviewed or formally published. When a salesman shows you .78, he is showing you a draft.

The largest field test of the whole question is not a laboratory study. Hoffman, Kahn and Li, working through the National Bureau of Economic Research, followed fifteen firms, 131 locations and roughly 300,000 hires into low-skill service jobs. Using a validated test score instead of manager discretion improved how long hires stayed by about 15% — green-scored hires stayed longer than yellow, yellow longer than red. And the detail every founder should memorise: a one-standard-deviation increase in how often a manager overrode the test was associated with about 5% shorter tenure for that manager's hires. The managers who trusted their gut over the score were not exercising superior judgment. They were importing bias, and the data billed them for it.

The sales twist: the register disagrees with the interview

Everything above is the general evidence, for jobs in general. Sales is not a job in general, and the sales-specific research changes the answer.

Vinchur, Schippmann, Switzer and Roth's 1998 meta-analysis of salesperson performance split the criteria that most studies blur together: what supervisors rate, and what the register records. Against supervisor ratings, general mental ability looks fine at .40. Against objective sales — money actually sold — it collapses to .04. Nearly nothing. What predicts the money is Achievement, the drive facet of conscientiousness — planful, persistent, self-propelled — at .41 against objective sales, and Potency, the assertive, energetic edge of extraversion, at .26. Biodata inventories reached .52 against ratings and sales-ability inventories .45. (An earlier meta-analysis, Hunter and Hunter 1984, reported .61 for cognitive ability in sales; the difference is the criterion — ratings and training scores, mostly, versus revenue. Believe the split.)

And extraversion itself refuses to behave. Adam Grant's 2013 study of 340 outbound call-centre representatives, published in Psychological Science, tracked revenue for three months and found an inverted U: mid-range ambiverts averaged $16,393 in revenue, against $13,227 for introverts and $12,401 for strong extraverts — the ambiverts outsold the strong extraverts by 32%, because they were assertive enough to close and quiet enough to listen. It is a single study, in a single company, never directly replicated, and we label it as such. But it is strong enough to kill one specific piece of folk wisdom: never hire for maximum extraversion. The loudest man on the corner is not your best seller. The one who can push and then shut up and hear the customer is.

Put together, the sales evidence says: measure drive, measure social potency, keep both in the middle-to-high range rather than at the extreme, and do not build your gate out of raw intelligence. Intelligence is what a seller uses to explain his numbers afterwards. Drive is what made them.

The success stories: companies that tested, and measured it

Google is the rarest case in this field: a company that ran the with-versus-without experiment on itself. Its people-analytics team correlated years of interview scores with later performance and found that its famous brainteaser questions predicted nothing — Laszlo Bock, its head of People Operations, called them a complete waste of time — and that unstructured interview scores had zero relationship to how people performed. What survived the audit: work samples, cognitive ability, and structured interviews with rubrics. Bock's figures, company-reported in his book Work Rules!, credit work samples with 29% of the variance explained, general mental ability and structured interviews with 26% each. Google's Rule of Four — four interviews predict the hiring decision with 86% confidence — cut its time-to-hire from roughly six months to about forty-five days.

The largest validation ever run belongs to the US military. The Army's Project A built hands-on performance tests for real jobs and checked the ASVAB aptitude battery against them: the correlation between test composites and core technical proficiency reached .69 for soldiers in their second tour, and across 23 military jobs the median corrected correlation was .38 — modest, the National Academies review concluded, and enough to justify the screen. Decades, millions of recruits, one lesson: a validated test, followed, beats the absence of one.

The vendor stories are louder, so read them with the label on. Unilever replaced CV screening for its graduate funnel with Pymetrics games and HireVue video interviews; the programme reports time-to-hire cut from four months to four weeks, 50,000 to 70,000 hours saved — the sources differ, so we print the range — more than £1 million a year saved, 96% candidate completion and a 16% rise in diversity. All of those numbers come from the vendor's and the company's own case materials, not an independent audit — and the same programme quietly dropped facial-expression analysis in 2019 when scientists objected, which is what a serious testing operation does when part of it fails review. Hilton reports its time-to-fill falling from six weeks, 42 days, to five days after introducing standardised video assessment — the company's own account. Xerox handed its call-centre hiring to an Evolv battery of personality, cognitive and situational measures and reports attrition down 20% in the pilot — along with the finding that embarrassed a generation of hiring managers: prior call-centre experience, the thing they had been selecting on for years, had no impact on retention or productivity at all. A companion Evolv case at Harte-Hanks reports score-selected hires missing 29% fewer hours and handling calls 15% faster. Vendor data, reported through the press; the direction matches the independent evidence, and the independent evidence is exactly what suggests discounting vendor magnitudes by half or more.

The failure stories: brilliance without a filter

Now the other side of the ledger — told honestly, because the honest version is more useful. None of the famous disasters is a simple story of a company that skipped a personality test. They are stories of selection weighted entirely toward brilliance and charm, inside systems that never measured integrity and could not hear a warning when one was sounded.

Enron hired, in the phrase from McLean and Elkind's account, "guys with spikes" — singular, dazzling talent, regardless of other flaws. Integrity was not a selection criterion; the internal rank-and-yank system forced out the bottom 15% a year and selected for aggression, while anyone who raised doubts was treated as not getting it. The end, in December 2001: roughly $74 billion in shareholder losses, about 20,000 jobs, and employee pension losses estimated at $1.2 to $2 billion, with the auditor Arthur Andersen destroyed in the blast. In Schmidt and Hunter's terms, Enron used the .51 predictor — raw ability — and ignored the .41 one — integrity — and thereby selected for people clever enough to commit sophisticated fraud.

Wells Fargo is the case that disciplines this whole article. Between 2011 and 2016, under a cross-selling target of eight products per customer, employees opened roughly 1.5 million unauthorised deposit accounts and more than 500,000 unauthorised credit cards — later estimates reach about 3.5 million accounts. Around 5,300 mostly junior employees were fired; the fines began at $185 million in 2016 and reached a $3 billion settlement in 2020, with a Federal Reserve asset cap that lasted seven years. Now the honest part: this was an incentives failure before it was a screening failure. A personality test at the door would not have prevented it, because the system inside paid for misconduct and punished refusal. What Wells Fargo lacked was any measurement of integrity anywhere in the system — and a structure in which a warning could stop anything. A test is not a conscience. It is, at best, one instrument on a panel that Wells Fargo flew without.

Uber's 2017 crisis followed a review of 215 complaints; 20 employees were fired, the Holder investigation told the company to rewrite values that had licensed bad behaviour — "Always Be Hustlin'", "Meritocracy and Toe-Stepping" — and the founder-CEO resigned. Inside the story sits a pure screening failure: an executive was asked to leave after it emerged he had not disclosed a prior harassment allegation at his previous employer. The reference check that would have caught it was never run.

Theranos is included with a warning label: it is a governance and verification failure, not a psychometric one. A board of famous names performed no technical due diligence; employees were hired into compartments where nobody could see the whole machine; questioners were fired. Nobody in the chain ran the elementary check — on the technology, or on the person selling it. The company went from a valuation of about $9 billion to zero, with $452 million in restitution ordered.

The courts supply the failures without celebrity. When a technician with a falsified, never-verified work history — prior terminations for harassing co-workers and forgery — murdered a customer, a Texas jury returned $375 million in compensatory and $7 billion in punitive damages against Charter Communications; judgment was entered at $1.147 billion before a confidential settlement. A Columbus bar that ran no background checks at all on its security staff took a $27 million wrongful-death verdict and went out of business. Baltimore landlords who handed apartment keys to a hire with violent assault convictions in 2009, 2011 and 2013 took a verdict of more than $21 million. A Florida jury, over drivers hired without adequate record checks, awarded $100 million compensatory and $900 million punitive. Industry compilations — vendor-compiled, so read them as directional — put employer losses in negligent-hiring cases at roughly 75 to 80% of cases tried, with average settlements around $1 million.

