Welfare secretaries to people operations. The Drucker debate, the Schmidt-Hunter findings, the structured-interview consensus, and the HR-tech stack that runs the modern firm.
Human resources is the corporate function that hires, pays, develops, manages, and (sometimes) fires the people who do the work. It has been called welfare, personnel, industrial relations, human resources, and most recently people operations — each renaming reflecting a shift in how firms understand their relationship to labour.
The discipline is older than its current name suggests. Welfare secretaries appeared in American factories in the 1890s. Frederick W. Taylor's Principles of Scientific Management (1911) put work design on a quantitative footing. The first PhDs in industrial psychology arrived in the 1910s. The American Management Association was founded in 1923.
This deck covers the historical sequence (welfare to personnel to HR to people ops), the central intellectual debates (Drucker on management, Pfeffer on evidence, Cappelli on the talent supply chain), the empirical findings that have held up (Schmidt-Hunter on selection validity, Locke and Latham on goal-setting), and the contemporary HR-tech stack.
The first paid HR-style role in American industry was the welfare secretary, employed from the 1890s onward by paternalistic firms — National Cash Register, H.J. Heinz, Filene's, Pullman — to look after the well-being of (largely female, largely immigrant) workers. The job mixed social work, employee counselling, hygiene inspection, and recreational programming.
The motivations were a blend of progressive-era concern, productivity calculus, and pre-emption of unionisation. The model peaked around 1915. By 1919 there were about 1,500 welfare secretaries in American firms. The Boston-based National Civic Federation and YMCA ran training courses.
The welfare-secretary role faded after WWI, replaced by the more bureaucratic employment manager — a title that emphasised hiring, record-keeping, and labour-relations rather than worker welfare. The 1916 founding of the Employment Managers' Association (later renamed many times; today's SHRM is a descendant) marked the shift.
By 1920 the language was already shifting again — to personnel, the dominant term for the next half-century.
From roughly 1920 to 1980, the function was called personnel. The personnel department managed hiring, payroll, benefits administration, training records, union relations, and grievances. The work was administrative, paper-based, and largely reactive.
Three parallel intellectual currents shaped it. Scientific management (Taylor, the Gilbreths) provided the time-and-motion methodology that made job analysis possible. Industrial psychology (Hugo Münsterberg, 1913; later the Army Alpha and Beta tests of WWI) gave personnel its testing tools. The Hawthorne studies (Elton Mayo and the Western Electric experiments, 1924–32) introduced the human-relations counter-current — the discovery that workers responded to attention and group dynamics, not only to wages and conditions.
Post-WWII, the personnel function grew with the firm. The 1950s and 60s saw the rise of large corporate benefits programmes (pension plans, health insurance), the codification of equal-employment law (Title VII, 1964; ADEA, 1967), and the expansion of in-house training departments. Personnel was an administrative support function, paid less than line management, and rarely strategic.
The intellectual ferment of the 1970s — Galbraith on organisation design, Mintzberg on managerial work, the rise of OB as an academic field — set up the rebranding that came next.
The phrase "human resources" was popularised by Raymond E. Miles in a 1965 Harvard Business Review article ("Human Relations or Human Resources?"). The argument: the human-relations school treated people as needing to be kept happy; the human-resources school treated people as untapped capability whose productivity could be raised by participation, training, and design.
The rebrand spread through the late 1970s and 1980s. The American Society for Personnel Administration changed its name in 1989 to the Society for Human Resource Management (SHRM). Corporate titles changed: VP of Personnel became VP of Human Resources, and eventually Chief Human Resources Officer (CHRO).
The intellectual scaffolding came from Dave Ulrich's Human Resource Champions (1997), which proposed the four-role model: strategic partner, administrative expert, employee champion, change agent. Ulrich's work pushed HR to position itself as a business partner rather than a back-office function — the so-called "HR business partner" model that became standard in Fortune 500 firms by 2010.
