EXCLUSIVE OFFER: Save 12% on all AI Certifications. Exclusive for Job Seekers. Offer Ends on May 31, 2024!
Use Voucher Code:  UNEM12AIC .
×
Generative AI in Banking: State of the Union/ai-insights/generative-ai-in-banking-state-of-the-union

Generative AI in Banking: State of the Union

March 14, 2024

Generative AI in Banking: State of the Union

2023 was the watershed year for Generative AI. Having debuted in early 2023, it is already seeing explosive growth in adoption across industries, including banking. 80% of the enterprises will have used Generative AI by 2026 (Gartner). For several decades now, banks have been at the forefront of technology adoption. Adoption of AI has moved beyond Proof of Concepts to the mainstream, cloud adoption being a significant enabler.

As per information available in the public domain, Santander Bank has identified over 60 use cases for Artificial Intelligence. JP Morgan Chase is working across 300 AI use cases. ING Bank, Deutsche Bank, Ally Financial, Citizens Bank, PNC Bank, Valley Bank, and several others have reported that they are actively exploring diversified AI use cases. A significant portion of these use cases involve Generative AI. Two-thirds of banks expect their IT spending to increase in 2024 (Cornerstone). The McKinsey Global Institute estimates that among industries globally, Gen AI could add the equivalent of $2.6 trillion to $4.4 trillion annually in value, with an annual potential of $200 billion to $340 billion in banking across all segments and functions.

Gen AI Use Cases in Banking

Gen AI enables the creation of new content based on existing data. It uses models that can learn existing patterns and structure of the data and generate new data with similar characteristics. It has many applications in banking.

Some of the typical use cases of generative AI in banking are: 

  • Content creation for personalized marketing: Gen AI can expedite content creation by automatically generating blog posts, social media updates, marketing emails, and other customer-facing content. 
  • Risk assessment: It can help banks assess the risk of lending, investing, or trading by generating realistic scenarios and outcomes based on historical data and current market conditions. 
  • Customer service operations: Gen AI can help banks improve customer service by generating natural language responses and dialogues for chatbots and voice assistants that can handle common queries and requests. 
  • AI use cases: It helps get adequate visibility to legacy applications and dependencies, understand legacy code, extract business rules, streamline documentation of the code, and even generate code.
  • Test data generation: Banks need quality test data that covers a myriad of test conditions to test their systems before pushing them into production. They typically use data from the Production environment. By regulations, they must protect the personally identifiable information (PII) of their customers. Hence, they apply various scrambling techniques to scramble the PII data. Gen AI helps generate test data in volumes, instead of depending on production data.

Implementation challenges

Being part of a regulated industry, banks need to consider several factors in their Gen AI adoption journey. Some of the key factors are:

  • Data quality and availability: Gen AI requires large amounts of high-quality and relevant data to train the models and generate accurate and diverse outputs. Banks may face challenges in accessing, collecting, and processing the data, especially if it is sensitive, or subject to regulations. Additionally, banks need to develop capabilities to use unstructured data.
  • Ethical and legal issues: Gen AI poses ethical and legal issues, such as privacy, security, accountability, and transparency, that banks need to address before deploying the technology. For example, banks need to ensure that the generated content is not misleading, harmful, or biased; and that the customers are informed and consent to the use of generative AI. A robust risk management framework for governing the models along with new controls are prerequisites for Gen AI adoption.
  • Human oversight and validation: Gen AI is not perfect and can produce errors, inconsistencies, or undesirable outputs that may affect the quality and reliability of the service. Banks need to have human oversight and validation mechanisms to monitor and correct the outputs of generative AI, as well as to provide feedback and guidance to the system. 
  • Need for training: McKinsey & Company have pointed out that the adoption of gen AI requires the management teams to learn new terminology such as reinforcement learning and convolutional neural networks. In addition, they need to understand the implications of the potential pathways gen AI could create.
  • Need to adopt new Operating models: In addition to the need for coordination between business and technology, successful adoption of gen AI requires close coordination with staff with expertise in Data Engineering and Data Science. While a centralized model might be more appropriate in the initial stages of Gen AI adoption, a federated model might be more beneficial as the technology advances.
  • Culture: The disciplines of AI and Gen AI are evolving very fast. Slow-moving banks will find it a challenge to align with new operating models with the necessary speed.
  • Talent: Scaling up Gen AI will require a talent pool skilled in areas such as quantitative analysis, modeling, prompt engineering, data engineering, and data science. The talent market is fiercely competitive, and recruitment may not keep pace with demand.
  • Change management: Successful deployment of Gen AI will typically involve significant changes to existing processes, mindset, priorities, and culture. Banks will need to plan for change management as they consider Gen AI.

