Market Drift Monitor Skill

This skill provides specialized procedures for auditing lead intelligence data quality and detecting market drift in real estate lead generation performance. It leverages the logic from the EnhancedLeadScoringService to identify shifts in lead behavior and source effectiveness.

Core Audit Procedures

1. Source Quality Distribution Audit

Monitor the distribution of lead sources and their relative quality scores to detect if high-performing channels are degrading.

Key Metrics:

• Source Mix: Percentage distribution across LeadSourceType (REFERRAL, ORGANIC_SEARCH, PAID_SEARCH, etc.).
• Quality Velocity: Change in average source_quality_score over a 30-day window.
• Attribution Gap: Ratio of UNKNOWN sources vs. trackable sources.

SQL Pattern:

SELECT 
    source_type, 
    COUNT(*) as lead_count, 
    AVG(source_quality_score) as avg_quality,
    AVG(conversion_likelihood) as avg_conv_likelihood
FROM lead_intelligence_scores
WHERE location_id = :location_id 
  AND scoring_timestamp > NOW() - INTERVAL '30 days'
GROUP BY source_type
ORDER BY avg_quality DESC;

2. Conversion Readiness Trend Analysis

Detect "Market Drift" by analyzing the shift in closing_probability and conversion_readiness_score.

Drift Indicators:

• Score Compression: Increasing volume of leads in the "Lukewarm" (50-69) range with fewer in "Hot" (85+).
• Signal Decay: Decrease in the frequency of positive_signals (e.g., "High financial readiness") relative to risk_factors.

3. Behavioral Pattern Audit

Monitor the evolution of prospect engagement by auditing BehavioralSignals.

Audit Targets:

• Urgency Frequency: Count of urgency_indicators (asap, urgent, etc.) per 1000 messages.
• Authority Signal Ratio: Presence of decision_authority_signals in initial qualification turns.
• Objection Density: Growth in objection_patterns (too expensive, market uncertainty).

Data Integrity Verification

Ensure the following integrity constraints are maintained across the PostgreSQL lead_intelligence_v2 schema:

1. Tenant Isolation: Verify all queries include location_id filtering.
2. Feature Completeness: Audit records for missing utm_source or utm_medium which causes attribution_confidence to drop below 0.5.
3. ML Feature Drift: Monitor the avg_message_length and question_count used as features for the closing_probability_model.

Recommended Actions on Drift Detection

• Source Pivot: If PAID_SEARCH quality drops >15%, recommend shifting budget to REFERRAL or ORGANIC_SEARCH.
• Nurture Adjustment: If urgency_indicators decline, suggest updating the "Warm-Lead" nurture sequences with more educational content.
• Model Retraining: If ml_confidence drops below 0.6 consistently, trigger a request for model retraining on recent 90-day data.

Integration Points

• Service: ghl_real_estate_ai.services.enhanced_lead_scoring.EnhancedLeadScoringService
• Model: ghl_real_estate_ai.ml.closing_probability_model
• Database: PostgreSQL (pgvector enabled for behavioral similarity)