Independent R&D · Agentic AI

Sales Decision
Intelligence Platform

SDIP transforms raw Salesforce CRM data into predictive, explainable, and actionable intelligence — combining machine learning, rule-based reasoning, and agentic AI into a single sales copilot.

Explore Architecture → View Modules
Why SDIP

Five Business Objectives

Built for enterprise sales organisations that need to move from gut-based forecasting to a data-driven, agentic intelligence model.

01
Improve Win Rates
Identify high-probability opportunities early and focus sales effort where it matters most.
02
Forecast Accuracy
Replace gut-based forecasting with probability-weighted, data-driven predictions.
03
Effort Optimisation
Reduce wasted effort on low-probability deals. Prioritise revenue-impacting opportunities.
04
Leadership Decisions
Give executives clear, explainable insights — not raw CRM reports.
05
Agentic Copilot
Move from dashboards to an intelligent assistant that reasons, explains, and recommends actions.
System Design

High-Level Architecture

Six logical layers from raw CRM data ingestion through to conversational UI — each with a single, clear responsibility.

Layer 1
📥
Data Ingestion Layer
Loads & validates Salesforce SFDC opportunity data. CSV-based today, SFDC-API ready for production.
Layer 2
🧠
Predictive Intelligence Layer
XGBoost V3 ML model + rule-based fallback. Predicts 90-day close probability per opportunity.
Layer 3
🎯
Recommendation Engine
Converts probability scores into actionable next steps: Close Quickly, Engage Customer, or Monitor.
Layer 4
⚡
Agentic Reasoning Layer
Prioritises deals, explains probability drivers, highlights revenue at risk, suggests focus areas. True agentic AI.
Layer 5
🔌
API Layer (FastAPI)
/login · /deals endpoints. Credential + token-based session security. Serves enriched predictions to UI.
Layer 6
📊
Conversational UI + Dashboard
3-panel layout: Top Opportunities · Chatbot · KPIs/Charts. Bar & pie visualisations via Chart.js.
End-to-End Intelligence Pipeline
1
SFDC Data
CRM Input
2
ML Model
XGBoost V3
3
Win Prob
0 – 100%
4
Agent Reason
What next?
5
Action
Close / Engage
Core Modules

Five Functional Agents

Each module is a dedicated agent with a single responsibility, enabling clean separation of concerns and future extensibility.

Data Layer
Data Loader Agent
Loads and validates opportunity data with schema consistency, auto-correction of missing values, and default probability initialisation.
Auto-correction of missing numeric fields
Schema validation & default probability init
SFDC API-ready in production mode
ML · Primary
Win Probability Prediction Agent
Predicts close probability using the trained XGBoost model. Falls back to transparent rule-based logic when the model is unavailable.
Deal amount · Historical win rate · Sales stage
Forecast confidence · Recent activity signals
CEO-friendly Win_Reason explanations
ML Training
ML Training Agent
Trains and continuously improves win probability models — from Logistic Regression baseline through to AutoML optimisation.
V1 Logistic Regression → V2 Random Forest
V3 XGBoost (current) → V4 AutoML (roadmap)
Target: Closed Won=1, all other stages=0
Execution
Recommendation Agent
Bridges analytics and execution by converting probability scores into clear, prioritised next actions for each opportunity.
High prob + High value → Close Quickly
Medium probability → Engage Customer
Low probability → Monitor / Low Priority
Intelligence
Agentic Reasoning Layer
The brain of SDIP. Answers "What should we do next?" with deal prioritisation, probability driver explanations, and revenue-at-risk signals.
Deal prioritisation with explainability
Revenue at risk identification
Focus area suggestions for leadership
Model Evolution

ML Model Progression

Version Model Strategic Value Status
V1 Logistic Regression High transparency — baseline interpretability Completed
V2 Random Forest Captures nonlinear patterns in deal data Completed
V3 XGBoost CURRENT Enterprise-grade accuracy with feature importance Active
V4 AutoML Continuous optimisation — self-improving pipeline Roadmap
Technology

Tech Stack

Production-grade, open-source technology choices that prioritise explainability, performance, and developer velocity.

Backend & Data
Python Pandas NumPy XGBoost Scikit-learn Joblib
API & Runtime
FastAPI Uvicorn REST API ASGI
Frontend & UI
HTML / CSS JavaScript Chart.js Responsive UI
Security
Credential Auth Token Sessions API Security
Business Value

Executive Dashboard KPIs

Four key metrics surfaced on the dashboard — giving leadership an instant read on pipeline health.

📦
Total Pipeline
All Opportunities
📈
Weighted Forecast
Prob × Deal Value
✅
High-Confidence Revenue
Win Prob > 70%
⚠️
Revenue at Risk
Stalled Deals
What's Next

Future Roadmap

Version 1 answers one high-impact question. The roadmap expands SDIP into a full autonomous sales intelligence platform.

Multi-Agent Deal Coaching
Specialised sub-agents that coach individual account executives with deal-specific guidance and objection handling strategies.
Risk Factor Decomposition
Granular breakdown of what is driving risk in each deal — competitive threats, stakeholder silence, budget freeze signals.
Customer Sentiment Ingestion
Pull email, call transcript, and support ticket sentiment into the probability model for richer deal signals.
Pricing & Discount Intelligence
Correlate discount patterns with win rates to recommend optimal pricing strategies per account segment.
Autonomous Pipeline Health Agent
A fully autonomous agent that monitors pipeline health 24/7, surfaces anomalies, and proactively alerts leadership before deals go dark.
VD
Viswanathan Durairaj
Independent R&D · Agentic AI & Enterprise Architecture

SDIP is an independent research and development initiative — exploring how agentic AI can transform the way enterprise sales organisations make decisions. Built from first principles: ML-first, explainability-driven, and designed for real leadership questions.

Connect on LinkedIn 🌐 Vision Lab
← Vision home