The oldest case in the file is Barings Bank, 1995: the dealing-licence check on Nick Leeson actually worked — it surfaced undisclosed County Court judgments for personal debts — and the bank shipped him to Singapore anyway, where he ran both the trading floor and the back office and lost £827 million. The bank, 233 years old, was sold for £1. A result nobody with power must obey is not a screen. It is documentation that you knew.

Even the cheerful version is expensive. Tony Hsieh of Zappos — a company famous for culture — estimated that bad hires, and the bad hires those hires went on to make, cost Zappos well over $100 million: a chief executive's own unaudited estimate, and an admission against interest, which is why it is worth printing.

The rebels: the companies that refuse tests entirely

The strongest counterargument to everything above is not an argument. It is a set of companies that use no psychological tests at all — and win. Look closely, though: none of them selects on nothing. Each one replaced the questionnaire, which infers traits, with a situation, which exhibits behaviour.

Automattic, the company behind WordPress.com, hires by paid audition. Finalists work a real contract trial — typically two to eight weeks, part-time, at a flat $25 an hour whatever the role, chief financial officer included — inside a real team, on real work. Matt Mullenweg's account in the Harvard Business Review reports that the company hires roughly 40% of the people it trials and that turnover is, in his words, ridiculously low. Both figures are company-reported. The logic is not: a trial shows the candidate the job too, which kills the mismatch that causes most early quits. Basecamp runs the same play smaller — applications read by humans, cover letters weighted at or above résumés, finalists paid for a week of real project work — with two decades of profitability at around fifty people to show for it: founder-reported, from a company too small for a validity study, and honest readers should note both halves of that sentence.

Zappos priced commitment instead of measuring it. Culture fit carried 50% of its hiring decision, and after four weeks of training every new hire met The Offer: $2,000 to quit on the spot. Only about 2 to 3% took it — company-reported — which Hsieh read as proof the funnel was already working. Amazon imported the device into its warehouses in 2014 as "Please Don't Take This Offer", $2,000 rising to $5,000 with service; Riot Games offered new hires up to $25,000 to leave inside sixty days. And there is peer-reviewed support for the mechanism's second effect: a study in the Journal of Management Accounting Research found that pay-to-quit offers increase the effort of the people who stay — declining the money is an act of recommitment, not a default.

Southwest Airlines built the same insight into a slogan: hire for attitude, train for skill. Its screen is human and structured — behavioural interviews and group exercises hunting three named dispositions, a Warrior Spirit, a Servant's Heart and a Fun-LUVing Attitude — behind Herb Kelleher's 1994 line: we can train you to work a gate; we can't train you to enjoy people. There is no controlled validity study behind Southwest's method. There is a fifty-year business case instead, and the reader can weigh a coefficient against half a century.

And the rebels have the law on their side, because the law was written by a test that failed. In Griggs v. Duke Power in 1971, the US Supreme Court looked at a plant where 58% of white applicants and 6% of Black applicants passed the Wonderlic and Bennett tests the company required — requirements never shown to predict performance in the plant's jobs — and ruled unanimously that a selection device which disproportionately excludes a protected group is unlawful unless it is proven related to the job. Tests, the Court said, must measure the person for the job, not the person in the abstract. Since 1971, a test in America is not evidence because it exists. It is evidence when it predicts.

Notice where the rebels and the scientists land. The best test-free funnel in this section — work sample plus structured interview — is the same pair that tops Sackett's 2022 table at .42 and .33. Google, the most data-rich employer on earth, audited itself into exactly that pair. The rebels are not anti-science. The best of them are what happens when the science is actually read.

The law in brief

American hiring law asks one question of every instrument in this article: is it job-related, and can you prove it? The Uniform Guidelines on Employee Selection Procedures, in force since 1978, say that any selection procedure producing adverse impact is discriminatory unless validated — by criterion evidence, by content evidence, or by construct evidence — against a documented analysis of the actual job.

The tripwire is the four-fifths rule: if the selection rate for any racial, sex or ethnic group falls below 80% of the highest group's rate, regulators generally regard it as evidence of adverse impact. It is a rule of thumb, a screening device for the screening devices — not a verdict — but a small company ignores it at the price of becoming a case study in somebody else's article.

