Predictive modelling

Explainable Credit Risk Decisions

Name: Credit Scoring
Brand: MII Microsoft

In trial

Interactive credit-risk dashboard with in-browser ML scoring and explainability for every decision.

Business value

Faster, more consistent credit decisions where every score can be traced back to the factors behind it, with applicant data staying within organizational boundaries.

Loan officers often assess applications with spreadsheets and scorecards that have not been updated in years, leaving decisions inconsistent and difficult to defend. Credit Scoring is an MII-built capability pattern that brings credit risk models directly into the lending workflow, so loan officers enter an applicant profile and get an instant creditworthiness score and default-probability estimate, each broken down into the factors that drove it. Gradient-boosted models are exported to ONNX and run in the browser so scoring is immediate and no applicant data leaves the session, and a model-performance view exposes accuracy, the confusion matrix, and feature importance so risk teams can review the model rather than take it on trust.

Where it helps

Scoring applicant creditworthiness instantly with per-factor explanations
Comparing model accuracy and behavior before trusting it for decisions
Keeping applicant data in-session instead of transmitting it externally

What the pattern does

In-browser ML model execution with instant scoring
Per-factor breakdown explaining what drove each decision
Model performance metrics including accuracy, confusion matrix, and feature importance

The challenge

Credit assessment in smaller lending institutions leans on manual review and static scorecards. They are slow to update, hard to explain to an applicant or a regulator, and give risk teams little visibility into why a decision was made.

What was built

Gradient-boosted models were trained on an applicant-profile feature set, then exported to ONNX so they run inside the browser alongside a transparent scoring fallback. The interface pairs every score with the factors that moved it, and a separate view reports model accuracy and feature importance.

What changed

A working dashboard covering the lending decision path end to end: portfolio overview, applicant scoring, default prediction, and model performance, with Microsoft Entra sign-in wired for enterprise rollout.

Why it matters

Shows how credit risk models can sit inside the tools lending teams already use, with every score traceable to the factors behind it, giving risk and compliance a basis to review the model rather than take it on trust.

Credit score range: 300–850
In-browser inference: ONNX
Per-factor score breakdown: Explainable

Back to MII-built products