Learning Recommendation Diversity Index (LRDI)

GMI-HCM-015Human Capital ManagementPilot-ReadyPattern O-6: Autonomous Coherence

Definition

LRDI measures the degree of variation in AI-recommended learning paths across employees in similar roles.

The Measurement Gap

Measurement Gap and Decision Impact

Using Shannon entropy normalized against maximum possible diversity, it quantifies whether the AI learning platform is producing diverse skill development across a role cohort or funneling everyone toward the same content. A score near 1.0 means maximum diversity in learning consumption. A score near 0 means the recommender is creating skill monoculture.

The Decision It Supports:

Is the AI recommender creating skill monoculture or diversifying workforce development? Organizations need teams with complementary capabilities, not identical skill profiles. If the AI learning platform recommends the same five courses to every data analyst, the organization builds depth in those five areas but creates a capability blind spot everywhere else.

Why It Requires a New Metric:

Legacy L&D metrics did not measure recommendation diversity because there were no recommendation engines to evaluate. Course catalogs existed, employees chose from them, and diversity of consumption was an emergent property of individual choice. AI recommendation engines actively shape consumption patterns, concentrating learning in areas the algorithm identifies as highest priority. Whether that concentration serves or undermines organizational capability development requires a metric designed to measure recommendation diversity, which did not exist before AI-driven learning.

The measurement gap LRDI addresses is the invisible homogenization of workforce capability that AI learning recommenders can produce: organizations are using recommendation engines that may systematically narrow the skill diversity of their workforce without any metric to detect it.

AI learning recommendation engines optimize for relevance, which typically means recommending content that aligns with inferred skill gaps and role requirements. The optimization is individually logical: each employee receives recommendations tailored to their profile. But when the algorithm identifies similar gaps across similar roles and recommends similar content to address them, the aggregate effect is homogenization. A cohort of 50 software engineers all receive recommendations for the same cloud architecture course, the same agile methodology module, and the same system design tutorial. Individually, each recommendation makes sense. Collectively, the organization has produced 50 engineers with identical supplemental skills.

This matters because organizational capability depends on diversity of expertise within role cohorts, not just depth of common expertise. When a novel problem arrives that requires skills outside the algorithm's recommendation pattern (a security challenge, a performance optimization approach, a domain-specific integration) no one in the cohort has been developing those capabilities because the recommender never surfaced them.

The recommender is not malicious. It is optimizing for a narrow objective (individual relevance) that conflicts with a broader organizational need (team capability diversity). LRDI makes this tension visible by computing Shannon entropy of learning content consumption within role cohorts, normalized against maximum possible entropy. Low entropy means the cohort is consuming nearly identical content. High entropy means diverse consumption patterns exist.

The comparison between AI-recommended content diversity and self-selected content diversity is particularly revealing. If employees who choose their own learning consume a more diverse portfolio than employees who follow AI recommendations, the algorithm is actively reducing the organization's skill diversity, a finding that would be invisible without LRDI.

Without LRDI, AI learning platforms can optimize individual learning recommendations while systematically narrowing organizational capability, and no dashboard will show the problem until the organization discovers it cannot staff a cross-functional initiative because everyone developed the same skills.

Formula

Formula:

LRDI = (1/C) × Σ Shannon_H(learning_content_c) / Shannon_H_max

Variable Definitions:

Variable Definition Unit Source
C Number of role cohorts included in measurement Integer HRIS role definitions
Shannon_H(learning_content_c) Shannon entropy of the learning content distribution consumed by cohort c: computed from the frequency distribution of distinct content items consumed across cohort members Nats or bits Computed from LMS/LXP consumption data
Shannon_H_max Maximum possible Shannon entropy given the available content pool (entropy under uniform distribution) Nats or bits Computed from LMS/LXP catalog size

Units: 0 to 1 ratio. 0 = perfect homogeneity (every employee consumed identical content). 1 = maximum diversity (uniform distribution across available content).

All thresholds are proposed starting points for organizational calibration, not industry standards.

Edge Cases and Exclusions:

  • Exclude mandatory training (safety, compliance, regulatory) from the content distribution. These are required regardless of recommendation and would artificially reduce diversity scores
  • Very small cohorts (fewer than 10 employees) produce noisy entropy estimates: aggregate into broader role families
  • Normalize against the available content pool size for each skill area, a cohort cannot diversify beyond what the catalog offers

Data Requirements:

Data Element Source System Classification
AI-recommended learning content by employee LMS/LXP NATIVE
Consumed/completed content by employee LMS/LXP NATIVE
Role cohort definitions HRIS NATIVE
Available content pool by skill area LMS/LXP catalog NATIVE

Proposed formula, pending validation pilots.

At a Glance

Unit
Ratio (0 to 1)
Trend Direction
Higher; away from homogeneity
Reporting Frequency
Quarterly
Category
AI Learning and Development Effectiveness
← Learning Competence Conversion Rate (LCCR)Homogeneity Amplification Index (HAI) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment