market-drift-monitor_skill

Overview

This skill monitors PostgreSQL lead intelligence data quality to detect market drift in lead source effectiveness, conversion readiness, and ROI metrics. It focuses on per-tenant audits, behavioral signal validation, and ML feature integrity for lead scoring models. The skill surfaces actionable alerts and remediation recommendations to preserve pipeline performance. It is designed for real-estate and multi-tenant CRM environments.

How this skill works

The skill runs periodic SQL audits against the lead_intelligence_v2 schema to quantify source mix, source_quality_score trends, and attribution gaps over configurable windows (default 30 days). It analyzes conversion readiness by tracking closing_probability, conversion_readiness_score, and signal patterns like urgency and authority indicators. Data integrity checks enforce tenant isolation, feature completeness (utm fields), and ML feature drift (avg_message_length, question_count, ml_confidence). When thresholds are crossed, the skill recommends concrete actions such as source pivoting, nurture changes, or model retraining.

When to use it

• Auditing lead generation performance across tenants after a campaign change or ad spend shift
• Validating data integrity before retraining or redeploying a closing probability model
• Detecting early signs of market drift in conversion readiness or behavioral signals
• Prioritizing remediation when attribution confidence or source quality degrades
• Auditing multi-tenant isolation to prevent data bleed between locations

Best practices

• Always include location_id filters in queries to enforce tenant isolation
• Use a rolling 30–90 day window for trend detection and a shorter window for alerts
• Monitor both aggregate metrics (avg_quality) and distribution shifts (score compression)
• Set actionable thresholds (e.g., PAID_SEARCH drop >15%, ml_confidence <0.6) and automate alerts
• Correlate behavioral drift with recent marketing or product changes before acting

Example use cases

• Detecting a decline in paid-search lead quality and recommending budget reallocation to referrals
• Triggering model retraining when ml_confidence falls below 0.6 across the last 90 days
• Identifying increased objection density that signals a need to update nurture content
• Flagging high UNKNOWN attribution ratios to prioritize UTM capture fixes
• Comparing urgency frequency and authority signal ratios before and after a pricing update