What is people analytics and why it matters

People analytics is the application of data analysis to people management decisions. It is not a new concept: companies have been measuring turnover and absenteeism for decades. What is new is AI's ability to cross-reference data, identify patterns and predict outcomes at a scale that was impossible with spreadsheets.

The difference between traditional HR and people analytics is the difference between "we have a turnover problem" and "we have a 23% turnover problem in the engineering department, concentrated in profiles with 2 to 4 years of experience, correlated with the lack of internal promotion and a 15% salary gap versus the market, and if we do not act in the next 90 days, we will lose 7 people from the product team".

The first is an observation. The second is an actionable diagnosis. AI takes you from the first to the second.

The 4 maturity levels of people analytics

1. Descriptive: what happened? Basic metrics (turnover, absenteeism, headcount). Most companies are here.
2. Diagnostic: why did it happen? Cause analysis (correlations between turnover and factors like salary, manager, department).
3. Predictive: what will happen? Models predicting who will leave, which teams will have problems, where hires are needed.
4. Prescriptive: what should we do? AI not only predicts but recommends actions and estimates their impact.

HR dashboards: the metrics that matter

An HR dashboard is not a collection of nice numbers. It is a decision tool. The key is choosing the right metrics, not the easiest to measure.

Operational metrics (the minimum)

• Headcount: number of employees by department, contract type, location. With trend (growth/decline).
• Turnover: percentage of employees leaving the company in a period. Voluntary vs involuntary. By department and tenure.
• Absenteeism: absence days per employee per month. Patterns by department and day of the week.
• Time-to-hire: days from vacancy opening to hire. By position and recruitment source.
• Cost per hire: total invested (job boards, recruiter, tools, time spent) divided by hires. By department.

Strategic metrics (the differentiator)

• Revenue per employee: total revenue divided by number of employees. The simplest global productivity metric.
• Engagement score: pulse survey results, monthly evolution, by department.
• Span of control: number of direct reports per manager. Research suggests 5-9 as optimal. More than 12 signals insufficient management infrastructure.
• Internal mobility rate: percentage of vacancies filled with internal talent. Target: > 30%. Indicator of professional development and retention.
• Diversity index: representation of underrepresented groups by hierarchical level and department.
• eNPS: Employee Net Promoter Score. Would you recommend this company as a place to work?

AI in the dashboard

A traditional dashboard shows data. An AI dashboard interprets it:

• Automatic anomalies: "Marketing team turnover increased 40% this quarter vs the previous one. This is statistically significant."
• Correlations: "Departments with span of control > 12 have 25% more turnover."
• Predictions: "If the current trend continues, the engineering team will lose 5 people in the next 6 months."
• Natural narrative: AI generates a text executive summary of the dashboard, ideal for presenting to leadership without having to interpret charts.

Workforce planning with AI

Workforce planning is the discipline of anticipating the organization's talent needs: how many people we need, with what skills, where, and when. Without planning, recruitment is reactive (someone leaves, we open a vacancy). With planning, it is proactive (we know in 6 months we will need 3 data engineers and start looking now).

How AI improves workforce planning

• Scenario modeling. What if we grow 20%? What if we lose 15% of the sales team? What if we automate process X? AI models multiple scenarios with their impact on headcount, skills and costs.
• Retirement and departure prediction. AI calculates how many employees will retire in the next 3-5 years, and how many are at risk of leaving (based on engagement and market data).
• Future skills analysis. What skills will the company need in 2-3 years based on industry trends? AI crosses labor market data, industry publications and internal strategic plans.
• Build vs buy vs borrow. For each skill gap: is it better to train someone internally (build), hire (buy), or outsource (borrow)? AI compares costs, timelines and risks of each option.

Prompt for workforce planning

"My company has [N] employees, distributed as follows: [breakdown by department]. Strategic plan for the next 12 months: [key objectives]. Current turnover rate: [X%]. Expected retirements: [N]. Planned automations: [processes]. Generate a workforce plan for the next 12 months: 1) Required headcount by department and quarter. 2) Critical skills to cover. 3) For each gap: build/buy/borrow recommendation with estimated cost. 4) Hiring timeline. 5) Risks and contingency plan."

Diversity and inclusion metrics

Diversity without measurement is a statement of intent. Diversity metrics are not a compliance exercise (though regulation increasingly requires them). They are a management tool: diverse teams make better decisions (McKinsey, "Diversity Wins", 2020), innovate more (BCG, 2018), and retain talent better.

What to measure

• Representation by level: not just the overall percentage of women (or minorities), but their representation by hierarchical level. A company with 50% women but 0% in leadership has a problem.
• Promotion pipeline: what percentage of each group is promoted in each cycle? Are there disparities?
• Pay gap: salary difference between groups for the same position, level and experience. Adjusted, not gross.
• Inclusion index: from surveys: do employees from all groups feel equally included, heard and valued?
• Hiring funnel by group: at which stage of the selection process are candidates from underrepresented groups lost?

