Promoting DX to utilize customer feedback in business operations through generative AI

Tokio Marine & Nichido Anshin Life Insurance (hereinafter "Anshin Life") has been working to utilize customer feedback to improve operations and service quality to realize fully customer-oriented business operations.

The company receives up to 18,000 customer feedback entries annually from customer contact points, including telephone calls, online and surveys. Until recently, the reading and classification of this unstructured data was done visually, one by one. In addition to taking an enormous amount of time, this process was also subject to human judgment that led to inconsistencies in classification, which was a barrier to promptly utilizing and reflecting customer feedback in business operations.

In the past, classification using non-generative AI was considered, but with conventional machine learning it was difficult to maintain expected accuracy due to considerable effort required to prepare large amounts of training data and retrain.

In response to this issue, Anshin Life co-created with IBM Client Engineering and began developing a classification and analysis model using IBM® watsonx.ai. The project progressed through collaboration between the business and IT departments, integrating their respective expertise. The business department was responsible for defining requirements and user tests, while the IT department was in charge of project management and architecture design. IBM presented implementation proposals and provided knowledge on generative AI. By proceeding with agile development on a weekly cycle, respective initiatives were realized with a sense of speed.

In the first phase focused on “business efficiency,” customer feedback from phone calls and other sources was captured by generative AI and classified into approximately 40 industry-wide codes. In the second stage, for the purpose of “advanced analysis,” important keywords set by the company were extracted as hashtags and cluster analysis was performed, making it possible to analyze text data, which was previously challenging.

In terms of classification and analysis accuracy, the combination of watsonx.ai’s natural language processing technology and Retrieval Augmented Generation (RAG) technology achieved both high accuracy and flexibility. Since watsonx.ai was running on the cloud, the company was also able to handle personal information with peace of mind.

The project’s smooth progress was supported by the “application owner system,” a framework under which the business and IT departments worked together to identify necessary functions and allocate appropriate resources by proceeding from PoC to full-scale development from a small start. In addition, by taking on the challenge of “trying it out first” with a target of 70-80% accuracy, rather than seeking perfection from the start, overall efforts were revitalized, helping the project to move forward with speed and agility.

Koji Imakiire, manager in charge of the project belonging to the Digital Support Group, Anshin Life IT Planning Department, speaks about how he worked to move it forward smoothly. “Controlling expectations for generative AI is key, to avoid falling into underestimation, or conversely overconfidence that it’s a ‘universal tool that can do anything’. To this end, it was important for everyone in the user department to engage the project with a sense of satisfaction. By performing user verification in each instance with products that actually work in weekly agile development, we made consistent improvements.”

The introduction of generative AI has resulted in improvements in four areas: operational efficiency, classification standardization, data utilization and maintenance.

• Operational efficiency
  In a demonstration test, the classification and approval process, which used to take 7-8 minutes manually, was reduced to 2-3 minutes, achieving an efficiency improvement of about 75%. The time saved was equivalent to 500 hours (approximately 20 business days) per year.
• Classification standardization
  AI presented the basis for judgment, making it readily visible. Variation in classification judgments among personnel, which had been an issue, was eliminated.
• Data utilization
  Cluster analysis through keyword extraction and visualization of related keywords enabled useful suggestions to be obtained from unstructured data. Trends in customer feedback that could not be seen through conventional analysis were clarified, leading to improved accuracy in formulating hypotheses and considering policies.
• Maintenance
  System prompts and RAG eliminated the need for fine tuning when launching new products and updating information, leading to a reduction in maintenance efforts.

Imakiire adds, "This initiative increased momentum for the use of generative AI within Anshin Life." The company actively reviewed other processes after the introduction of generative AI, moving to further improve operational efficiency. In fact, more than 200 types of code were used in the PoC during the analysis stage, but after actual operation, they considered streamlined operations without them. "Before using generative AI, classification alone took a lot of time, so we didn't have time to think about efficiency. Today we can afford to review other existing processes," Imakiire concludes.

Koji Yamamoto, Director of Anshin Life, also comments, "We now see trends in customer feedback that went unnoticed before. We plan to formulate hypotheses regarding them in the future, allowing us to take even more effective measures."

Moving forward, Anshin Life plans to expand this system to other departments and continue to challenge itself by testing the potential of generative AI in new areas.

Anshin Life is building a system that can quickly read diversifying customer needs through the use of generative AI and reflect them flexibly in business strategies. By transferring these results to the production environment, the company aims to create new value in the insurance industry by strengthening "customer-oriented business operations." They will also expand customer support, products and services through this demonstration of the value of human work facilitated by transitioning AI to the production environment.

Tokio Marine & Nichido Anshin Life Insurancesupports peace of mind through life insurance as a member of the Tokio Marine Group. With customer trust as the starting point of all business activities, they provide peace of mind through the life insurance business, contributing to the development of an affluent, comfortable society and robust economy.