Banks, NBFCs and fintech companies hire many analysts because their business runs on data: every loan, payment and repayment is a record. Knowing the core metrics helps you understand projects in this domain and talk confidently in interviews.
Lending basics in one paragraph
A lender gives a loan (disbursal), the borrower repays in instalments (EMIs) with interest, and the lender earns the interest minus its costs and losses. The main risk is that some borrowers stop paying. Much of lending analytics is about growing loans while keeping that risk under control.
Key lending metrics
| Metric | What it means |
|---|---|
| Disbursals | Amount of new loans given in a period |
| Loan book / AUM | Total outstanding loans |
| DPD (days past due) | How many days an EMI is overdue |
| Delinquency buckets | Loans grouped by DPD, such as 1-30, 31-60, 61-90, 90+ |
| NPA | Loans overdue beyond a regulatory threshold, treated as non-performing |
| Collection efficiency | Amount collected ÷ amount due |
| LTV (loan to value) | Loan amount ÷ value of the collateral, important for gold and home loans |
| Approval rate | Approved applications ÷ applications |
| Roll rate | Share of loans moving to a worse bucket from one month to the next |
Payments metrics
For UPI, cards and wallets:
- Success rate: successful transactions ÷ attempts.
- Failure reasons: technical decline, insufficient balance, wrong PIN, timeout.
- Failures by bank, app, time of day and amount band.
- Settlement and reconciliation gaps between internal records and partner reports.
Typical analysis questions
- Why did delinquency in a product or region rise in the last quarter?
- Which customer segments roll from early to late delinquency fastest?
- Are we approving riskier customers than before?
- Which branches have collection efficiency well below the rest?
- When gold prices fall, how many gold loans cross the safe LTV limit?
- Which partner bank causes most payment failures, and at what time of day?
Common analysis techniques
- Vintage analysis: group loans by disbursal month and track how each group performs over time, so you compare like with like.
- Cohort and bucket movement: roll-rate matrices showing how loans move between DPD buckets.
- Segmentation: by product, ticket size, region, channel and customer profile.
- Risk models: logistic regression and tree models to estimate default probability.
Skills that matter
SQL for large transaction tables, careful date handling for DPD and vintages, Excel or Power BI for management reports, and a clear understanding of definitions. In banking, a metric calculated with the wrong definition can mislead a credit committee.
Practise with a realistic brief
Our banking and finance projects put these ideas to work: the gold loan LTV risk project is about collateral risk, and the UPI payment failures project is a payments investigation. For default prediction, start with the logistic regression lesson.
Practise what you just read
Reading helps; writing the code yourself is what sticks. Start here: