Ozon Bank is a separate division of Ozon where everything related to finance and IT is closely intertwined. We create new products and services for individuals and legal entities. We are proud of the atmosphere in our teams: every employee can influence processes and paths to results.
In the B2B Risk Analytics team, we are looking for an analyst with a strong technical background – this is a role at the intersection of ML and business. Scoring models here are not an abstraction, but systems that make decisions involving money every day: you will validate them as ML systems – ranking, calibration, stability, correctness of application – and turn insights into limit strategies, cut-offs, and stop rules with a measurable impact on P&L.
If you are a DS or ML engineer who has outgrown pure modeling and wants to own not only model metrics but also its financial impact, this is the role for you. We can teach you the credit domain; bring your deep understanding of models and data with you.
Our stack: Python (pandas, numpy, scikit-learn, CatBoost/LightGBM, statsmodels), SQL (Vertica, ClickHouse, PostgreSQL), Airflow, Superset.
Your responsibilities will include
- Validating scoring and classical ML models: ranking ability (Gini, KS), calibration (calibration curves, Brier score), stability over time (PSI/CSI by score and features), out-of-time and out-of-sample backtests, detecting target leakage and checking the correctness of model application in the business process.
- Participating in model development and improvement: error analysis, feature engineering on transactional data and bureau data, interpretability (SHAP, WoE/IV), champion/challenger comparisons.
- Developing and optimizing limit strategies, cut-offs, and stop rules: threshold optimization through expected loss, strategy simulation on historical data, swap-set analysis.
- Monitoring strategy effectiveness across cohorts, vintages, and segments: Approval Rate, Take Rate, FPD/SPD, roll rates, default rates, reserves — and automating the monitoring of model and portfolio drift.
- Building forecasts of key portfolio metrics (volume, risk, profitability): from vintage matrices to statistical models.
- Designing and launching A/B tests: power calculation, stratification, correct estimation of effects on rare events like defaults.
- Performing cohort and vintage analysis, building scoring funnels, segmenting the client base to identify growth areas and risk concentrations.
- Formulating recommendations for product and business based on data and evaluating the actual effects of implemented changes — not only before/after, but also with adjustments for biases.
- Writing complex optimized SQL queries to Vertica and ClickHouse for working with large volumes of data.
- Building multi-level dashboards in Superset for monitoring the state of the portfolio and models.
- Automating metric calculations and validation processes in Airflow.
We expect you to have
- Higher education in technical or mathematical fields.
- At least 2 years of experience in data science, ML, or advanced analytics — experience with models is more important than experience in lending.
- Solid understanding of binary classification metrics (ROC-AUC/Gini, KS, precision/recall, calibration) and which metric is important for which business decision.
- Practical experience in training and/or validating models: gradient boosting, logistic regression; understanding of overfitting, target leakage, time-based validation.
- Strong SQL skills: window functions, query optimization on large datasets.
- Python proficiency at the level of confidently working with pandas, numpy, scikit-learn.
- Statistics: hypothesis testing, confidence intervals, experimental design.
- Ability to translate technical results into business recommendations and defend them persuasively to stakeholders.
This will be a plus
- Experience in credit risk or knowledge of risk metrics (PD/LGD/EAD, roll rates, vintage analysis) — if you don't have it, we'll teach you.
- Familiarity with scoring cards, WoE/IV binning, IFRS 9 / Basel requirements.
- Econometrics and causal inference: statsmodels, diff-in-diff, uplift modeling.
- Experience in building ETL processes in Airflow and dashboards in Superset / Tableau / Power BI.
- Experience launching A/B tests in production.
