HR analytics

In short

HR analytics (people analytics) is the systematic use of employee data, statistics and business logic to measure what HR is doing, explain why workforce outcomes happen and predict what will happen next — so that people decisions are evidence-based rather than intuitive. It turns HRIS records, payroll, attendance, appraisal scores and engagement surveys into metrics, dashboards and models that finance and the board can read. Indian universities examine it in HRM, HRIS and strategic HR papers (AICTE model BBA curriculum, UGC CBCS), and it is the fastest-growing HR skill tested in placement rounds at TCS, Infosys, Wipro, Accenture, Deloitte, EY, Genpact, WNS, Flipkart and bank/NBFC HR teams. Interviews test metric formulas, Excel/SQL comfort and the ability to read a number and recommend an action.

Explained

What it is and the four levels of analytics

• Definition: the application of statistical, data-mining and visualisation methods to HR data to improve workforce and business outcomes. Also called people analytics, talent analytics or workforce analytics.
• HR metric vs HR analytics: a metric is a single measured number (attrition = 18%); analytics is the analysis that links it to a driver and a business result (attrition is 18% because first-year supervisor quality is low, costing ₹2.1 crore a year).
• Level 1 — Descriptive (what happened): headcount, attrition, absenteeism, cost per hire. Historical reporting and dashboards; about 70-80% of what most Indian HR teams actually do.
• Level 2 — Diagnostic (why it happened): drill-down, correlation, cross-tabs, exit-interview text analysis — e.g. attrition by manager, location, tenure band.
• Level 3 — Predictive (what will happen): regression, logistic regression, decision trees, survival analysis — flight-risk scores, hiring-demand forecasts.
• Level 4 — Prescriptive (what should we do): optimisation and scenario simulation — the best mix of hiring, internal movement and retention spend under a budget.
• A fifth level, cognitive/AI-based (NLP on feedback, resume parsing, chatbots), is often added in newer syllabi.

Process of an HR analytics project (examinable sequence)

1. Frame the business question: state it as a decision, not a report request — "which stores lose most new joiners in 90 days, and why?"
2. Identify measures and data sources: HRIS/HRMS master (SAP SuccessFactors, Darwinbox, Zoho People, Keka), payroll, biometric attendance, ATS, LMS, appraisal, engagement survey, exit interviews, plus business data (sales, quality, CSAT).
3. Collect and clean: remove duplicates, fix employee-ID mismatches across systems, treat missing values, standardise date formats, flag outliers. Typically the longest step.
4. Define metrics precisely: fix numerator, denominator, period and population (does attrition include contractors? interns? involuntary exits?). Inconsistent definitions are the commonest cause of disputed numbers.
5. Analyse: descriptive statistics → segmentation → correlation → model. Test significance; separate correlation from causation.
6. Visualise and communicate: dashboard and one-page story with an insight, a rupee impact and a recommendation.
7. Act and evaluate: implement, then re-measure to prove the intervention worked (ROI of the action).
8. Tools in Indian practice: Excel (PivotTables, VLOOKUP/XLOOKUP, Power Query), Power BI / Tableau, SQL, Python/R, HRIS inbuilt analytics; SPSS in academic projects.

Core metric families

• Workforce/demographic: headcount, FTE, span of control, average tenure, average age, diversity ratio, contingent-worker share.
• Recruitment: time to fill, time to hire, cost per hire, offer-acceptance ratio, offer-decline/shopping ratio, source of hire, selection ratio, yield ratio, quality of hire, early attrition (90-day/6-month).
• Retention: attrition/turnover rate (voluntary, involuntary, total), regretted vs non-regretted attrition, retention rate, stability index, survival rate, flight risk.
• Learning: training hours per employee, training cost per employee, participation rate, pre/post test gain, Kirkpatrick levels, training ROI.
• Performance and productivity: revenue per employee, profit per employee, HR-to-employee ratio, HR cost per employee, utilisation/billability (IT and ITES), performance-rating distribution.
• Engagement and wellbeing: engagement score, eNPS, absenteeism rate, Bradford factor, grievance rate, internal-mobility rate, accident frequency rate.
• Compensation: compa-ratio, range penetration, pay-band compression, labour cost as % of revenue, gender pay gap, salary-increase percentage.
• A good metric is actionable, benchmarkable, consistently defined, and tied to a business outcome; vanity metrics (number of training programmes run) are not.