The critics: Keith Hammonds's Why We Hate HR (Fast Company, 2005) and a recurring complaint in the McKinsey/HBR literature that HR remained tactical despite the rebrand. The function still struggles with the same identity tension.
The current rebrand. Coined at Google under Laszlo Bock (SVP of People Operations 2006–2016) and popularised by his book Work Rules! (2015). The argument: the function should be run with the same data discipline as engineering — A/B-test interview structures, measure the lift from training, instrument every people decision.
Google's People Analytics team — Prasad Setty, then John Lin — produced internal studies that became part of the public record: Project Oxygen (2008–10) on what makes a great manager, Project Aristotle (2012–15) on what makes a great team. Aristotle's headline finding — that psychological safety (Edmondson's construct, see OB deck) was the strongest predictor of team effectiveness — became one of the most cited HR findings of the 2010s.
The "people ops" branding spread through tech (Facebook/Meta, Airbnb, Stripe, Spotify, Uber) and from there into start-ups and adjacent industries. By 2020 most large tech companies had a Chief People Officer rather than a CHRO, and "people analytics" was a recognised speciality with its own conferences (Wharton's People Analytics Conference, founded 2013) and master's programmes.
Whether "people ops" is substantively different from HR or merely a rebrand for technical workforces is debated. The data-driven discipline is real; the work of policy, payroll, and compliance is unchanged.
Peter Drucker (1909–2005) defined modern management. His The Practice of Management (1954), The Effective Executive (1967), and Management: Tasks, Responsibilities, Practices (1973) were the field's founding texts.
Drucker on HR was deeply ambivalent. In 1954 he wrote that personnel administration was "partly a file clerk's job, partly a housekeeping job, partly a social worker's job, and partly fire-fighting" — and that the personnel department had "constitutionally" failed to grasp the strategic potential of its remit.
His positive contribution: Management by Objectives (MBO) in the same 1954 book. The idea — managers and workers jointly set specific, measurable objectives, then evaluate performance against them — became the basis of every subsequent goal-setting and performance-management system. Locke and Latham's later experimental research (see Chapter X) confirmed MBO's intuition empirically.
Drucker's deeper claim: knowledge workers would be the dominant labour category of the 21st century, and managing them required a different model than managing manual labour. The HR functions of the 1950s — designed for blue-collar workforces — were structurally unsuited. Most of the 70 years since have been HR slowly adjusting to that prediction.
The most consequential empirical finding in HR research. Frank Schmidt and John Hunter's 1998 meta-analysis ("The Validity and Utility of Selection Methods in Personnel Psychology," Psychological Bulletin) synthesised 85 years of validation studies on hiring methods.
The headline ranking, by predictive validity for job performance:
Then: integrity tests (.41), conscientiousness tests (.31), unstructured interviews (.38 but with much higher variance), reference checks (.26), years of education (.10), graphology (.02 — i.e. zero).
The strong claim from the data: structured interviews, work samples, and tests of general mental ability are the best-validated selection methods, and combinations of them outperform any single method. Unstructured interviews — the dominant practice in most firms — perform substantially worse than the alternatives.
A 2016 update (Sackett et al., Journal of Applied Psychology) revised some coefficients downward — GMA validity is closer to .30 in modern data, structured interviews still ~.45 — but the relative ranking is robust. The implication: most firms hire badly, by methods their own HR departments could in principle replace.
What "structured" means: the same questions in the same order, asked by trained interviewers, scored on pre-defined rubrics, with results combined arithmetically rather than discussed. The key features:
Job analysis-based questions. Questions derived from a competency model for the role, not from the interviewer's intuition.
Behavioural and situational formats. "Tell me about a time when…" (past behaviour, the STAR-method format) or "What would you do if…" (situational judgement). Both outperform open-ended biographical questions.
Standardised probes. Trained interviewers are allowed to follow up — but only with pre-approved probes. No tangents, no rapport-driven detours.
Independent ratings, mechanical combination. Each interviewer scores independently on the rubric. Scores are averaged or summed by formula — not discussed in a "huddle." Robyn Dawes's research on clinical vs statistical prediction (1979) shows the formula consistently beats the discussion.
Google's interview process, as described by Bock, runs four interviews per candidate (the marginal value of the fifth interview is essentially zero by their analysis), each structured around three or four core competencies, with calibrated scoring rubrics and a hiring committee that reviews the packets without re-interviewing.
Most companies don't do this. The reasons are cultural: managers prefer "chemistry" to checklists; structured interviewing feels mechanical; the interviewer's confidence in their own intuitive judgement is unreasonable but persistent.
The annual performance review is the most-disliked HR ritual, and the most-studied. The findings are not flattering.
Edwin Locke and Gary Latham's goal-setting theory (1968 onward; A Theory of Goal Setting and Task Performance, 1990) is the strongest empirical pillar. Specific, difficult goals reliably outperform "do your best" goals across hundreds of studies. Goals must be accepted, the individual must have ability, and feedback must be available — but given those conditions, the effect is robust and large (d ≈ 0.5 to 0.8).
The annual review, however, is a poor implementation. The 2010s saw a wave of large firms — Deloitte (2015), Adobe (2012), GE (2015), Accenture (2015) — abandoning forced ranking and annual numerical ratings in favour of more frequent, conversation-based check-ins. The driver was internal data showing low correlation between rating and actual performance, plus the Wisdom-of-Crowds problem that ratings reflected rater idiosyncrasy more than ratee behaviour.
Forced ranking ("rank-and-yank") — Jack Welch's GE practice of culling the bottom 10% — was the dominant fashion of the 1980s-90s. Microsoft's stack ranking (abandoned 2013) and the broader trend away from these systems reflect a recognition that they damaged collaboration without improving overall performance.
The state of the art (Cappelli & Tavis, HBR 2016; Pulakos and O'Leary): frequent feedback, separated developmental and evaluative conversations, calibrated team-level discussions to control rater bias.
Pay is the largest line item in most firms' P&L. The HR sub-discipline that designs it (compensation, or "total rewards") draws on labour economics, psychology, tax law, and accounting.
The classic framework: internal equity (pay people similarly for similar work within the firm), external competitiveness (benchmark against the market), individual incentive (vary pay by performance and tenure). Most pay structures balance the three imperfectly.
Edward Lawler's research (1971 onward) established the empirical case that pay-for-performance works for jobs with measurable individual output and weakly or perversely for jobs with team-based, qualitative, or innovation-driven output. Sales commissions, piece rates, and executive equity grants belong to the first category. Most knowledge work belongs to the second.
The pay structure of a modern firm has at least four layers: base salary (fixed; benchmarked to market), annual bonus (variable; tied to firm and individual performance), equity (long-term incentive; the dominant form for senior roles in tech), and benefits (health, retirement, stock purchase plans, learning stipends, parental leave).
The broadbanding trend of the 1990s collapsed dozens of pay grades into a few wide bands; the more recent trend toward pay transparency (Buffer's open salaries, 2013; Colorado's 2021 pay-transparency law, EU pay-transparency directive 2023) is reversing decades of confidentiality.
Peter Cappelli (Wharton, founder of the Center for Human Resources) has been the most influential academic voice on HR strategy of the 2000s and 2010s. His Talent on Demand (2008) reframed workforce planning as a supply-chain problem.
The argument: the post-1970s collapse of the corporate-loyalty contract (lifetime employment, internal labour markets, predictable promotion ladders) meant that firms could no longer rely on internal pipelines for senior talent. They had to manage workforce inflows and outflows actively, with the same uncertainty-quantification discipline that operations managers brought to inventory.
Practical implications: build vs buy decisions for senior roles (do we develop internally or hire externally? — and at what cost?); portfolio of bets on uncertain skill demand; flexibility through contingent work (contractors, consultants, fractional executives); retention as the limiting factor, not selection.
Cappelli's 2014 book Why Good People Can't Get Jobs argued that the "skills gap" complaint of US employers was largely a refusal to train — firms had outsourced workforce development to candidates and educational institutions, then complained when the supply was unfit. The argument that firms should re-invest in apprenticeship, structured on-the-job training, and tuition support has only intensified post-pandemic.
Jeffrey Pfeffer (Stanford GSB) and Robert Sutton's Hard Facts, Dangerous Half-Truths, and Total Nonsense (2006) launched the evidence-based management movement. The complaint: HR practice was driven by fads (TQM, Six Sigma in HR contexts, "best places to work" rankings, leadership-style flavours of the month) without serious accounting for what the underlying research did and didn't support.
Pfeffer's earlier work — The Human Equation (1998) — synthesised the evidence on "high-performance work practices": employment security, selective hiring, self-managed teams, high contingent compensation, training, reduced status differentials, information sharing. Firms that combined these practices showed measurable performance advantages, with effect sizes in the 15–30% range across multiple studies.
The movement borrows from evidence-based medicine. Its premise: managers should know the empirical support for the practices they implement, weigh that evidence against context, and not confuse anecdote, ideology, or charisma with proof.
Critics argue HR is an applied art with too few RCTs to be evidence-based in the medical sense. The defence: meta-analyses (Schmidt-Hunter, Locke-Latham, Tett & Meyer on commitment) cover enough territory to discipline the worst nonsense, and the meta-analytic literature is much underused in practice.
Employee engagement is the dominant HR metric of the 2000s and 2010s. The concept — William Kahn's 1990 paper "Psychological Conditions of Personal Engagement and Disengagement at Work" — was operationalised by Gallup's Q12 survey (twelve questions correlating with business outcomes).
Gallup's findings (now from millions of respondents): engaged employees show 21% higher productivity, 22% higher profitability, 41% lower absenteeism, and 59% lower turnover than disengaged employees. The base rate of full engagement is low — Gallup's State of the Global Workplace consistently puts it at ~20–35%.
The Q12 questions cluster around: knowing what's expected, having the resources to do it, doing what one does best, recent recognition, feeling cared about, having a development conversation. None of them is about pay or benefits — the predictors are managerial, relational, and developmental.
Critics (Schaufeli, Bakker, others in the academic engagement literature) note that Gallup's instrument is proprietary and the construct overlaps substantially with job satisfaction, organisational commitment, and burnout's inverse. The academic JD-R model (Job Demands-Resources, Bakker & Demerouti 2007) provides a more analytically tractable framework but has less corporate uptake.
Workforce diversity programs date in their modern form to the post-Civil Rights Act 1960s, with the establishment of the EEOC and affirmative-action requirements for federal contractors (Executive Order 11246, 1965). The discipline expanded through the 1980s and 1990s under "managing diversity" branding, then again from ~2010 under the DEI label.
The empirical record on diversity training is mixed. Frank Dobbin and Alexandra Kalev's research (Harvard, 2006 onward) using federal EEO-1 data on ~800 firms found that mandatory diversity training was uncorrelated or negatively correlated with subsequent representation gains. What worked: mentoring programs, diversity task forces with line-management accountability, college recruitment from HBCUs and women's colleges, and managerial accountability for diversity outcomes.
The 2020s post-Floyd surge in DEI spending — Ford, Meta, JPMorgan and others made multi-billion-dollar commitments — was followed by a 2023–25 pullback under political and legal pressure (the SFFA v. Harvard Supreme Court ruling on race-conscious admissions; state-level anti-DEI legislation; corporate retreats by Walmart, Boeing, Toyota, Ford).
The remaining empirical question is what works. The Dobbin-Kalev evidence suggests that structural interventions (recruitment pipelines, mentoring, accountability) outperform attitudinal ones (unconscious-bias training).
The contemporary HR function runs on software. The major categories:
Core HRIS / HCM (the system of record): Workday (founded 2005, dominant in upper-mid and enterprise), SAP SuccessFactors, Oracle HCM Cloud, ADP, BambooHR (mid-market), Rippling, Gusto.
