Building a Practical Model for "Responsible Generative AI Utilization" and Initiating System Implementation for Automated Underwriting

Nissay Plus Short-Term Mutual Aid Insurance Co., Ltd. (Representative Director and President: Hideyuki Imanishi, hereinafter "Nissay Plus"), part of the Nippon Life Group, which aims to realize "a society where everyone can live with peace of mind, always," and Finatext Ltd. (Representative Director and CEO: Akane Kinoshita, hereinafter "Finatext"), part of the Finatext Group, which provides financial infrastructure for the AI era, are pleased to announce that they have built a practical model to automate underwriting operations while preventing the "black box" nature of generative AI utilization, and have begun system implementation.

This initiative realizes a practical model that addresses the particularly important challenges for financial institutions in generative AI utilization, such as "black box nature" and "ensuring reliability and accountability."

1. Background of the Initiative

In response to the rapid advancement of generative AI technology, financial institutions such as insurance companies are increasingly utilizing generative AI. However, the following challenges have been pointed out regarding its use:

Opacity of the decision-making process (black box nature)

Difficulty in controlling variations in judgments (hallucinations, fairness, bias, etc.)

Ensuring accountability for results

The challenge of financial institutions being unable to "ensure the validity of decision grounds and processes and fulfill accountability," which is essential, becomes apparent as the scope of tasks delegated to generative AI expands simply because it "can make plausible judgments."

As the focus shifts from "experimentally using and evaluating advanced technologies" to "practically applying AI to operations with governance," resolving these issues has become a critical imperative for financial institutions.

Furthermore, financial institutions need to address the following challenges to utilize generative AI in insurance services:

A mechanism to update the AI engine applied to operations without affecting business, ensuring it does not become obsolete.

A mechanism that does not incur significant learning costs or quality assurance costs for AI each time new products are introduced or existing products are revised.

A mechanism to prevent customer information from being unintentionally disclosed to third parties.

Simply building operations that temporarily utilize generative AI is insufficient for long-term insurance services. It is required to continuously keep up with the rapid technological advancements in AI, respond flexibly to the rapidly changing business environment, and responsibly provide these as reliable services demanded by financial institutions.

2. Building a Practical Model for "Responsible Generative AI Utilization"

We have built a "practical model" that resolves these challenges in generative AI utilization, provides responsible insurance services to customers, and is achievable at a cost level that enables efficiency improvements in underwriting operations.

● The role of generative AI is limited to image analysis, and the AI's decision-making process is verifiable.

Assigning broad and complex roles to generative AI leads to problems such as "black box nature of the decision-making process," "uncontrollable judgment fluctuations," and "inability to verify the grounds for judgment." Therefore, the role of generative AI is limited to image analysis (enabling image analysis functions with significantly lower learning costs compared to conventional learning-based AI).

In addition, generative AI should output not only the judgment results but also the grounds for judgment (information extracted from images and the content of the judgment) for verification. (Limiting the role to image analysis ensures the verifiability of the grounds for judgment, avoids the black box nature, and guarantees accountability).

● Underwriting logic is implemented in a coding-based system and controlled through visualization.

By implementing and controlling it through a coding-based system with visualization, judgment fluctuations and the black box nature are eliminated in underwriting logic, which requires strict judgments based on policy terms. Since the logic is visualized, even if unexpected results occur, the cause and problematic parts can be clearly identified, and the scope of modification can be determined. This ensures accountability required by financial institutions.

● Designed with "full automation" in mind, without human intervention.

Since the logic is visualized as code and its quality can be confirmed through testing, the design allows for full automation without assuming continuous human checking of AI judgment results.

Initially, operations will start with approvers confirming the results, and gradually expand to automated underwriting without approval in areas where quality is deemed stable.

● Agile response to new products, product revisions, etc.

The logic is visualized, and additional or modified underwriting logic can be implemented by only handling the differences.

Unlike methods that require retraining generative AI with policy terms, this allows for flexible additions and changes to operations in a short period and at a low cost.

● A reliable generative AI analysis platform that can withstand long-term operation.

A dedicated, closed environment is constructed, and no micro-adjustments through retraining are made, preventing unintentional disclosure of customer information to third parties.

AI models and versions are generally fixed and updated systematically. Judgment results before and after updates are compared and evaluated to verify the impact on operations.

The AI engine and analysis processing are kept loosely coupled, allowing the AI to be updated without affecting the analysis processing.

3. Realizing Seamless AI Underwriting Based on "Inspire"

Nissay Plus utilizes Finatext's insurance business platform "Inspire" as its core system. Inspire is an "AI-Ready" system with a loosely coupled design, a flexible and extensible data model, and an API-first architecture.

In this initiative, an AI underwriting platform was built based on Inspire's design philosophy and added to Inspire's standard functions as an extension. This allows for the rapid launch of the AI underwriting platform and enables flexible and agile additions to underwriting operations and modifications to underwriting logic after service launch.

The Finatext Group has been operating an "underwriting support LLM" that automatically creates underwriting documents using generative AI at Smartplus Short-Term Mutual Aid Insurance Co., Ltd. since 2025. Additionally, Nowcast Inc. has accumulated technology for AI reading and structuring of non-standard documents in areas such as real estate.

Furthermore, they have experience in designing and implementing AI governance policies and human-in-the-loop concepts in these initiatives. As a partner with high affinity for "responsible generative AI utilization," which Nissay Plus considers an important guideline for this initiative, Finatext is responsible for the technical foundation of this project.

4. Approach Towards Full-Scale Operational Application

For the full-scale operational application of the "Responsible Generative AI Utilization" practical model, a phased expansion of application will be implemented.

Prior introduction to the underwriting operations for "Smartphone Insurance," which is highly standardized and has a large number of processing cases.

Starting with a system where "automated underwriting results are approved by humans," it will transition to automated underwriting without human approval once the quality is deemed stable.

Gradual expansion of applicable products and operations.

5. Future Outlook

This initiative aims to build a practical model that financial institutions, which are required to provide reliable services over the long term, can implement and quickly apply to operations, while considering both the societal trend of unavoidable generative AI utilization and the increasing risks of AI utilization alongside rapid technological advancements. This is in the context of moving from the "experimentation and evaluation of advanced technologies" phase to the "practical application to operations with governance" phase.

Through this initiative, Nissay Plus and Finatext aim to establish "practical financial infrastructure that can fulfill accountability" and "best practices for generative AI utilization in insurance operations," and to maximize the potential of generative AI for the benefit of customers and partner companies.

Company Overview

Name

Nissay Plus Short-Term Mutual Aid Insurance Co., Ltd.

Representative

Representative Director and President: Hideyuki Imanishi

Location

Otemachi Building FINOLAB, 1-6-1 Otemachi, Chiyoda-ku, Tokyo

Established

April 30, 2021

Official Website

https://www.nissay-plus.co.jp/

Name

Finatext Ltd.

Representative

Representative Director and CEO: Akane Kinoshita

Location

Sumitomo Fudosan Kudan Building 9F, 1-8-10 Kudan Kita, Chiyoda-ku, Tokyo

Established

December 2018

Official Website

https://finatext.com/fn