Data Science · Indian Banking

How Data Science Is Used in Indian Banking Companies Today

SECTION 01Fraud detection at UPI scale

Fraud detection is the highest-impact data science application in Indian banking — and the scale is staggering.

• UPI processes over 14 billion transactions per month — roughly 500 million a day. Every transaction is scored in real time.
• Real-time scoring: Models must decide in under 100 milliseconds whether to approve, flag, or block a transaction.
• Gradient boosting dominates: XGBoost, LightGBM, and CatBoost are standard for tabular transaction data.
• Feature engineering matters most: Transaction velocity, device fingerprint, merchant risk, geolocation, and time since last transaction.
• Class imbalance: Fraud is often less than 0.1% of transactions. Handling imbalance is a core skill.
• Graph analytics: Banks detect fraud rings using network analysis of shared devices, IPs, and accounts.
• Real example: HDFC Bank uses ML models to score card and UPI transactions, flagging suspicious ones for manual review.
• Real example: ICICI Bank reported a significant reduction in fraud losses after deploying real-time ML scoring on digital transactions.
• Real example: NPCI uses ML for UPI fraud detection across the entire network, in partnership with banks and payment providers.
• False positive cost: Declining a legitimate transaction loses a customer. Precision matters as much as recall.
• Continuous retraining: Confirmed fraud cases feed back into models — often daily or weekly.

SECTION 02Credit scoring and financial inclusion

India's credit market has been transformed by machine learning and alternative data.

• Traditional scoring: CIBIL, Experian, and Equifax scores based on payment history and credit utilisation.
• Alternative data: UPI transaction history, mobile usage, utility payments, and e-commerce behaviour.
• Thin-file borrowers: ML models can score customers with little or no formal credit history — a huge opportunity in India.
• Real example: HDFC Bank uses ML models for personal loans and credit cards, combining bureau data with internal behavioural signals.
• Real example: ICICI Bank's instant personal loan product uses ML to approve loans in minutes for pre-qualified customers.
• Real example: Axis Bank uses ML for MSME lending, evaluating cash-flow patterns from GST and bank statements.
• Real example: Kotak Mahindra Bank uses ML to personalise credit offers and pricing for retail customers.
• Account Aggregator framework: RBI's Account Aggregator system lets customers share financial data securely — enabling much richer credit models.
• Explainability is mandatory: RBI and internal risk teams require reason codes. SHAP and LIME are standard.
• Fairness and bias: Models must not discriminate by gender, region, or caste. Fairness auditing is now standard practice.

SECTION 03Customer analytics and personalisation

Indian banks use data science to keep customers, grow revenue, and compete with fintechs.

• Churn prediction: Identifying customers likely to leave so retention teams can act with targeted offers.
• Customer lifetime value (CLV): Predicting the total revenue a customer will generate over their relationship with the bank.
• Next best action: Recommending the next product or offer for each customer — credit card, insurance, or investment product.
• Cross-sell and up-sell: Identifying which customers are likely to buy additional products.
• Real example: HDFC Bank uses ML to identify pre-approved customers for credit cards and personal loans, reducing marketing cost per acquisition.
• Real example: ICICI Bank uses ML for its "Insta" product family — instant savings accounts, credit cards, and loans.
• Real example: SBI uses data analytics for its YONO digital banking platform to personalise offers across millions of customers.
• Segment-of-one: ML models can tailor pricing, offers, and communication to individual customers, not just broad segments.
• Digital behaviour analytics: Understanding app usage, drop-off points, and feature adoption to improve digital products.
• Complaint and sentiment analysis: NLP on customer feedback, call centre transcripts, and social media.

SECTION 04Operations, AML, and compliance

Beyond customer-facing work, data science drives back-office efficiency and regulatory compliance.

