a four-rung ladder · most HR is on rung 1 · the board asks for rung 3
Predictive analytics & AI in HR strategy
The ladder doesn\'t skip rungs. You can\'t buy your way from descriptive to predictive — predictive requires clean historical data, statistical capability, and the decision authority to act on the forecast. Most failed AI-in-HR programs skipped a rung.
The maturity ladder
04
Prescriptive
“What should we do?”
Recommendation engines that propose interventions. Requires real-time data, AI engineering, audit + ethics framework. Front of the field.
03
Predictive
“What will happen?”
Statistical models — flight risk, hiring demand, comp drift, succession readiness. Requires multi-year clean data + data science team + decision authority.
02
Diagnostic
“Why did it happen?”
Root-cause drill-downs — why turnover spiked in this region, why this manager's engagement dropped. Requires clean historical data + BI capability.
01
Descriptive
“What happened?”
Headcount, turnover, time-to-fill, training completion. Lagging indicators. Where most HR functions live today.
Each rung needs three things, not just a tool: clean data, analytical capability, and decision authority to act on what the data says. Buying a predictive analytics platform without those gives you a screen-saver, not insight.
the compliance overlay senior HR owns
Audit before deploy. Document the audit. Explain the model.
NYC LL 144 · 2023
Annual third-party bias audit for automated employment decision tools. Public summary required. Candidate notice mandatory.
EU AI Act · 2024
Employment AI is “high-risk.” Risk management, data governance, human oversight, accuracy, robustness all required.
Title VII · ongoing
Algorithm-driven adverse impact still violates Title VII. EEOC tech initiative. Multiple US states drafting comparable rules.
Removing protected fields doesn\'t remove bias
Models learn proxies. Zip code becomes a race proxy. College name becomes a socioeconomic proxy. Language patterns become an age proxy. Stripping out gender or race from the training data leaves every other field free to encode the pattern. Bias mitigation requires auditing the predictions for adverse impact, not just cleaning the inputs.
Exam Traps
Predictive ≠ prescriptive
Predictive forecasts what will happen. Prescriptive recommends what to do. Different capability levels.
AI bias is not solved by removing protected fields
Models learn proxies (zip, college, language patterns). Bias mitigation requires audit, not data scrubbing.
AI explainability is regulated
NYC LL 144 + EU AI Act require explainability. Black-box models fail compliance.
Strategic AI use vs operational AI use differ
Strategic AI informs board decisions. Operational AI automates transactions. Senior HR governs both.
Don't buy your way to maturity
Buying a predictive analytics tool does not produce predictive analytics. Data, capability, and decision authority must precede.
Audit before deploy
Bias audits, explainability reviews, privacy compliance — before AI hits production. Senior HR owns governance.
Headcount, turnover, time-to-fill. Lagging indicators.
Root cause analysis. Drill-down. Why turnover by manager.
Statistical models. Flight risk, hiring demand, comp drift.
Recommendation engines. Optimal interventions.
Ready to test your SPHR knowledge?
1,500+ practice questions written by certified professionals.
Start Practicing SPHR