Financial Services 360 — Churn Analytics with Salesforce Data Cloud & Einstein Discovery

Purpose

Enablement-style demo showing how to model customer churn using Salesforce Data Cloud (DMOs + Calculated Insights), explain predictions with Einstein Discovery, and tell the story in CRM Analytics (Tableau CRM). Data is fully synthetic.

---

Architecture

flowchart LR
  A[Data Sources (CSV)] --> B[Data Cloud DLOs]
  B --> C[DMOs: Customer, Account, Transaction]
  C --> D[Calculated Insights (features per customer)]
  D --> E[Einstein Discovery Model (churn + drivers)]
  E --> F[CRM Analytics Dashboard (FS 360)]
  F --> G[Action Plan (Next-Best Action)]

See: assets/architecture.mmd

---

Repository Layout

financial-services-360-data-cloud-ai/
├── README.md
├── requirements.txt
├── .gitignore
├── data/                      # synthetic CSVs
├── notebooks/
│   ├── 01_generate_synthetic_data.ipynb
│   └── 02_feature_engineering.ipynb
├── datacloud/
│   ├── calculated_insights.sql
│   └── dmo_relationships.md
├── ed/
│   ├── training_dataset.csv
│   ├── scoring_dataset.csv
│   └── model_readme.md
├── crm_analytics/
│   ├── dashboard_spec.md
│   └── screenshots/
└── assets/
    └── architecture.mmd

---

Quickstart

1) Local setup

python -m venv .venv
# Windows: .venv\Scripts\activate
source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook

2) Generate synthetic data

Open notebooks/01_generate_synthetic_data.ipynb and run all cells.
CSV outputs land in /data.

3) Configure Data Cloud

1. Create Data Streams from /data/*.csv
2. Define DMOs: Customer (Profile), Account (Profile Object), Transaction (Event)
3. Relationships: Customer 1-* Account 1-* Transaction
4. Run Calculated Insights (datacloud/calculated_insights.sql) → produces feature table per customer

4) Einstein Discovery

• Train model with target churn_90d using /ed/training_dataset.csv
• Capture top drivers and reason codes; add notes in ed/model_readme.md

5) CRM Analytics

• Build an exec dashboard per crm_analytics/dashboard_spec.md
• Add screenshots to crm_analytics/screenshots/

6) Summary

High-fee, low-tenure, low-balance customers are 2.3× more likely to churn. Prioritize fee reversal outreach and savings cross-sell for retention.

---

---

Databricks (PySpark)

See /databricks/ for a PySpark notebook that computes the same features at scale in Databricks Community Edition.

Tableau

See /tableau/ for steps to create a lightweight executive dashboard using a sample CSV.

Data Wrangling

Primary wrangling is done in Python (pandas notebooks). The Databricks path demonstrates PySpark for scale.