Frameworks and models

• HR scorecard (Becker, Huselid, Ulrich): links HR deliverables, HR practices, HR system alignment and HR efficiency to firm performance; derived from the Balanced Scorecard (financial, customer, internal process, learning and growth).
• LAMP framework (Boudreau and Ramstad): Logic (the causal story), Analytics (correct method), Measures (right numbers), Process (influence and change management). Used to explain why good numbers still fail to change decisions — the P is missing.
• Service-profit chain / causal chain: employee engagement → service quality → customer satisfaction → revenue; the logic behind most HR-to-business linkage studies.
• Human capital ROI and utility analysis: value of an HR intervention in money; HC ROI and Human Economic Value Added (HEVA).
• Workforce planning models: demand forecast (ratio analysis, trend analysis, regression on business driver, Delphi), supply forecast (skills inventory, Markov chain transition matrices, replacement and succession charts), gap analysis.
• HR audit vs HR analytics: an audit is a periodic compliance and process review (are statutory registers and PF/ESI filings in order?); analytics is continuous measurement for decisions.
• Benchmarking: internal (over time, across units) and external (industry data from NASSCOM, Aon, Mercer, Deloitte India salary-increase surveys, SHRM India).

Statistical methods examined

• Descriptive: mean, median, mode, range, standard deviation, percentiles, frequency distribution. Median salary is preferred over mean because pay is right-skewed.
• Correlation (r): −1 to +1; strength and direction of linear association. Correlation is not causation — attrition may correlate with training hours because high-potential leavers get more training.
• Simple and multiple regression: predict a continuous outcome (salary, performance score) from drivers; read coefficients, R², p-values (significant at p < 0.05).
• Logistic regression: predicts a binary outcome (will leave / will stay); output is a probability — the basis of flight-risk scoring.
• Chi-square test: association between two categorical variables (gender vs promotion); t-test/ANOVA: compare means between two or more groups (ratings across departments).
• Survival analysis (Kaplan-Meier): probability an employee is still with the firm at month t; handles employees who have not yet left (censored data).
• Clustering/segmentation: group employees by behaviour for targeted policies; decision trees for interpretable attrition rules.
• Text analytics: sentiment and theme extraction from exit interviews and open-ended survey comments.

Data privacy, ethics and the Indian legal frame

• Digital Personal Data Protection Act, 2023 (DPDP Act): India's data-protection law; employee data is personal data, the employer is the Data Fiduciary and the employee the Data Principal. Obligations include lawful purpose, notice, purpose limitation, data minimisation, accuracy, security safeguards and breach reporting to the Data Protection Board and affected principals. Employment-related processing is a "legitimate use" (Section 7), so consent is not always required, but purpose limits, accuracy, security and breach duties still apply. The DPDP Rules were notified in November 2025 with most obligations phased in over roughly 18 months — check the current notified position before quoting a compliance date.
• Earlier basis: Information Technology Act, 2000 and the SPDI Rules, 2011 (sensitive personal data — passwords, financial, health, biometric); plus the Supreme Court's K. S. Puttaswamy (2017) judgment recognising privacy as a fundamental right.
• Statutory sources of HR data in India: EPF (UAN), ESI, Shops and Establishments and Factories Act registers, Payment of Wages and Minimum Wages records, POSH Act annual report, maternity-benefit records, and the four labour codes (Code on Wages 2019; Industrial Relations Code 2020; Code on Social Security 2020; OSH&WC Code 2020), which the Centre brought into force on 21 November 2025 with central and state rules still being notified — verify current status.
• Ethical rules: anonymise and aggregate (suppress groups below about 5 employees), use data only for the stated purpose, avoid proxy discrimination (pin code, college tier, marital status as a hidden proxy for caste, class or gender), keep a human decision-maker in the loop, audit models for bias, and never use surveillance-style monitoring without disclosure.
• Limitations of HR analytics: small samples in single firms; poor data quality and siloed systems; soft outcomes hard to quantify; correlation mistaken for causation; Hawthorne and survivorship bias; low HR statistical skill; employee trust eroded if analytics feels like policing.

