people-analytics

# People Analytics ## Core Principle: Data Hygiene Before Analytics A people analytics program built on bad data produces confidently wrong answers. Before investing in dashboards or specialized tools, audit: - HRIS as single source of truth - One field for level, one for location, one for manager, etc. - Demographic data consistent and complete (with employee consent) - Job titles standardized to a controlled vocabulary - Audit cadence quarterly A clean HRIS beats sophisticated analytics on a messy one. ## The 6–8 Reports That Matter For a CEO and Head of People, monthly: 1. **Headcount and trend**: by function, level, geo 2. **Open reqs and time-to-fill** 3. **Hiring funnel**: applications, interviews, offers, accepts; by demographic 4. **Attrition**: regrettable vs. all; by function, tenure, demographic 5. **Engagement / eNPS**: trend 6. **Comp distribution**: by level / function (ranges, compa-ratios); audit by demographic 7. **Manager span and layer audit** 8. **Manager scorecard summary** Each report: - Refreshed automatically (no manual extracts) - Versioned and archived - Reviewed in a recurring meeting - Tied to action when signal is concerning ## Beyond the Basics For larger / more sophisticated programs: - Time-to-productivity (ramp curves) - Promotion velocity by demographic - Internal mobility rates - Retention curves by cohort, source, manager - Pay equity (under privilege) - Performance rating distributions by manager × demographic - 9-box / talent reviews (with skepticism) ## Pitfalls ### Streetlight Effect Optimizing for what's measurable instead of what matters. Easy to measure: time-to-fill, eNPS, training completion rates. Hard but matters: quality-of-hire, behavior change, fairness perception. Don't replace the meaningful with the measurable. ### Demographic Re-Identification Cell-size minimums (n ≥ 5–8) at every cut. Intersectional cuts often run into limits — surface in aggregate or qualitatively. ### Spurious Correlations With dozens of metrics × dozens of cohorts, correlations appear by chance. Distinguish: - Repeated patterns (multiple periods, multiple cuts) - One-off blips (often noise) ### Engagement-Driven Compensation Don't tie comp to engagement scores. Gaming destroys signal. ### Manager Scorecards Used Punitively Without context, manager scores become a popularity contest. Use as input to coaching and development; not as the sole performance metric. ## Privacy and Trust Boundaries: - Aggregate trends shareable - Individual data restricted - Anonymity in surveys preserved (cell-size; vendor agreements) - Demographic linkage with consent only - Don't use AI tools that ingest sensitive data without clearance ## Tools | Stage | Pattern | |-------|---------| | < 100 | Spreadsheets + HRIS reports | | 100–500 | HRIS reports + light BI tool (Looker / Tableau / Mode) | | 500–2000 | Dedicated tools (Visier, ChartHop) or BI on data warehouse | | 2000+ | Data warehouse + dedicated people analytics function | ## Manager-Level vs. Aggregate Manager-level reporting at scale enables manager-specific accountability. Norms: - Managers see their own team results (where n permits) - Managers don't see individual responses - Manager-of-manager reviews trends with the manager - Persistent issues escalate Watch for managers pressuring teams to score them well — destroys signal. ## Pay Equity Analytics (Specialized) Pay equity analyses run under attorney-client privilege when remediation will follow. See `pay-equity-analysis` skill. ## Common Failures - Multiple "sources of truth" disagreeing - Reports built by hand each month - Cell-size violations - Spurious correlations driving decisions - Manager scorecards without context - People analytics function with no path to action - Vendor dashboards that show signal nobody acts on ## Cross-References - `hr-operations-architect` agent - `pay-equity-analysis` skill - `engagement-survey-design` skill - `dei-strategist` agent ## Key References - Industry practitioner work from Visier, ChartHop, One Model - Boudreau, J., & Ramstad, P. (2007). *Beyond HR*. - McKinsey People Analytics studies (with appropriate skepticism)