Hockey Recruitment with Machine Learning

Traditional scouting relies on subjective evaluations and limited observations. Machine Learning gives us the ability to analyze thousands of players systematically and objectively.

The Problem with Traditional Scouting

Human Limitations:

• Bias based on previous experience
• Limited time to observe all players
• Difficult to remember and compare performances over time
• Influence of individual games rather than trend analysis

Costs:

• Expensive travel for scouts
• Time-intensive process
• Risk of “group think” within the organization

My ML Approach to Recruitment

1. Player Similarity Models

I use k-nearest neighbors and clustering to find players who resemble already successful profiles:

# Simplified version of similarity scoring
def find_similar_players(target_player, player_database):
    features = ['goals_per_game', 'assists_per_game', 'corsi_for_pct', 
               'zone_starts_off_pct', 'shooting_pct']
    
    scaler = StandardScaler()
    normalized_data = scaler.fit_transform(player_database[features])
    
    similarity_scores = cosine_similarity([target_player], normalized_data)
    return similarity_scores

2. Development Trajectory Prediction

By analyzing development patterns, we can predict which young players will “break out”:

Key indicators:

• Shot attempt rate progression
• Deployment evolution (ice time, zone starts)
• Linemate quality adjustments
• Age-adjusted performance curves

3. Value vs. Cost Optimization

ML helps us find “bargains” - players who perform better than their salary/draft position indicates.

Case Study: Successful Identification

Last year I helped a team identify a 20-year-old defenseman in a European league:

Scouting report: “Not physical enough for North American hockey”

ML analysis showed:

• Top 5% in pass completion under pressure
• Excellent shot suppression metrics
• Development curve similar to established NHL defensemen
• Undervalued due to low point production

Result: Player signed for minimal cost and became a top-4 defenseman within 18 months.

Feature Engineering for Recruitment

Traditional stats are just the beginning. Important ML features include:

Contextual Adjustments

• League quality adjustments
• Age-relative performance
• Linemate/teammate quality impact
• Situational usage (PP, PK, even strength)

Advanced Metrics

• Expected goal rates (both for and against)
• Shot quality metrics
• Puck possession indicators
• Transition play efficiency

Injury Risk Factors

• Historical injury patterns
• Playing style risk indicators
• Workload sustainability metrics

Implementation for Organizations

Step 1: Data Collection & Cleaning

• Integrate multiple data sources
• Standardize metrics across leagues
• Historical performance database

Step 2: Model Development

• Train on your organization’s successful picks
• Validate against missed opportunities
• Continuous model improvement

Step 3: Scout Integration

• ML as complement, not replacement
• Prioritization of scouting resources
• Red flags for human review

ROI & Results

Organizations implementing ML-driven scouting see:

• 25-40% improvement in draft success rate
• Reduced scouting costs through smart resource allocation
• Faster identification of emerging talents
• Less bias in evaluation process

Next Steps

Ready to revolutionize your recruitment process? I help organizations implement customized ML solutions for talent identification.

Book Strategy Meeting →