Applicant Tracking Systems: Greenhouse (the tech-industry default), Lever, Workday Recruiting, iCIMS, Ashby (newer), SmartRecruiters.
Performance and goals: Lattice, Culture Amp, 15Five, BetterWorks. Increasingly absorbed into HCM platforms.
Learning Management Systems: Cornerstone, Workday Learning, Docebo, LinkedIn Learning, Coursera for Business.
Compensation: Pave, Figures, Carta (for equity), Aeqium.
People analytics: Visier, ChartHop, OneModel, Tableau/PowerBI built on top of the HCM data warehouse.
Engagement and survey: Culture Amp, Glint (Microsoft), Peakon (Workday).
The total addressable market is large — Gartner estimates worldwide HR-tech spending at ~$30B annually as of 2024. The integration problem is hard: most HR data lives in 8-12 systems that don't natively talk to each other, and the people-analytics function exists partly to plumb the integrations.
The dominant enterprise HCM platform was founded in 2005 by Dave Duffield and Aneel Bhusri, both former PeopleSoft executives, after Oracle's hostile acquisition of PeopleSoft. Their thesis: the next generation of HR software would be cloud-native, multi-tenant, and unified with finance — none of which the on-premises ERP incumbents (SAP, Oracle) could deliver without rebuilding.
Workday went public in 2012, hit $1B in revenue in 2016, and was at ~$8.4B in FY2025 with a market cap of ~$70B. It is the system of record for a majority of large US enterprises, including most of the Fortune 500's people functions.
The product is conservative by tech standards — long upgrade cycles, limited customisation, expensive implementations (a Workday HCM rollout for a 50,000-person firm typically costs $20–80M and takes 18–36 months). The bet is on stability and integration depth rather than on UX innovation. The bet has paid off: Workday is a near-default purchase at the Fortune 1000 tier.
The competitive frontier in 2024–25 is AI-augmented HR work — résumé screening, performance-summary drafting, internal-mobility matching, copilot interfaces. Workday, SuccessFactors, Oracle, and a wave of new entrants (Eightfold, Gloat, Phenom) are competing on this dimension.
The hiring function — sometimes called talent acquisition — is HR's most visible operational machine. The flow:
Sourcing: identifying candidates. LinkedIn (founded 2002; Microsoft acquisition 2016) is the dominant inbound and outbound sourcing platform. Indeed (founded 2004; Recruit Holdings acquisition 2012) is the largest job board. Specialised tools (Gem, hireEZ, SeekOut) layer search-and-CRM on LinkedIn.
Screening: filtering applications. Modern ATS systems use rule-based filters and (increasingly) ML models. The 2018 Reuters report on Amazon's scrapped resume-screening AI — which had learned to penalise the word "women's" — is the canonical case study in why this is hard.
Interviewing: typically 4–6 rounds for senior tech roles, 2–4 for mid-level. The Schmidt-Hunter findings argue for fewer, more structured rounds; most firms run too many, too unstructured.
Offer and close: compensation negotiation, reference checks, background verification. The decline trade-off — every additional day to offer correlates with higher decline rates — is a real operational metric.
Onboarding: first 90 days. The empirical literature (Bauer; Saks; Klein) shows substantial impact of structured onboarding on first-year retention and productivity ramp.
The cost of replacing an employee runs roughly 50–200% of annual salary, depending on role complexity (SHRM, Center for American Progress estimates). For senior technical and managerial roles, the cost is closer to 200%; for hourly retail and food service, closer to 30–50%.
The empirical predictors of voluntary turnover, in roughly descending order of strength:
Manager quality. The cliché "people leave managers, not companies" has substantial empirical backing (Buckingham & Coffman, First Break All the Rules, 1999; Gallup Q12 data; Google's Project Oxygen). The manager is the strongest single environmental predictor of intent to stay.
Pay relative to market. Especially in tight labour markets and especially for technical roles. Compa-ratio (employee pay ÷ market median for role) below 0.95 substantially raises turnover risk.
Growth opportunity. Internal mobility, learning, advancement. Employees who make an internal move show ~30% lower 12-month turnover (LinkedIn Workplace Learning Reports).
Job-role fit. Selection mistakes are retention mistakes — most "bad culture fit" claims are mis-selection.
Burnout and workload. Demerouti's JD-R model, Maslach's burnout inventory: chronic high demands without adequate resources predict exit.
The intervention design follows: improve manager quality (training, feedback, accountability), monitor pay competitiveness (compa-ratio dashboards), build internal mobility, fix selection, monitor workload.
The L&D sub-function manages employee skill development. The 2024 LinkedIn Workplace Learning Report puts global L&D spend at ~$370B annually (corporate + government).
The empirical literature is sobering. The famous 70-20-10 model (Lombardo & Eichinger, attributed to Center for Creative Leadership 1996 research) — 70% of development from on-the-job experience, 20% from relationships and feedback, 10% from formal training — is widely cited but methodologically thin. The number is closer to a heuristic than a finding.
What does work, per the meta-analytic literature:
Spaced practice over massed practice. The Hermann Ebbinghaus forgetting-curve work (1885) and modern reproductions consistently show that distributed learning beats one-shot training.
Active over passive. Application, problem-solving, and teaching others outperform passive reading or listening.
Manager involvement and follow-up. Training without manager reinforcement decays rapidly. Burke and Hutchins (2007) found pre/during/post-training manager support is the strongest moderator of training transfer.
The current frontier is AI-tutoring (Khan Academy's Khanmigo, Coursera Coach), microlearning (Workday Learning, Degreed), and skills-based career marketplaces (Gloat, Eightfold, Workday Skills Cloud) that match employees to internal projects rather than to courses.
HR sits inside a thick legal envelope. The major US statutes:
Title VII of the Civil Rights Act (1964). The foundation of US employment-discrimination law. Race, colour, religion, sex, national origin. The 2020 Supreme Court ruling in Bostock v. Clayton County extended sex discrimination to sexual orientation and gender identity.
FLSA (Fair Labor Standards Act, 1938). Minimum wage, overtime, child labour. The exempt/non-exempt classification — which determines who gets overtime — is one of HR's most operationally consequential rulebooks.
FMLA (1993). Twelve weeks of unpaid job-protected leave for medical and family reasons.
ADA (Americans with Disabilities Act, 1990). Disability accommodations and non-discrimination.
NLRA (1935). Collective bargaining and union activity. The recent Starbucks, Amazon, and Apple unionisation activity (2021 onward) has revived NLRB enforcement after decades of decline.
ERISA (1974). Pension and benefits administration. The reason your 401(k) plan looks the way it does.
Internationally: the EU's GDPR (2018), Pay Transparency Directive (2023), and AI Act (2024) all impose HR-relevant obligations. State-level US law (California in particular — CCPA, AB 5, FAIR Act) layers on additional complexity.
The compliance burden is real and growing. Most large HR departments now have a dedicated employment-law counsel function alongside HR business partners.
The bleakest part of the function. The 2022–24 tech layoff wave — Meta cut 21,000, Amazon ~27,000, Google 12,000, Microsoft 10,000, Salesforce 8,000 — produced one of the largest synchronised white-collar workforce reductions in US economic history.
The empirical literature on layoffs (Cascio, Pfeffer) is unambiguous: most layoffs do not improve subsequent firm performance. Cascio's 2002 study of 4,000 firms found that "downsizers" had lower 3-year ROA and stock-price performance than peers who held headcount. Pfeffer's Layoffs and Lives argues that layoffs produce persistent damage to the survivors' productivity, engagement, and health.