The Americans with Disabilities Act draws a second line: no medical examination before a conditional offer. A psychological test designed to reveal a mental impairment can count as a medical examination — the Seventh Circuit ruled exactly that about the MMPI in Karraker v. Rent-A-Center in 2005, and the scores were ordered destroyed. Target paid $2.8 million in a 2015 settlement involving an assessment treated as a pre-offer medical exam. Ordinary trait and honesty measures are generally not medical exams, but the rule for a small employer is simple: no clinical items, no mental-health screening, before an offer. Ever.

The newest layer governs machines. New York City's Local Law 144, enforced since July 2023, requires an independent bias audit within a year before an automated employment decision tool is used on city residents, a public summary of the results, and notice to candidates at least ten business days ahead — at penalties of $500 for a first violation and $500 to $1,500 per day for later ones. Illinois requires consent before AI analyses a video interview, and from January 2026 treats discriminatory AI use in hiring as a civil-rights violation. The EEOC's first AI-hiring settlement, iTutorGroup in 2023, cost $365,000 for software that simply auto-rejected older applicants; and in Mobley v. Workday, a federal court held in 2024 that an AI vendor can be liable as the employers' agent, with a nationwide age-discrimination collective certified in 2025 and the merits still undecided. The principle transfers to every employer, however small: "the algorithm decided" is not a defence. Neither is "the test decided". A person decided to use it.

The verdict: how Magic Royal Flash hires

Here is what all of this buys a street photography company with a phone funnel and no research department. Not a battery. A stack of four cheap, honest layers — the layers the evidence, the successes, the failures and the rebels all point at together.

First, a realistic job preview: two minutes of the truth about the job — the weather, the rejection rate, the lone working, how the pay actually works — before any test at all. McDonald's UK built exactly this, paired with a situational test, into its funnel and reported crew applications down 35% and management applications down 50%, by design: the wrong people removing themselves is the cheapest screening ever invented, because a candidate who quits at the landing page costs nothing and a candidate who quits on day forty costs the full bad-hire bill.

Second, a short situational test built from our own street: six to ten scenarios copied from real days on the square — the third no in a row, the hostile stranger, the teammate skimming — scored against the answers of our own best operators. The research says a job-analysis-built situational test reaches .38, an intuition-built one .29. Ours is built from the job because it is the job.

Third, the free one. The IPIP Big Five, fifty items, about seven minutes, public domain — used to sort the queue and to arm the interviewer, never to reject a human being by itself. We read it the way the sales evidence instructs: achievement and drive first, emotional stability second, extraversion potent but not maximal — and a low score does not close a door; it writes the questions the interviewer must ask. A personality score is a flashlight, not a judge.

Fourth, the gate that cannot be faked: a live role-play. Sixty seconds of pitch to a stranger, one hostile objection, one rejection and the recovery from it — the candidate doing a slice of the actual work, scored on a written rubric, followed by a structured interview: the same questions, in the same order, for everyone. The two methods at the top of both meta-analyses, .42 and .33 in the modern table, .51 and .54 in the classic one. Total candidate effort before the interview: under twenty minutes, on a phone.

And then the part almost nobody does, which is the real advantage: every score — situational, personality, role-play — is recorded against what happens next. Who is still standing at ninety days. Who sells. Who causes an incident. Within a few hundred hires, the funnel stops borrowing other companies' science and starts running its own validation study, which is the strongest legal position available and the only way to know whether our cut-scores deserve to exist. And once a measure has proven itself in our own numbers, our interviewers do not override it casually. Hoffman, Kahn and Li measured where overrides lead.

The doctrine in one line

We do not hire the résumé, the charm, or the four-letter type. We hire the evidence — a person watched doing the work, measured against the work — and then we keep our side of the bargain: strict standards, the best people kept, and pay that arrives on time, without exception. A funnel that honest does not need to beg people to stay, and a test that honest does not need to pretend it can see the future. It just has to see the person in front of it, clearly, once — before the street does it for us, at street prices.