Key points

• HR analytics - evidence-based measurement and modelling of workforce data to improve people and business decisions.
• Metric vs analytics - a metric reports a number; analytics explains the driver and the rupee consequence.
• Four levels - descriptive (what), diagnostic (why), predictive (what next), prescriptive (what to do).
• Attrition rate - separations ÷ average headcount × 100, always stated with period and population.
• Regretted attrition - exit of employees the firm wanted to keep; the number that matters for action.
• Early attrition - exits within 90 days or 6 months; signals selection, onboarding or supervisor problems.
• Cost per hire - (internal + external recruiting costs) ÷ number of hires in the period.
• Time to fill vs time to hire - requisition approval to offer acceptance, vs first contact with the selected candidate to acceptance.
• Revenue per employee - total revenue ÷ average FTE; the headline productivity metric in Indian IT.
• Compa-ratio - actual pay ÷ midpoint of the pay range; below 1 means paid under the range midpoint.
• Absenteeism rate - days lost ÷ scheduled workdays × 100; the Bradford factor weights frequent short absences more heavily.
• HR scorecard - Balanced-Scorecard logic applied to HR deliverables, practices, alignment and efficiency.
• LAMP framework - Logic, Analytics, Measures, Process — analytics changes nothing without the Process (influence) leg.
• Logistic regression - standard technique for flight-risk scoring because the outcome is leave/stay.
• Survival (Kaplan-Meier) analysis - probability of remaining employed at a given tenure, handling censored records.
• Human capital ROI - (revenue − operating expense excluding employment cost) ÷ employment cost.
• DPDP Act, 2023 - employer is Data Fiduciary, employee is Data Principal; notice, purpose limitation, minimisation, security and breach reporting.
• Biggest practical risk - inconsistent metric definitions and dirty data across HRIS, payroll and ATS.

Formulas and examples

Units: headcount in persons or FTEs; money in ₹; time in days; rates in %.

Example 1 — Attrition, annualised attrition and retention A BPO in Pune starts the financial year with 1,200 employees and ends with 1,400. During the year 260 employees left (210 resignations, 50 terminations) and 460 were hired.

• Average headcount = (1,200 + 1,400) ÷ 2 = 1,300
• Total attrition = 260 ÷ 1,300 × 100 = 20.0%
• Voluntary attrition = 210 ÷ 1,300 × 100 = 16.15%
• If 1,010 of the opening 1,200 are still present at year end: retention rate = 1,010 ÷ 1,200 × 100 = 84.2% (so 190 exits came from the opening cohort and the other 70 from employees hired during the year)
• If one month shows 24 exits on an average headcount of 1,300: annualised attrition = 24 ÷ 1,300 × 12 × 100 = 22.2% — a monthly spike annualises into a large-looking figure, so always label it "annualised".

Example 2 — Cost per hire and cost of attrition A Bengaluru IT firm hired 50 engineers in a quarter. Costs: job-portal and LinkedIn subscriptions ₹3,00,000; agency fees ₹9,00,000; campus drive ₹1,50,000; referral bonuses ₹2,50,000; recruiter salaries apportioned ₹4,50,000; assessment tools ₹50,000.

• Total recruiting cost = 3,00,000 + 9,00,000 + 1,50,000 + 2,50,000 + 4,50,000 + 50,000 = ₹21,00,000
• Cost per hire = 21,00,000 ÷ 50 = ₹42,000
• Cost of one mid-level exit: separation and settlement ₹40,000 + 45 days vacancy coverage ₹90,000 + replacement hiring ₹42,000 + training and ramp-up ₹1,28,000 + lost productivity during 3-month ramp ₹1,50,000 = ₹4,50,000
• If 60 such employees leave in a year, cost of attrition = 60 × 4,50,000 = ₹2.70 crore. Cutting attrition by one-fifth (12 exits) saves ₹54,00,000 — the sentence that gets a retention budget approved.

