A featured contribution from Leadership Perspectives, a curated forum for enterprise technology leaders, nominated by our subscribers and vetted by the CIOApplications Editorial Board.
LCL
From Theory to Practice: Generative AI in Banking
At our company, the integration of generative AI began with a dual awareness. On the technical side, at the AI Factory level, these models and their potential were studied as soon as they were released. On the business side, there was a clear mandate from the highest levels to adopt them quickly, starting with the identification of the first suitable use case. At the product level, this has taken shape through joint efforts involving the Business Lines, AI delivery teams and IT, following a fairly traditional model.
When it comes to integration, everything revolves around computing. The core challenge for organizations is to ensure the security of data and information shared with increasingly essential service providers.
This requires airtight processes, encrypted data in transit and strong contractual guarantees from providers. Implementing generative AI is fundamentally an engineering issue. If the goal is to scale it by issuing prompts, the key question becomes: What is the bandwidth? For users, it also raises another concern: What is the response time and is it compatible with realworld applications?
User training is just as critical. While technology can be deployed once the right skills are assembled, successful adoption depends on more than just tools. Technology should serve users, not dictate how they work. This underscores the primacy of business needs, which must be fully supported. Moreover, AI use must comply with the applicable regulatory and legislative frameworks. That is especially important in the banking sector.
Among the use cases scaled to production, the most emblematic is the writing assistant deployed over a year ago. Designed for all 12,000 advisors in the commercial network, it marked the first application of generative AI to its full potential. This became the cornerstone of what we, among others, call the “augmented advisor.” In its first version, the AI generates a draft response to incoming client emails in the secure messaging system. The advisor provides a natural language instruction (the prompt) and receives a proposed reply, which can be validated or modified. Millions of emails are now handled this way.
This tool was just the first building block. A speechto-text feature was soon added. This allows advisors to dictate instructions, improving convenience and shaping prompts in a way that enhances the AI’s output. The next, more ambitious step is integrating access to the document base to retrieve precise product information and other key data.
A review of this work would be incomplete without examining the types of value generated. Two categories stand out:
1. Tangible gains such as increased revenue or cost savings through reduced personnel or operational expenses.
2. Strategic or qualitative gains, including differentiation, regulatory compliance and improved operational efficiency, though these are harder to quantify and may not always be replicable.
Managing use cases by value means being able to assign measurable outcomes to each initiative. This is necessary, although not sufficient. Since these projects require investment and are expected to yield returns, measuring performance is critical. It is not just about credibility; it is about responsible business management.
If a generative AI tool is expected to free up resources, then once implemented, those resources should be redeployed to new initiatives. If not, the investment should be reconsidered. If a tool promised to save 15 minutes a day, that is a measurable outcome and it should be tracked.
We already have a wide range of relevant metrics: efficiency, time saved, adoption rate, employee satisfaction and customer satisfaction. Even when gains do not translate directly into revenue, they often appear in how human capital is redirected toward growth-oriented initiatives. That said, some projects are valuable even when their benefits are less tangible, such as driving acculturation or spreading new technologies.
After all, money is not the measure of all things. And yes, there is a certain irony in hearing that from someone who has spent years in the banking sector.
Among the use cases scaled to production, the most emblematic is the writing assistant deployed over a year ago. Designed for all 12,000 advisors in the commercial network, it marked the first application of generative AI to its full potential. This became the cornerstone of what we, among others, call the “augmented advisor.” In its first version, the AI generates a draft response to incoming client emails in the secure messaging system. The advisor provides a natural language instruction (the prompt) and receives a proposed reply, which can be validated or modified. Millions of emails are now handled this way.
This tool was just the first building block. A speechto-text feature was soon added. This allows advisors to dictate instructions, improving convenience and shaping prompts in a way that enhances the AI’s output. The next, more ambitious step is integrating access to the document base to retrieve precise product information and other key data.
A review of this work would be incomplete without examining the types of value generated. Two categories stand out:
1. Tangible gains such as increased revenue or cost savings through reduced personnel or operational expenses.
2. Strategic or qualitative gains, including differentiation, regulatory compliance and improved operational efficiency, though these are harder to quantify and may not always be replicable.
Managing use cases by value means being able to assign measurable outcomes to each initiative. This is necessary, although not sufficient. Since these projects require investment and are expected to yield returns, measuring performance is critical. It is not just about credibility; it is about responsible business management.
If a generative AI tool is expected to free up resources, then once implemented, those resources should be redeployed to new initiatives. If not, the investment should be reconsidered. If a tool promised to save 15 minutes a day, that is a measurable outcome and it should be tracked.
We already have a wide range of relevant metrics: efficiency, time saved, adoption rate, employee satisfaction and customer satisfaction. Even when gains do not translate directly into revenue, they often appear in how human capital is redirected toward growth-oriented initiatives. That said, some projects are valuable even when their benefits are less tangible, such as driving acculturation or spreading new technologies.
After all, money is not the measure of all things. And yes, there is a certain irony in hearing that from someone who has spent years in the banking sector.
The articles from these contributors are based on their personal expertise and viewpoints, and do not necessarily reflect the opinions of their employers or affiliated organizations.
