Data Science, Credit Risk and Decision Intelligence

Business problem

A card portfolio needed earlier visibility of accounts likely to migrate into non-performing status, so collections and risk teams could act before delinquency hardened.

Challenge

Existing indicators reacted only after accounts were already significantly overdue, giving limited time for intervention.

Analytical approach

Built behavioural and vintage-based early-warning indicators, combining delinquency migration patterns with account-level risk signals to flag deteriorating accounts earlier in the cycle.

Decision supported

Prioritisation of collections outreach and adjustment of risk-based treatment strategies for at-risk segments.

Outcome

Earlier identification of at-risk accounts, enabling proactive rather than reactive collections treatment.

Tools

Python, Oracle SQL

Certain details and organisational identifiers have been generalised to protect confidentiality.

Business problem

Credit risk signals were often interpreted in isolation, without a consistent framework connecting portfolio structure, behaviour and macro context.

Challenge

Fragmented analysis made it difficult to distinguish structural portfolio shifts from short-term noise.

Analytical approach

Developed a structuralist framework linking portfolio segmentation, delinquency dynamics and macroeconomic overlays into a single analytical lens for interpreting risk movements.

Decision supported

More consistent interpretation of portfolio risk trends in management and risk committee discussions.

Outcome

A repeatable analytical lens for distinguishing structural risk shifts from temporary fluctuations.

Tools

Python, Oracle SQL, Power BI

Certain details and organisational identifiers have been generalised to protect confidentiality.

Business problem

Risk and management reporting relied on manual, dispersed data pulls, slowing down monitoring cycles.

Challenge

Inconsistent data definitions across sources made portfolio metrics difficult to reconcile and trust.

Analytical approach

Designed a structured credit risk data mart with standardised definitions, feeding automated dashboards for portfolio and delinquency monitoring.

Decision supported

Faster, more reliable portfolio monitoring for risk management and reporting cycles.

Outcome

Reduced manual reporting effort and a single consistent source of portfolio risk metrics.

Tools

Oracle SQL, Python, Power BI

Certain details and organisational identifiers have been generalised to protect confidentiality.

Business problem

Management needed forward-looking visibility of non-performing loan flow under different macroeconomic conditions.

Challenge

Portfolio-level NPL flow is influenced by macro conditions that are difficult to translate into a usable forecast.

Analytical approach

Applied time-series and econometric modelling (including SARIMAX-style approaches) to link NPL flow with macroeconomic indicators and generate scenario-based forecasts.

Decision supported

Scenario planning and portfolio risk provisioning discussions.

Outcome

A forward-looking view of NPL trajectory under alternative macroeconomic scenarios.

Tools

Python (statsmodels), Oracle SQL

Certain details and organisational identifiers have been generalised to protect confidentiality.

Business problem

Risk teams needed a consistent, visual way to track how loan vintages performed and migrated over time.

Challenge

Vintage and roll-rate analysis was previously ad hoc, rebuilt manually for each reporting cycle.

Analytical approach

Built a reusable vintage and roll-rate dashboard, standardising cohort definitions and migration-state transitions across products.

Decision supported

Ongoing portfolio quality monitoring and early identification of vintage-level deterioration.

Outcome

A recurring, self-service view of vintage performance for risk and management stakeholders.

Tools

Power BI, Oracle SQL, Python

Certain details and organisational identifiers have been generalised to protect confidentiality.