Example 3 — Productivity, HR ratio and human capital ROI A company reports revenue ₹500 crore, total operating expense ₹440 crore of which employment cost is ₹180 crore, average FTEs 2,500, HR team 25.

• Revenue per employee = 500 ÷ 2,500 = ₹0.20 crore = ₹20 lakh per employee
• Profit per employee = (500 − 440) ÷ 2,500 = 60 ÷ 2,500 = ₹0.024 crore = ₹2.4 lakh
• Non-employment operating expense = 440 − 180 = ₹260 crore
• Human capital ROI = (500 − 260) ÷ 180 = 240 ÷ 180 = 1.33 — every ₹1 of employment cost returns ₹1.33 of gross value; above 1.0 is the minimum acceptable.
• HR-to-employee ratio = 25 ÷ 2,500 = 1 : 100, i.e. 1.0 HR staff per 100 employees (Indian services firms commonly run about 0.7-1.4 per 100).

Example 4 — Absenteeism and the Bradford factor A plant schedules 26 workdays a month for 400 workers; 520 worker-days were lost to absence.

• Available workdays = 400 × 26 = 10,400; absenteeism rate = 520 ÷ 10,400 × 100 = 5.0%
• Worker A: 1 spell of 10 days → B = 1² × 10 = 10. Worker B: 5 spells totalling 10 days → B = 5² × 10 = 250.
• Same days lost, very different disruption: the Bradford factor flags the frequent short-absence pattern for counselling, while A's single long spell is likely genuine illness.

Example 5 — Correlation between engagement and attrition Five branches report engagement score (x, out of 100) and annual attrition % (y): (60, 28), (70, 22), (75, 18), (85, 12), (90, 10).

• x̄ = (60+70+75+85+90) ÷ 5 = 380 ÷ 5 = 76; ȳ = (28+22+18+12+10) ÷ 5 = 90 ÷ 5 = 18
• Deviations (dx, dy): (−16, 10), (−6, 4), (−1, 0), (9, −6), (14, −8)
• Σdx·dy = −160 − 24 + 0 − 54 − 112 = −350; Σdx² = 256 + 36 + 1 + 81 + 196 = 570; Σdy² = 100 + 16 + 0 + 36 + 64 = 216
• r = −350 ÷ √(570 × 216) = −350 ÷ √123,120 = −350 ÷ 350.9 = −0.998 — a very strong negative association.
• Regression slope b = −350 ÷ 570 = −0.614; intercept a = 18 − (−0.614 × 76) = 18 + 46.7 = 64.7, so attrition % ≈ 64.7 − 0.614 × engagement. A 10-point engagement gain predicts about 6.1 percentage points lower attrition — but with n = 5 and no controls this is an association, not proof of cause.

Where it is used

• IT and ITES (TCS, Infosys, Wipro, HCLTech, Cognizant India): quarterly attrition disclosure to investors, bench and utilisation analytics, skill-supply forecasting, flight-risk models for billable staff.
• BPO, banking and NBFC (Genpact, WNS, HDFC Bank, ICICI, Bajaj Finance): shrinkage and absenteeism analytics for workforce management, branch-level attrition dashboards, sales-productivity-to-incentive linkage.
• Consulting and HR advisory (Deloitte, EY, PwC, KPMG, Aon, Mercer, Korn Ferry India): people-analytics and rewards-benchmarking engagements, salary-increase and attrition surveys.
• E-commerce, retail and quick commerce (Flipkart, Reliance Retail, Zomato, Swiggy): gig and delivery-partner churn models, store-level staffing optimisation, seasonal hiring forecasts.
• Manufacturing and plants (Tata Steel, Maruti Suzuki, L&T): safety and absenteeism analytics, skill-matrix gap analysis, contract-labour compliance reporting.
• Job roles: HR analyst, people analytics analyst, HRIS/HRMS analyst, HR MIS executive, compensation and benefits analyst, workforce management (WFM) analyst, talent acquisition analyst — mostly Excel and Power BI first, SQL and Python next.

