DealSense AI: Production ML Deal Risk Intelligence

$1.4M/quarter revenue impact – ML-powered deal risk scoring for B2B SaaS CROs

Random Forest predicts deal loss probability (0-100 score)
Slack alerts + "Exec sponsor TODAY" actions
Production-ready: Airflow ETL + joblib models + API endpoints
Custom metrics uncover sales cycle + rep performance gaps
ROI: Saves 20 high-risk deals/quarter = $1.4M revenue

---

Table of Contents

1. The Business Problem
2. System Architecture
3. Quick Setup Steps
4. Project Structure
5. Technical Deep Dive
6. Sample Output
7. Business Results
8. Production Architecture
9. Assignment Parts
10. Tech Stack

The Business Problem

Sales cycles up 214%, win rates volatile. Sales leaders need:

• Early warning on at-risk pipeline
• Explainable risk factors (rep performance, stage delays)
• Actionable interventions ("Exec sponsor call TODAY")

DealSense AI delivers daily intelligence that turns data into revenue protection.

System Architecture

How to Run and Use This Project

Follow these steps to set up, run, and use DealSense AI:

1. Clone the repository

   git clone https://github.com/username/dealsense-ai
   cd dealsense-ai

2. Install dependencies

   pip install -r requirements.txt

3. Run the EDA notebook ( for data exploration)

   jupyter notebook 02_eda_insights_skygeni.ipynb

4. Run the main risk scoring notebook

   jupyter notebook notebooks/03_decision_engine_deal_risk.ipynb

5. Check outputs

   • Risk reports and feature importance will be generated in the results/ folder.
   • Example: open results/feature_importance.png for feature importance visualization.

6. Production code

   • The main ML engine is in src/deal_risk_scoring.py and can be imported or run as a script for integration into other systems.

7. Documentation

   • See the docs/ folder for business framing, system design, and project reflection.

Project Structure

.
├── docs/
│   ├── 01_problem_framing.md         # Business context and challenge
│   ├── 04_system_design.md           # Architecture 
│   └── 05_reflection.md              # Project learnings and insights
├── notebooks/
│   └── 03_decision_engine_deal_risk.ipynb  # ML risk scoring 
├── results/
│   └── feature_importance.png        # Visualized feature importance
├── src/
│   └── deal_risk_scoring.py          # Main ML scoring logic 
├── 02_eda_insights_skygeni.ipynb     # EDA notebook:data analysis 
├── CRM_to_Risk_Scoring_Engine.png    # System architecture diagram
├── LICENSE                           # Project license
├── README.md                         # Project overview 
├── requirements.txt                  # Python dependencies
└── skygeni_sales_data.csv            # Raw sales data (5K deals)

Note: Documentation files use their actual project file names (e.g., 01_problem_framing.md, 04_system_design.md, 05_reflection.md) to match the naming convention in the docs folder.

Technical Deep Dive

ML Pipeline

Raw CRM Data → Feature Engineering → Hybrid Scoring → Exec Outputs

Feature Engineering (15+ features):

• rep_historical_winrate, industry_winrate, deal_age_percentile
• is_long_cycle, is_large_deal, lead_source_quality

Hybrid Scoring Engine:

• Rule-based (40 pts): Cycle length, rep performance, stage delays
• ML-based (60 pts): Random Forest (ROC-AUC validated)
• Combined: 0-100 risk score + top 3 factors

Production Features:

• joblib model serialization
• Airflow/cron scheduling ready
• Slack/email alert integration
• Fallback logic (rules if ML fails)

Sample Output

CRITICAL: D12345 ($45K ACV)

Risk Score: 87/100 | Level: CRITICAL

Top Factors:

1. Rep bottom 20% win rate (25 pts)
2. 95 days in Demo (30 pts)

→ ACTION: Exec sponsor TODAY

Business Results

• 214% sales cycle increase diagnosed
• Custom metrics: PVS, REI created
• $1.4M/quarter revenue protection potential
• Daily exec reports + Slack alerts ready

Production Architecture

Salesforce API → Airflow ETL (6AM) → ML Scoring → Slack/Email → Tableau

        ↓

      CRO Dashboard (real-time)

Full system design: docs/04_system_design.md

Why This Matters

Most sales tech reports data. DealSense AI prescribes actions:

"This rep needs coaching"

"Escalate this deal to VP Sales"

"Partner leads failing → audit channel"

File Descriptions

1. Problem Framing
   docs/01_problem_framing.md
   Defines the business context, objectives, and challenges for deal risk scoring.

2. EDA & Insights
   02_eda_insights_skygeni.ipynb
   Exploratory data analysis notebook: uncovers key patterns, trends, and actionable insights from the sales data.

3. Decision Engine & Risk Scoring
   notebooks/03_decision_engine_deal_risk.ipynb, src/deal_risk_scoring.py
   ML pipeline and production code for scoring deal risk, including model training, evaluation, and demo analysis.

4. System Design
   docs/04_system_design.md
   Technical documentation of the system architecture, data flow, and integration points.

5. Reflection
   docs/05_reflection.md
   Project learnings, challenges faced, and key takeaways from the development and deployment process.

Tech Stack

• ML: scikit-learn, Random Forest, joblib
• Visualization: Plotly, Matplotlib
• Data: pandas, numpy
• Production: Airflow-ready, API endpoints

---

Final Note

Thank you for exploring DealSense AI! This project is designed to empower sales teams with actionable intelligence, robust analytics, and production-ready machine learning. Whether you’re a data scientist, engineer, or business leader, we hope this solution inspires you to drive smarter decisions and unlock new revenue opportunities.

Innovate boldly, automate wisely, and let data lead your growth!