How AI helps

• Intersectional analysis: not just gender or ethnicity separately, but combinations (woman + over 50 + technical department). AI can cross multiple variables that would be impossible to analyze manually.
• Hidden bias detection: patterns in promotions, project assignments, training access that correlate with demographic data.
• Benchmarking: compare your diversity metrics with your sector and market. Not to meet a quota, but to understand where you stand relative to the reference.

Compensation benchmarking

Salary is the number 1 reason employees change jobs (not the only one, but the first in nearly every study). A compensation system misaligned with the market is a turnover machine.

How AI improves salary benchmarking

• Market comparison. AI crosses your salary data with market sources (Glassdoor, Levels.fyi, industry salary surveys) and calculates the percentile for each position. An employee at the 25th percentile is below 75% of the market. They are a flight candidate.
• Internal equity. Do employees in the same position, with the same experience and performance, earn the same? AI detects unjustified deviations.
• Adjustment simulation. If I adjust salaries to the market 50th percentile, how much does it cost? And to the 75th? AI models the budget impact of different salary policies.
• Total compensation. Not just base salary. Bonus, equity, benefits, training, flexibility. AI can calculate the total compensation package value and compare it with the competition.

Prompt for compensation analysis

"Analyze the salary equity of my team: [table with position, experience, base salary, bonus, gender, tenure]. Generate: 1) Internal equity map (are there employees in the same position with significant salary differences without justification?). 2) Estimated market positioning (based on public data for [sector] in [location]). 3) Employees at risk of leaving due to compensation (low percentile + high profile demand). 4) Estimated budget to adjust to the market 50th percentile."

Predictive analytics for HR

Predictive analytics uses historical data to predict future outcomes. In HR, the most common applications are:

• Turnover prediction: which employees have the highest probability of leaving in the next 6 months? (Covered in HR04, here we go deeper with data.)
• Performance prediction: from onboarding and first-month data, predict 12-month performance. Useful for evaluating hiring quality.
• Absenteeism prediction: patterns predicting absenteeism increases (manager change, workload increase, team problems).
• Training needs prediction: based on industry changes and current skill gaps, predict which training will be critical in 12-18 months.

Practical implementation

You do not need a data science team for basic predictive analytics. An LLM with access to your HR data (exported to CSV or connected via API) can do analyses that previously required weeks of analyst work.

The key: start simple. A regression model with 5 variables (tenure, department, salary gap vs market, engagement score, time since last promotion) predicts turnover with 70% accuracy. It is not perfect, but it is better than intuition.

HR data quality: the bottleneck

The biggest obstacle to people analytics is not technology. It is data quality. If your HR data is in 5 different systems, with inconsistent formats, empty fields and duplicates, no AI algorithm will produce useful results.

Common HR data problems

• Data in silos: payroll in one system, evaluations in another, training in another, recruitment in another. Without a common identifier connecting them.
• Inconsistent fields: the department is called "Marketing" in one system, "Mkt" in another, and "Marketing and Communications Department" in a third.
• Outdated data: an employee who changed positions 6 months ago but still has the old position in the HRIS.
• Missing data: 40% of employees have no registered performance evaluation. 60% have no updated skill profile.

Data hygiene plan (before doing analytics)

1. Source inventory. List all systems containing people data. Identify the "single source of truth" for each data type.
2. Standardization. Define unique naming conventions for departments, positions, locations, levels.
3. Integration. Connect systems with a common identifier (employee ID). AI can help reconcile duplicate or inconsistent records.
4. Completeness. Identify critical fields and their completeness percentage. Plan to reach 90%+ on fields needed for analytics.
5. Maintenance. Processes to keep data current: automatic triggers when a position changes, reminders to update skill profiles, etc.

From data to stories: HR data storytelling

People analytics only has impact if it convinces leadership to act. A perfect dashboard nobody looks at is useless. AI can help you turn data into narratives that drive action.

Prompt to generate HR narrative

"Convert this HR data into an executive report for the leadership team: [paste key data]. Structure: 1) Headline (the main conclusion in 1 sentence). 2) Context (why this matters for the business, 2 sentences). 3) Key data (3-5 bullets with the most relevant metrics). 4) Risks if we do not act (2 bullets with quantified impact). 5) Recommendations (3 concrete actions with estimated cost and expected ROI). Tone: business-oriented, not HR-oriented. The reader is a CEO, not an HR manager. Length: 1 page."

Common people analytics mistakes

1. Measuring everything, acting on nothing. Having 50 metrics in a dashboard does not make you data-driven. Having 5 metrics that generate 5 concrete actions does.
2. Confusing correlation with causation. "Employees who use the company gym have 30% less absenteeism." That does not mean the gym reduces absenteeism. Healthier employees may use both.
3. Ignoring privacy. People analytics with non-anonymized individual data is a GDPR risk and a trust risk. If employees feel they are being monitored, engagement drops.
4. Analytics without context. A 20% turnover may be excellent in hospitality and disastrous in engineering. Numbers without sector, market and timing context mean nothing.
5. Starting with the tool. "We need Visier/Tableau/PowerBI." No. You first need clean data and clear questions. The tool is last, not first.

Ejercicio practico

Puntos clave