Lead Scoring Models

Choose between rule-based, AI-powered, or hybrid scoring to match your sales process.

Overview

Grader.io offers three primary scoring models, each with distinct advantages depending on your business needs, data availability, and team preferences.

Rule-Based Scoring

When to Use

• Clear qualification criteria - You know exactly what makes a good lead
• Predictable patterns - Your ideal customers have consistent characteristics
• Compliance requirements - Need transparent, auditable scoring logic
• Limited historical data - Don't have enough conversion data for ML training

How It Works

Explicit rules define point values for form field conditions:

// Example rule-based scoring
if (jobTitle.includes('VP', 'Director', 'Chief')) {
  score += 25; // Decision maker bonus
}

if (employees >= 100) {
  score += 20; // Company size bonus  
}

if (email.domain.in(fortuneCompanies)) {
  score += 30; // Enterprise company
}

Pros & Cons

Advantages:

• Transparent and explainable
• Quick to set up and modify
• Works with small datasets
• Consistent results

Disadvantages:

• Requires manual tuning
• May miss subtle patterns
• Static unless manually updated
• Can become overly complex

AI-Powered Scoring

When to Use

• Large datasets - 1,000+ historical submissions with conversion outcomes
• Complex patterns - Multiple variables interact in non-obvious ways
• Dynamic markets - Customer profiles change frequently
• Optimization focus - Want maximum predictive accuracy

How It Works

Machine learning algorithms analyze historical data to identify patterns:

# Training data example
features = [
  'email_domain_type', 'job_seniority', 'company_size', 
  'industry', 'form_completion_rate', 'time_on_site'
]

# ML model learns: 
# VP + Healthcare + 100-500 employees = 87% conversion probability
# Manager + Manufacturing + <50 employees = 23% conversion probability

Model Types

1. Logistic Regression - Simple, interpretable probability predictions
2. Random Forest - Handles complex interactions, provides feature importance
3. Gradient Boosting - High accuracy, good with mixed data types
4. Neural Networks - Best for very large datasets with complex patterns

Pros & Cons

Advantages:

• Discovers hidden patterns
• Improves over time with more data
• Handles complex feature interactions
• Adapts to changing conditions

Disadvantages:

• Requires significant historical data
• Less transparent/explainable
• Can overfit to training data
• Needs ongoing monitoring

Hybrid Scoring

When to Use

• Best of both worlds - Want transparency AND pattern discovery
• Partial automation - Some criteria are clear, others need ML
• Risk management - Need guardrails on AI decisions
• Gradual transition - Moving from rules to AI over time

How It Works

Combines rule-based constraints with ML-powered insights:

// Hybrid approach example
let baseScore = mlModel.predict(formData); // AI prediction

// Apply business rules
if (email.includes('competitor.com')) {
  baseScore = 0; // Hard rule: block competitors
}

if (jobTitle.includes('Student', 'Intern')) {
  baseScore = Math.min(baseScore, 40); // Cap student scores
}

if (industry === 'target_vertical' && baseScore > 60) {
  baseScore += 15; // Boost for target industry
}

Implementation Strategies

1. Rule-First Hybrid

• Rules provide base score (0-60 points)
• AI provides bonus points (0-40 points)
• Transparent foundation with AI enhancement

2. AI-First Hybrid

• AI provides primary score (0-100 points)
• Rules apply adjustments and overrides
• ML handles complexity, rules handle edge cases

3. Ensemble Approach

• Multiple models vote on final score
• Different models for different scenarios
• Meta-model combines individual predictions

Choosing Your Model

Decision Matrix

Business Context Questions

Choose Rule-Based if:

• "We have clear qualification criteria from sales"
• "We need to explain every scoring decision"
• "We're in a regulated industry requiring transparency"
• "We don't have much historical data yet"

Choose AI-Powered if:

• "We have thousands of historical leads with conversion data"
• "Our sales team says lead quality is unpredictable"
• "We want maximum accuracy and don't mind black boxes"
• "We have data scientists to monitor and tune models"

Choose Hybrid if:

• "We want AI insights but need some business logic control"
• "We're transitioning from manual to automated scoring"
• "We have some clear rules but think there are hidden patterns"
• "We need a balance of transparency and accuracy"

Implementation Examples

Rule-Based Configuration

{
  "model_type": "rules",
  "criteria": [
    {
      "field": "jobTitle", 
      "condition": "contains",
      "values": ["VP", "Director", "Chief", "President"],
      "points": 25,
      "weight": 1.0
    },
    {
      "field": "company",
      "condition": "employee_count",
      "min": 100,
      "points": 20,
      "weight": 1.0
    },
    {
      "field": "email",
      "condition": "domain_type", 
      "values": ["business"],
      "points": 15,
      "weight": 1.0
    }
  ],
  "score_range": [0, 100],
  "grade_thresholds": {
    "hot": 80,
    "warm": 50,
    "cold": 0
  }
}

AI-Powered Configuration

{
  "model_type": "ml",
  "algorithm": "random_forest",
  "features": [
    "email_domain_category",
    "job_title_seniority",
    "company_size_bucket", 
    "industry_vertical",
    "form_completion_percentage",
    "geographic_region",
    "referral_source",
    "time_on_site_minutes"
  ],
  "training_data": {
    "min_samples": 1000,
    "conversion_window_days": 90,
    "positive_class_ratio": 0.15
  },
  "model_params": {
    "n_estimators": 100,
    "max_depth": 10,
    "min_samples_split": 50
  }
}

Hybrid Configuration

{
  "model_type": "hybrid",
  "primary_model": {
    "type": "ml",
    "weight": 0.7,
    "algorithm": "logistic_regression"
  },
  "secondary_model": {
    "type": "rules", 
    "weight": 0.3,
    "criteria": [...]
  },
  "business_rules": [
    {
      "condition": "email.domain.in(['competitor1.com', 'competitor2.com'])",
      "action": "set_score",
      "value": 0
    },
    {
      "condition": "jobTitle.includes('Student')",
      "action": "cap_score",
      "value": 30
    }
  ]
}

Model Performance

Evaluation Metrics

Monitor these metrics for all models:

A/B Testing

Compare model performance with controlled tests:

1. Split Traffic - Route 50% to Model A, 50% to Model B
2. Track Conversions - Monitor actual sales outcomes
3. Statistical Significance - Wait for adequate sample size
4. Roll Out Winner - Implement better-performing model

Best Practices

Starting Simple

1. Begin with rules based on sales team input
2. Collect data while running rule-based scoring
3. Train ML models once you have sufficient conversion data
4. Transition gradually to hybrid or AI-powered approaches

Continuous Improvement

• Monthly Reviews - Analyze model performance vs. actual conversions
• Quarterly Retraining - Update AI models with fresh data
• Feature Engineering - Add new data sources and derived fields
• Threshold Tuning - Adjust grade boundaries based on sales capacity

Avoiding Common Pitfalls

• Overfitting - Don't create rules for every edge case
• Bias Introduction - Be careful about demographic or geographic bias
• Stale Models - Regularly retrain AI models as markets change
• Ignoring Edge Cases - Have fallback logic for unusual submissions

Next Steps

After choosing your scoring model:

• Define evaluation criteria that align with your model choice
• Set up routing rules based on scores
• Monitor performance and optimize over time
• Implement A/B testing to compare approaches