• AML (Anti-Money Laundering): ML models detect suspicious transaction patterns for reporting to FIU-IND.
• KYC (Know Your Customer): OCR, face matching, and document verification automate customer onboarding.
• Collections optimisation: ML models predict which delinquent customers are likely to pay, and which need early intervention.
• Real example: Indian banks use ML-driven collections models to prioritise outreach and reduce non-performing assets (NPAs).
• Loan default prediction: Early warning systems that flag accounts at risk before they become NPAs.
• Regulatory reporting: ML and automation support RBI reporting, Basel capital calculations, and IFRS 9 provisioning.
• Operational efficiency: Automating back-office processes like reconciliation, reporting, and document processing.
• Call centre analytics: NLP on call transcripts to identify customer issues, compliance risks, and coaching opportunities.
• Fraud ring detection: Graph analytics to detect coordinated fraud across accounts, devices, and merchants.
• Real example: SBI and other large banks use ML for AML transaction monitoring — a regulatory requirement.

SECTION 05Real examples from Indian banks

Here is a summary of how specific Indian banks and institutions use data science today.

• HDFC Bank: Fraud detection for cards and UPI, ML-based credit scoring, cross-sell models, collections optimisation, and a dedicated data science team.
• ICICI Bank: Instant personal loans, credit cards, pre-approved offers, fraud detection, and the "Insta" product family driven by ML.
• State Bank of India (SBI): YONO digital platform analytics, AML monitoring, credit scoring, and large-scale analytics across its customer base.
• Axis Bank: MSME lending models, credit card analytics, fraud detection, and customer segmentation.
• Kotak Mahindra Bank: Personalised offers, pricing models, fraud detection, and digital customer analytics.
• Punjab National Bank: Credit risk models, AML, and operational analytics.
• NPCI: Fraud detection across the UPI network, transaction analytics, and real-time monitoring.
• Bajaj Finserv: ML for lending decisions, collections, and cross-sell across its NBFC and insurance businesses.
• Paytm: Fraud detection, credit scoring, and customer analytics for its payments and financial services products.
• HDFC Life and ICICI Prudential: Insurance-specific ML for underwriting, claims, and customer analytics.

SECTION 06Skills and jobs in banking data science

Here is what Indian banks actually hire for — and what they pay.

• Python + SQL: Non-negotiable. Every banking data science role requires both.
• Gradient boosting: XGBoost, LightGBM, and CatBoost dominate tabular banking data.
• Feature engineering: The skill that separates good models from great ones in banking.
• Imbalanced learning: SMOTE, class weights, focal loss, and appropriate metrics.
• Explainability: SHAP, LIME, and reason codes for regulatory compliance.
• Time series: For collections, forecasting, and risk models.
• Cloud ML: AWS SageMaker, Azure ML, or GCP Vertex AI for deployment.
• Domain knowledge: Credit, risk, fraud, or payments — pick one and go deep.
• Fresher salaries: ₹6–12 LPA for strong candidates from top institutes or with strong portfolios.
• Mid-level (3–5 years): ₹15–30 LPA in private banks and NBFCs.
• Senior (6+ years): ₹30–60 LPA for lead and manager roles in large banks and fintechs.
• Consulting firms: Accenture, Deloitte, and EY hire data scientists for BFSI projects, often at comparable salaries.

SECTION 07How to break into banking data science

Banking data science is competitive but accessible. Here is how to get in.

• Learn the fundamentals: Python, SQL, statistics, and machine learning. No shortcuts.
• Understand banking basics: Loans, interest, risk, credit, and regulatory concepts.
• Build a fraud detection project: Use the Kaggle Credit Card Fraud dataset as a starting point.
• Build a credit scoring project: Use Lending Club or Home Credit data, and explain the model with SHAP.
• Learn explainability: SHAP and LIME are standard in Indian banking because of RBI requirements.
• Practice class imbalance: Fraud and default datasets are highly imbalanced. Learn SMOTE, class weights, and evaluation metrics.
• Understand Account Aggregator: RBI's framework is reshaping credit data in India. Learn how it works.
• Target the right companies: Banks, NBFCs, fintechs, and BFSI-focused consulting firms.
• Network in BFSI communities: LinkedIn, analytics meetups, and industry conferences.
• Consider certifications: FRM and CQF are valuable for risk-focused roles. Not required, but they help.

SECTION 08Test yourself — Data science in Indian banking

SECTION 09Frequently asked questions

SECTION 10Related reads