Context
Manual credit-limit decisions can be inconsistent. This historical Blue Star analysis explored whether a quantitative framework could support more standardised and updateable decisions.
My role
Identified process inconsistencies and formulated a data-led credit-policy approach using financial and vendor information.
Approach
The work covered exploratory analysis, variable selection, a 75/25 training/test split, and comparison of linear regression models using significance, adjusted R², and mean absolute percentage error (MAPE). The selected model used customer type and total billing.
Results
The historical selected model reported adjusted R² of 81.9% and MAPE of 39.2%. A billing-only alternative reported adjusted R² of 81.6% and MAPE of 40.3%. Adjusted R² describes explained variance under the fitted model; it is not prediction accuracy.
Limitations
The material supports a proposed decision framework, not a claim of realised financial impact. MAPE is a separate error measure, and these historical results do not guarantee future performance. Source data, coefficients, customer identities, and credit limits remain private.