What firms do anyway: layoffs are forced by capital-market expectations, post-merger integration logic, and the speed of revenue decline relative to cost structure. The internal HR work — selection criteria, severance design, WARN Act compliance (60-day notice for 100+ layoffs at a single site), benefits continuation, outplacement — is tactical and detailed.
The post-layoff "survivor syndrome" is well-documented: reduced trust, reduced organisational commitment, increased withdrawal behaviour, and an immediate 12-month productivity drop in the surviving workforce. The remediation is communication intensity, manager support, and structural reassurance about future stability — none of which firms in financial distress execute well.
The COVID-19 pandemic produced the largest workforce-mode transition in modern history. Nicholas Bloom's ongoing Stanford WFH research tracks the trajectory: pre-2020 ~5% of US paid workdays were remote; April 2020 ~60%; 2024 has stabilised at ~30% (with hybrid the dominant pattern in white-collar work).
The policy question splits roughly four ways: fully remote (GitLab, Automattic, some Stripe teams), remote-first / hybrid by default (Atlassian, Spotify "Work From Anywhere", many post-2020 startups), structured hybrid (3 days office / 2 home — JPMorgan, Amazon's 2025 5-day RTO, Google, Apple), fully in-office (Goldman, much of finance, much of pharma, most retail and manufacturing).
The empirical evidence on productivity is mixed and depends on task type. Bloom's call-centre RCTs (Ctrip, 2013; subsequent studies) found a 13% productivity gain for fully-remote workers on routine output. Knowledge-work studies (Emanuel & Harrington 2021 on customer-service; Yang et al. 2022 on Microsoft) found mild collaboration-network costs offsetting individual-productivity gains. The honest read: hybrid is a coordination problem, not a solved problem.
HR's role: redesigning policy (location, equipment, expense), managing tax and labour-law complexity (multi-state and multi-country employment), reshaping performance management to work asynchronously, and managing the manager training required to lead distributed teams.
The discipline that brought statistical practice to HR. The seminal text is Tom Davenport, Jeanne Harris, and Jeremy Shapiro's 2010 HBR article Competing on Talent Analytics. The flagship example is Google's People Operations.
The maturity ladder, roughly: descriptive (how many people, how many leaving, how diverse), diagnostic (why are they leaving, what predicts engagement), predictive (who is at risk of leaving, who will succeed in this role), prescriptive (what intervention will most improve retention or performance for this segment).
Most firms are still in descriptive territory. The technical bottleneck is rarely modelling — it is data infrastructure (the people data is in 12 systems and the keys don't match), data privacy (the EU GDPR, employee-monitoring norms), and the scarcity of HR practitioners with quantitative backgrounds.
The institutional form is the People Analytics team — a small (typically 3–15 person) group reporting to the CHRO or sometimes to the CFO, staffed with industrial-organisational psychologists, data scientists, and former HR business partners. Wharton's Center for Human Resources, MIT Sloan's Institute for Work and Employment Research, and the Insight222 / Myers-Briggs / Visier consultancies form the practitioner ecosystem.
The frontier is causal inference (most observational HR data is confounded), machine-learning model fairness (bias propagation through hiring and promotion models), and the integration of organisational-network analysis (who actually talks to whom, from email and Slack metadata).
The 2023–25 generative-AI wave has hit HR earlier and harder than most corporate functions. The use cases:
Resume screening and ranking. Long-standing (HireVue, Pymetrics, Eightfold), now LLM-augmented. The Amazon 2018 example — gender-biased screening model — and the NYC 2023 Local Law 144 (algorithmic-hiring audits) frame the regulatory environment.
Job-description writing and interview-question generation. Heavily adopted: ~70% of recruiters report using ChatGPT or equivalent for drafting (LinkedIn Future of Recruiting 2024).
Performance summary drafting. Microsoft Copilot, Workday Illuminate, Lattice AI all offer manager-facing summary tools. The risk: review homogenisation and removal of the cognitive forcing function of writing.
Internal mobility matching. Eightfold, Gloat, Phenom — match employees to internal opportunities by inferred skills graph rather than by manager's mental model.
Employee Q&A copilots. Replace tier-1 HR helpdesk. Moveworks, ServiceNow Now Assist, Workday Assistant.
The risks: discrimination law applies to algorithmic decisions; the EU AI Act categorises HR uses as "high risk"; explainability requirements are tightening; and the HR data the models train on encodes the firm's existing biases.
The Chief Human Resources Officer (or, increasingly, Chief People Officer) sits on the executive team, reports to the CEO, and owns the people strategy. The role expanded substantially in scope through the 2010s and especially after the 2020 pandemic.
Korn Ferry and Russell Reynolds executive-search data: CHRO compensation has risen faster than other C-suite roles since 2015. Median Fortune 500 CHRO total compensation is now in the $2–4M range. CHRO tenure is shorter than other C-suite roles (~4 years median vs ~5–7 for CFO/CEO).
The role's scope today: workforce strategy, organisational design, total rewards, talent acquisition, learning, diversity and inclusion, employee relations, HRIS, people analytics, internal communications (often), and increasingly real estate / workplace experience.
The strategic-vs-administrative tension that Drucker identified in 1954 is still live. The CHROs who break out of the administrative frame — Laszlo Bock at Google, Patty McCord at Netflix, Tracy Layney at Levi's, Christy Pambianchi at Verizon and then Intel — are visible because they are exceptional. The modal CHRO still spends most of their time on tactical operations rather than strategy.
The "CHRO to CEO" pipeline that some have predicted (the function as a path to the top) has produced a few examples (Mary Barra was at one point GM's HR head; Anne Mulcahy at Xerox) but is not yet a routine pattern.
HR is a function with chronic credibility problems. The complaints come in three flavours:
From employees. HR works for the company, not for them. The grievance system protects the firm. Whistleblowers learn this the hard way. The structural critique — Hammonds's 2005 Why We Hate HR, the regular discussion on Glassdoor and similar — is fair as far as it goes. The mitigation is transparency about role, plus genuine ombuds and ethics functions outside the HR reporting line.
From line management. HR is process-heavy, slow, risk-averse, and disconnected from the work. The Ulrich-business-partner model was supposed to fix this. The mixed verdict on its execution after thirty years is that the partnership often becomes a bureaucratic intermediary rather than a strategic one.
From the academic literature. Pfeffer, Cappelli, Boudreau — the function is unscientific, fad-driven, and underuses the substantial empirical knowledge it could be built on. Boudreau's "decision sciences" reframe (the HR function should make people-decisions with the same analytical rigor finance brings to capital decisions) has had limited uptake beyond a handful of leading firms.
The strongest defence: the work is necessary, the regulatory and legal complexity is real, and the function is improving — slowly, and from a low baseline.
A compact orientation to the function: what HR does, how the role splits between operations and strategy, and how the field has moved from administrative to analytic.
Then the deeper cuts:
· Peter Cappelli on Talent Management & the New Workplace — the talent-supply-chain framing applied directly. Cappelli is the most influential academic voice on HR strategy of the last twenty years.
· How to Conduct a Performance Review — the operational mechanics of the most-disliked HR ritual. Watch with an eye to where the conversation about goal-setting (Locke-Latham) and the conversation about evaluation are intentionally separated.
Read: Bock's Work Rules! for the Google operational view; Pfeffer's The Human Equation for the evidence-based-management argument; Cappelli's Talent on Demand for the strategic reframe; the Schmidt-Hunter 1998 paper for the selection-validity facts.
Volume VII, Deck 14. The function with many names — welfare, personnel, human resources, people operations — and one durable problem: how to organise the work of people who can quit.
The discipline has accumulated more empirical knowledge than its practice reflects. The Schmidt-Hunter findings on selection, the Locke-Latham findings on goal-setting, the Dobbin-Kalev findings on diversity, the Bloom findings on remote work — all are usable, all are underused.
Set in Inter and Tiempos. Drafted in May 2026.