Mitsubishi Research Institute Begins Practical Provision of "AI Scoring Model" to Mebuki Financial Group

Mitsubishi Research Institute, Inc. (President and CEO: Kenji Yabuta, hereinafter MRI) began providing its "AI Scoring Model" for practical use to Mebuki Financial Group, Inc. (President and CEO: Tetsuya Akino, hereinafter Mebuki Financial Group) on April 1st. This initiative supports the digital transformation (DX) of underwriting operations.

**1. Background**
In mortgage underwriting, scoring models play a crucial role in assessing an applicant's credit risk based on application and credit information, which is then used for screening decisions and setting interest rates and guarantee fees. In recent years, against the backdrop of a rapidly changing economic climate, including rising interest and inflation rates, as well as the accumulation of data and advancements in AI technology, there has been a growing need for new models that are more accurate and flexible than conventional ones.
For over 20 years, MRI has supported underwriting decisions and risk management in the retail loan sector for regional financial institutions by building and monitoring scoring models post-implementation. Additionally, through its "Underwriting AI Service," which automates loan approval decisions, MRI has been working on the automation and efficiency of underwriting operations using AI. Building on this track record, MRI has now developed an advanced "AI Scoring Model" for mortgages for Mebuki Financial Group, which began practical use in April. Mebuki Financial Group is the first to implement MRI's "AI Scoring Model."

**2. Overview of the AI Scoring Model**
**(1) Features**
**① Improved Accuracy in Predicting Default Probability**
MRI's "AI Scoring Model" is an AI model that quantifies credit risk by predicting the probability of default based on mortgage applicants' application and credit information. Compared to conventional models that rely on about 5 to 10 items, this model can capture complex relationships from a more diverse range of items, thereby improving the accuracy of default probability prediction. By precisely assessing credit risk, it becomes possible to make objective evaluations in cases where lending decisions were difficult under previous standards.

**② Design Tailored to Each Financial Institution's Characteristics**
A key feature of the AI scoring model provided to Mebuki Financial Group is that it was built and adjusted based on the group's own lending performance data, taking into account the specific realities of their mortgage underwriting and portfolio characteristics. Furthermore, leveraging MRI's long-standing expertise in supporting the retail loan sector of regional financial institutions, the AI is designed not only for predictive accuracy but also with consideration for the ease of confirming the basis of judgment and for continuous operation.

**③ Ease of Implementation and Operation via API Integration**
The evaluation results from this model are provided through API integration (*2) with the financial institution's personal loan business support system (*1). Since the model's judgment results can be linked between systems, it can be applied to practical operations without requiring large-scale modifications to the personal loan business support system. This allows for implementation within a realistic timeframe and cost, and enables regular and rapid verification and updating of the model in response to changes in the external environment or portfolio.

**(2) Relationship with the Underwriting AI Service**
This initiative expands the lineup of the "Underwriting AI Service" already in practical use at Mebuki Financial Group. With a single API call from the personal loan business support system, it can link the credit risk quantification results from the "AI Scoring Model" and the comprehensive approval/rejection judgment from the "Underwriting AI Model." This not only speeds up underwriting responses and reduces the man-hours required for screening decisions but also supports the sophistication of credit risk assessment and pricing (setting of interest rates and guarantee fees) in an integrated manner. Mebuki Financial Group achieved a more efficient implementation by leveraging the connection infrastructure established during their initial Underwriting AI Service deployment, but the "AI Scoring Model" can also be introduced and used as a standalone product.

**API Integration Image of Underwriting AI Model and AI Scoring Model**
(Image caption: Created by Mitsubishi Research Institute)

**3. Future Plans**
Going forward, MRI will continuously verify the operational status and accuracy of this scoring model, aiming to further enhance credit risk management in retail loan underwriting and contribute to the further promotion of DX for regional financial institutions.