How AI is transforming financial services & banking

As organizations look to build and scale their artificial intelligence (AI) efforts in a cloud-enabled enterprise, it’s more important than ever to take an integrated approach that mobilizes cross-functional partners, uses cutting-edge tools and stays at the forefront of the latest science and applied research in finance and AI. 

At AWS re:Invent 2024, I presented on successfully building programs for AI at scale, including specific insights from our own evolution here at Capital One around enterprise AI capabilities and applications.

AI use cases in banking and financial services

To set the stage, it’s helpful to understand how vast the different use cases are when it comes to banks using AI. Specific areas where AI systems are improving Capital One’s operations include:

• Anti-money laundering
• Cybersecurity
• Digital and call center servicing
• Fraud detection
• Multichannel marketing
• Product valuations

AI has proven to be a key tool in our ability to deliver next-level experiences for our customers, and we’ve seen great success in enabling real-time modeling for the above use cases in financial services.

The power of cross-functional partnerships in AI for banking and finance

Enterprise-level strategic investments play a key role in fostering an innovative culture at Capital One. These investments help us establish models that use high-quality data to deliver timely, relevant and accurate insights. One of our biggest investments has been in cross functional partnerships.

Aligning internal teams to accelerate our AI research agenda

When we think about scaling AI technology in a responsible and well-managed way, having diverse perspectives that can help anticipate and catch problems that may otherwise be overlooked are critical. With this in mind, we prioritize research teams that include a mix of:

• Data scientists 
• Machine learning engineers
• Software and data engineers
• Applied researchers
• Product managers

We also think about how different teams work together with our applied AI research teams – specifically, the various lines of business and risk, legal and regulatory compliance teams. When working with the lines of business, research teams can gain insight into the problems relevant to each individual business, the hurdles that they might not have been aware of and how they are connecting data to meet their goals. On the other hand, risk, legal and regulatory compliance teams help us better understand the best practices we need to follow to ensure that the research we are working on aligns with company priorities. 

When it comes to AI technology, we involve our partner teams early and often. We keep the lines of communication open with these partners, listen to them and consider every avenue to leverage their expertise to optimize our AI innovations. We also conduct AI research and design with deployment top of mind. We prioritize making our AI systems and tools easy to use, and account for the customer experience at each stage of the product development process. 

Collaborating with industry partners to advance AI

The diversity of thought we get through our partnerships with business, risk and other research teams is a core tenant of how we scale our AI model, but that diversity of thought doesn’t stop at the walls of Capital One.

We partner with some of the country's top research universities and institutions to tackle AI challenges in banking and finance. For example, we’ve established centers for responsible AI research in finance in conjunction with Columbia University and the University of Southern California (USC).

Using AI banking to enhance the customer and associate experience

As our research needs scaled over the years, so did the innovative tech stack supporting our work. Our current stack is the product of continued investment that enables us to continually enhance the customer experience and how our associates work.

Understanding and meeting customer needs in real time

We’ve thought a lot about being personalized and proactive in our communications with customers and how to get a holistic understanding of customers before and during the engagement. To do this, we’ve investigated the use of machine learning (ML) for temporal data in finance. What we found is that AI and ML can be used to study temporal patterns across various financial data streams. From there, we can identify meaningful patterns and bring form and structure to the data within them, helping us to find unique ways to assist our customers through AI applications.   

Connecting research to customer-facing products

The next step is how we actually take this research to deploy products that result in a positive customer experience in real time. To achieve this, we’re evaluating the best way to use low latency streaming data, reliable hosting of LLMs and fault-tolerant systems in our products.

Using generative AI to enhance efficiency

We’re also thinking about how to improve the efficiency of our associates. Our proprietary generative AI agent servicing tool is helping our agents get access to information to resolve customer questions more quickly and efficiently.

For example, if a banking customer calls to ask about whether a declined card transaction will count against their daily card limit, our agents can use this AI tool to quickly search for the relevant information in real-time, enabling them to deliver reliable information more quickly than ever before.

We’re also leveraging generative AI tools to help our software developers supercharge their coding experience and enhance efficiency—letting them focus on more creative problem solving while staying well managed.

Key takeaways from scaling AI in financial services

Beyond pure R&D work, we’ve learned from the practical application of LLMs that data, evaluation, tooling and modeling have been critical to our continued success in AI:

• Data: A use case will make or break depending on the data. To meet our business goals, we found that we need clean and curated data. We’ve invested in tools and talent to help us get just that. Effective data management plays a critical role in maximizing AI potential.
• Evaluation: Evaluation is key to staying well-managed to accurate information retrieval, and to identifying potential failure modes. We’ve set up rigour testing and guardrails for our models to evaluate them adequately.
• Tooling: State-of-the-art inference optimization tools and reliable access to GPUs matter a lot, so we’ve invested in engineering expertise and continued R&D in high performance computing.
• Modeling: Rigorous testing and continuous R&D has helped us fine-tune our models, which can help us uncover performance deterioration in unexpected areas.