What You Will Do

- Build and own credit risk strategies for the lending portfolio underwriting policy, cut-offs, risk-based limits and

pricing, and line management.
- Design decisioning logic end-to-end — bureau rules, affordability checks (FOIR/DBR), fraud screens, policy waterfalls

— quantifying the approval-rate vs. loss trade-off of every rule before it goes live.
- Build risk models and scorecards hands-on — application, behavioural and collections scoring, from data prep and

feature engineering through to a deployed, monitored model. You write the code.
- Own the models after they go live — retrain, recalibrate and rebuild as they decay. Apply model risk management

discipline: validation, back-testing, documented assumptions/limitations, stability monitoring, and governance on when a model can be trusted or must be retired.
- Evaluate and onboard alternate data (AA / bank statements, device, telecom, GST, UPI, employer signals) — proving

incremental lift over the existing stack before recommending spend.
- Run controlled experiments — champion/challenger, swap-set analysis, structured test-and-learn — so every

strategy change has an attributable outcome; and monitor portfolio health (vintage curves, roll/flow rates, PAR/DPD, early-warning indicators, FPD, cohort loss forecasting).
- Communicate to senior stakeholders and partner cross-functionally — translate a KS statistic into a business

decision, and work with Product, Engineering, Business, Collections and Compliance to take strategies from analysis to production. Bureau & Credit Decisioning Depth (non-negotiable)
- Hands-on with CIBIL, CRIF High Mark, Experian and/or Equifax data — raw report structure,



tradeline-level parsing,

DPD strings and enquiry data.
- Strong grasp of individual bureau parameters and their marginal impact on decisioning — score bands and

thin-file/NTC handling, DPD history and worst delinquency, trade vintage, enquiry velocity, credit hunger, secured/lending mix, utilisation, write-off/settlement/suit-filed flags, and overleverage.
- Ability to build derived bureau attributes, test predictive power (IV, WOE, univariate lift), and defend why a variable

belongs — or does not — in a policy or scorecard.
- Understanding of bureau data limitations (reporting lags, quality gaps, duplicate/merged files, matching logic) and

the regulatory frame — RBI Digital Lending Guidelines, fair lending, DPDP — well enough to build strategies that survive an audit.

What We Are Looking For

Quantitative & statistical foundation

- Command of statistics — distributions, hypothesis testing, confidence intervals, correlation vs. causation, and

knowing when a result is noise.
- Modelling — logistic regression, gradient boosting (XGBoost/LightGBM), decision trees; bias-variance and

overfitting in low-default settings.
- Risk measurement — KS, Gini/AUC, PSI/CSI, IV & WOE binning, lift/gains, segmentation, survival and

vintage/cohort analysis.
- Reject inference,



sample bias and selection effects — why a model trained only on approved customers can quietly

mislead you.

Technical proficiency

- Python (pandas, NumPy, scikit-learn, statsmodels) — production-grade — and advanced SQL (window functions,

CTEs, complex joins on large loan-level tables).
- R for statistical modelling/validation; Tableau (or Power BI/Looker) for actionable risk dashboards; advanced Excel.
- Git, scripting/automation for recurring monitoring, and big-data tooling (PySpark, Hive/Presto) for data at scale.
- Valuable to have: decisioning/rule engines (Blaze, Drools, FICO, in-house BRE), API-based bureau integrations, and basic

MLOps/model deployment. Experience & domain

- Minimum 3 years in credit risk analytics and modelling within lending (NBFC, bank, fintech, digital lending).
- Hands-on credit risk modelling on Indian lending data is a hard requirement — you know this market's quirks

(bureau coverage, thin-file/NTC, informal-income borrowers, how Indian repayment behaviour breaks textbook assumptions). Unsecured lending portfolio experience strongly preferred.
- Demonstrated ownership of a model or scorecard that went live and moved a number (approval rate, loss rate, or

profitability) — ready to walk through data, modelling choices, validation and outcome.
- Bachelor's/Master's in a quantitative discipline (Statistics, Maths, Economics, Engineering). Working grasp of unit

economics — how approval rate, loss, yield, opex and collections combine into portfolio profitability.
- Great to have: CQF (Certificate in Quantitative Finance) or an equivalent formal quant credential.

📌 Credit and Risk Analyst (Bengaluru)
🏢 Benchire
📍 Bengaluru