Skip to content

04_system_design.md

Latest commit

 

History

History
75 lines (62 loc) · 2.8 KB

04_system_design.md

File metadata and controls

75 lines (62 loc) · 2.8 KB

SkyGeni Deal Risk Decision Engine: System Design

Overview

The SkyGeni Deal Risk Decision Engine is designed to provide daily, actionable risk insights for B2B SaaS sales teams. It combines rule-based and machine learning (ML) models to score deals, generate risk reports, and highlight key risk factors for executive decision-making.

🎯 Data Flow

flowchart TD
    CRM[CRM (Salesforce/HubSpot)] --> ETL[Daily ETL Job (6AM PST)]
    ETL --> FeatureStore[Feature Store]
    FeatureStore --> RiskEngine[Risk Scoring Engine\n(Rule-based + ML)]
    RiskEngine --> Reports[Risk Reports + Alerts]
    Reports --> Channels[Email/Slack/Tableau]
Loading

🏗️ Architecture Components

See below for detailed architecture.

⏰ Scheduling

  • Daily 6AM PST via Airflow/cron
  • Real-time scoring on deal stage changes (future)
  • SLA: Reports delivered by 7AM PST

🚨 Example Alerts & Insights

🚨 CRITICAL ALERT (3 deals @ risk):

  • D12345 (ACV $45k) - Rep bottom 20% win rate, 95 days in Demo → ACTION: Executive sponsor call TODAY
  • D67890 (ACV $28k) - Partner lead + Proposal stage 60+ days → ACTION: Verify buyer authority + SE review

📊 DAILY SUMMARY: 12 high-risk deals ($1.4M ACV @ risk)

⚠️ Failure Cases & Mitigations

  1. No CRM data sync → Send "data unavailable" alert
  2. ML model drift → Auto-fallback to rule-based scoring
  3. High-risk deal volume spike → Escalate to CRO dashboard
  4. Data quality issues → Flag deals with >3 missing fields

Architecture

  • Data Ingestion:
    • Source: CRM export (CSV or API integration)
    • Automated daily data pull (future integration)
  • Feature Engineering:
    • Custom feature generation (e.g., deal age, activity counts, engagement metrics)
    • Handled in src/deal_risk_scoring.py
  • Risk Scoring:
    • Rule-based scoring for transparency and quick wins
    • ML-based scoring (Random Forest) for predictive accuracy
  • Reporting:
    • Daily risk report CSVs
    • Feature importance plots
    • Executive summaries
  • Output:
    • Results saved to results/ folder
    • Ready for dashboard or presentation integration

Integration & Automation

  • Current: Manual notebook execution
  • Future:
    • Schedule as a daily job (e.g., with Airflow, cron, or cloud scheduler)
    • Integrate with CRM via API for real-time scoring
    • Automated email or dashboard delivery of risk reports

Monitoring & Maintenance

  • Track model performance (ROC-AUC, precision/recall)
  • Monitor data drift and retrain ML model as needed
  • Log daily outputs and errors for auditability

Security & Compliance

  • Handle sensitive sales data securely
  • Ensure compliance with company data policies

This system is designed for extensibility, automation, and executive usability, supporting SkyGeni's goal of data-driven sales excellence.