Interview questions

• Q: What is the difference between an HR metric and HR analytics? A: A metric is a measured number such as 18% attrition; analytics interprets it — identifying the driver, testing it statistically and quantifying the business impact so a decision can be taken.
• Q: How do you calculate attrition rate, and what is the trap? A: Separations ÷ average headcount × 100 for a stated period. The trap is the denominator — using opening or closing headcount in a fast-growing firm, or annualising one bad month without labelling it, distorts the figure badly.
• Q: Why is regretted attrition more useful than total attrition? A: Total attrition mixes exits the firm wanted with exits it did not; regretted attrition isolates the loss of good performers and critical skills, which is what retention spend should target.
• Q: Which technique would you use to predict whether an employee will resign, and why? A: Logistic regression (or a decision tree / gradient-boosted model), because the outcome is binary — leave or stay — and the output is an interpretable probability that can be banded into risk tiers.
• Q: Attrition and training hours are positively correlated in our data. What do you conclude? A: Nothing causal — high-potential employees usually get more training and are also the most marketable, so a third variable drives both; I would control for performance rating, tenure and market demand before concluding anything.
• Q: Explain the LAMP framework. A: Logic (the causal story), Analytics (the correct method), Measures (the right numbers) and Process (influencing decision-makers); projects usually fail on Process, producing accurate analysis that nobody acts on.
• Q: How would you build the business case for a retention programme? A: Cost one exit fully (separation, vacancy, replacement hiring, training, lost productivity), multiply by regretted exits, then compare the rupee saving from a realistic reduction against the programme cost and show the payback period.
• Q: What is compa-ratio and how do you act on it? A: Actual pay ÷ range midpoint. Below about 0.90 for a strong performer signals a flight risk and a correction case; well above 1.10 for an average performer signals over-payment and a case for restricting further increases.
• Q: What privacy rules apply to employee data in India? A: The Digital Personal Data Protection Act, 2023 makes the employer a Data Fiduciary with duties of notice, lawful purpose, data minimisation, security and breach reporting; the IT Act and SPDI Rules cover sensitive data, and the DPDP Rules notified in November 2025 phase these obligations in, so the current position should be verified.
• Q: Give an example of bias entering an HR model. A: Using college tier, pin code or marital status as predictors can act as a proxy for caste, class or gender, so the model reproduces past discrimination even though no protected attribute was included.
• Q: What is survival analysis used for in HR? A: To estimate the probability that an employee is still with the firm at a given tenure, using Kaplan-Meier curves that correctly handle current employees whose exit has not yet occurred (censored data).
• Q: Your dashboard shows attrition down but the business is still short-staffed. What do you look at? A: Net headcount movement — attrition can fall while hiring and internal transfers lag; I would check time to fill, offer-acceptance ratio, vacancy ageing and the demand forecast rather than the attrition rate alone.
• Q: What is the single biggest practical obstacle in an HR analytics project? A: Data quality and definition mismatch — HRIS, payroll, ATS and attendance systems rarely agree on employee IDs, joining dates or employee categories, so most of the effort goes into cleaning and agreeing definitions.

Common mistakes

• Using opening or closing headcount instead of average headcount in attrition, or forgetting to state the period and whether contractors and involuntary exits are included.
• Annualising a single month's attrition and presenting it as the annual rate without the word "annualised".
• Treating correlation as causation — the classic exam trap with engagement, training and attrition data.
• Reporting metrics without a benchmark, trend or rupee impact, so the number tells management nothing.
• Confusing time to fill with time to hire, or cost per hire with cost of attrition.
• Building a predictive model on a tiny, imbalanced sample and quoting accuracy instead of precision, recall or AUC for a rare event like resignation.
• Ignoring privacy and ethics — publishing team-level results for groups too small to anonymise, or using surveillance data without disclosure under the DPDP Act.

Related topics

Human resource management · Recruitment and selection · Performance management and appraisal · Compensation and benefits · Talent management and succession planning · Business statistics and research methods