Technology considerations

Most banks still run on legacy core banking systems. These systems have evolved over a long time. They are complex and are tightly coupled with other systems. Changes to these systems are time-consuming. Gen AI applications are built on top of new technology stacks. Banks need to have a view of the overall architecture to support Gen AI, that takes into consideration the integration and co-existence between the new technology stacks and the existing legacy stack. Most banks are actively pursuing migration to the Cloud, often with multiple cloud providers. This adds another level of architectural complexity. Critical choices will need to be made between Build and Buy options for the Gen AI models and applications. Third-party risks must be carefully considered if the Bank decides to go the Buy route.

AI Regulations

In October 2023, the White House issued an executive order emphasizing content labeling and transparency of the AI models. The order requires the Department of Commerce to develop guidance for labeling AI-generated content. AI companies will use this guidance to develop labeling and watermarking tools that the White House hopes federal agencies will adopt. The executive order also calls for federal agencies to develop rules and guidelines for different applications, such as supporting workers’ rights, protecting consumers, ensuring fair competition, and administering government services. The US Government has launched the AI Safety Institute to develop standards for the safety, security, and testing of AI models, develop standards for authenticating AI-generated content, and provide testing environments for researchers to evaluate emerging AI risks and address known impacts. The use of AI will be regulated by the AI Act. Banking being a highly regulated sector, adopters of Gen AI need to look out for evolving standards, rules, and regulations.

The Journey

Banks should start with a clear business objective and value proposition. They should identify the specific use cases that Gen AI can address and the expected benefits and outcomes that it can deliver. They should define the key performance indicators and success metrics to measure the impact of Gen AI. They should start with small-scale and low-risk pilot projects to test and validate the feasibility and effectiveness of generative AI. Thereafter, they should continuously improve and refine the system based on the feedback from the users and stakeholders. 

Banks should have a holistic framework that includes architecture, operating model, oversight, risk management and controls, talent, and change management. They should leverage existing platforms and tools that offer generative AI capabilities, such as Microsoft Azure OpenAI Service, ChatGPT, AWS BedRock, AWS Sagemaker, or Google Gemini for the development and deployment of generative AI solutions, as well as provide enterprise-grade security and compliance features. 

They should aspire to build a culture of innovation and collaboration that encourages and supports the adoption of Gen AI. They should also invest in upskilling and reskilling their workforce to enable them to use and benefit from Gen AI, as well as to collaborate with the system. 

Conclusion

Banks are at the nascent stages of AI adoption. Some have identified use cases and initiated pilot projects; others are still exploring and have not yet begun. AI and Gen AI should not be viewed merely as new technologies. Banks must plan for the right operating model, hire the right talent, and pay attention to new risks introduced by these technologies. They also need to get more agile. The technologies are evolving fast, and the legacy culture of most banks will not be able to cope with the required agility. When used properly for the right use case, they can improve efficiencies significantly, help mitigate risks, and create personalized service offerings. We can expect to see some successful Pilots in 2024 and 2025, and acceleration of adoption in 2026 and beyond. It is also expected that the Fintech space will get very active and highly competitive, addressing specific use cases leveraging AI and Gen AI.