Advancing from AI-Ready to AI-Powered: Meeting the Challenge of Transforming Business Processes by Training AI

In fiscal 2024, the IT Promotion Division and the dX Strategy Division prepared an environment for safely using generative AI. Building on that foundation, we encouraged company-wide adoption under the motto “try it first,” and average individual usage reached around 60%. (For more on this, please see the previous article.) At the same time, precisely because use expanded, the next challenge became clear: How can we embed AI into actual operations? The challenge for fiscal 2025 is moving from “try it first” to “apply it in day-to-day work.”

The Digital Promotion Group was established in April 2025 as a specialist organization to address these issues and to be more strategic in promoting the use of data and AI. The group has two major roles. The first is to strengthen competitiveness and enhance corporate value through advanced technologies such as AI. The second is to improve administrative operations through digitalization. We also collaborate with business units such as the Business Process Planning Division to develop specialists and raise AI literacy among employees, aiming to serve as a core driver of Company-wide digital transformation.

As part of our efforts to expand the use of generative AI, we conducted interviews with the relevant divisions to identify operational issues. These discussions revealed that many departments were spending a lot of time handling inquiries. In particular, pre-transaction verification handled by the Business Process Planning Division involves complex rules and a high volume of transactions, requiring significant workloads. So we launched a joint project with that division, as the issues were clear and the potential benefits of automation were expected to have the greatest impact there.

Tasks such as pre-transaction verification must be performed using closed internal information. Unlike Gemini, which draws from a wide range of online sources, NotebookLM generates responses solely from the data we provide. We determined that this characteristic made it the best fit for this work. We also needed to reexamine the source materials the AI references in order to further improve the accuracy of its responses.

During development, we presented a wide variety of questions to the AI. When the results fell short of expectations, we examined what additional information was needed to guide it toward correct responses and incrementally refined the inputs, case by case, to make steady progress.

Release is just the starting line. Training AI is like raising a new employee. No matter how capable someone may be, no one is perfect from day one. New hires do not initially understand company-specific rules or culture; they learn and grow through ongoing conversations with senior colleagues and managers. AI is no different. As it gains “experience” through data and repeated refinement of its output, it evolves into a form that best serves our needs.

Using generative AI as a sounding board during the planning phase can also enhance the quality and accuracy of outputs. In some cases, this approach has already led to new insights and ideas for new business.

According to the five-level framework published by Keidanren, our company is now at Level 2: the initial stage of being AI-ready. This is a small-start phase in which certain divisions have begun to use AI for some tasks. During this fiscal year, we aim to link AI with core systems and multiple data sources that are currently dispersed. For example, when receiving a business inquiry or investment consultation, AI could perform credit assessments based on corporate information and past records, reducing the need for manual review. This would address a strong internal need, and we want to implement it as soon as possible.

In April 2025, we began sharing AI-related updates as part of internal training, aiming to enhance employees’ AI literacy. Their interest has been strong, and going forward it will be important to deepen understanding of the practical and knowledge-based aspects of AI, including the fact that it is not all-powerful.

One key skill in the coming AI era will be the ability to articulate thoughts precisely. AI will be used across all types of work to generate and refine ideas, making effective interaction with AI essential. The more clearly problems and concepts can be expressed in words, the higher the quality of the AI’s output. Just as with prompts in Gemini, the ability to clearly convey intent will increasingly determine outcomes.

The Pre-transaction Verification AI handles inquiries by accumulating the information needed to generate the most appropriate responses. In other words, the same approach can be applied to other business tasks by changing the data used to train the AI. We have already begun collaborating with divisions beyond the Business Process Planning Division and are discussing plans to launch a new internal service by the end of the year. Through these efforts, we also hope to strengthen connections across divisions and among employees.

The use of AI as a business support tool, particularly through AI agents, is advancing rapidly. We believe AI can transform our working environment as it processes increasing volumes of internal and external data and acquires autonomous, multifaceted capabilities, such as computer vision and task execution. This will free employees from repetitive routine tasks, allowing them to concentrate on work that requires creativity and advanced judgment. The IT Promotion Division aims to lead this transformation by formulating and executing the next strategic steps to help guide the industry forward.