Scaling Generative AI in the Banking Sector: Selecting the Optimal Operating Framework

The advent of generative AI is revolutionizing the banking industry, presenting avenues for enhanced efficiency and innovation. As financial institutions hasten to integrate generative AI, adopting the appropriate operating framework is crucial for leveraging its full capabilities.

Generative AI (Gen AI) is transforming the banking sector as institutions employ this technology to enhance customer service chatbots, combat fraud, and streamline labor-intensive activities like code development, drafting pitch books, and condensing regulatory reports.

The Kamiweb Project projects that Gen AI could contribute an annual value of $200 billion to $340 billion to the worldwide banking industry, accounting for 2.8 to 4.7 percent of total sector revenue, primarily through productivity gains. Nonetheless, as banks and financial entities swiftly move to adopt Gen AI, they encounter various challenges. Properly leveraging Gen AI can unlock significant value, whereas mishandling it may introduce complex issues. Industries face multiple risks with Gen AI, such as producing inaccurate or nonsensical information, violating intellectual property rights, lacking transparency in system operations, grappling with bias and fairness concerns, and encountering security vulnerabilities.

In a prior piece, we discussed strategies for banks to fully harness the potential of Gen AI. To maintain value over the long term, beyond preliminary experiments, it's essential to have strong capabilities in seven key areas:

• Strategic planning
• Talent acquisition and development
• Operational framework
• Technological infrastructure
• Data management
• Risk management and control measures
• Adoption and change leadership

These areas are interdependent and necessitate cohesive alignment throughout the organization. For example, an effective operational framework is futile without the requisite talent or data infrastructure.

This piece focuses on one of these critical areas: the operational model, essentially the strategic execution blueprint for businesses. Future articles will explore the remaining dimensions. We define what an operational model is, its significance, and then discuss the various operational model archetypes that have evolved for Gen AI in banking, highlighting the most successful archetype. We also cover crucial decisions that financial institutions must make when establishing a Gen AI operational model.

Our findings indicate that a high level of centralization is optimal for Gen AI operational models across different industries. Without centralized management, pilot projects risk becoming isolated, hindering scalability. Specifically, in the financial services sector, institutions adopting a centrally managed Gen AI operational model have seen the most substantial benefits. Although a shift towards a more distributed model is expected as the technology advances, to date, centralization has yielded the most favorable outcomes.

Several advantages arise from adopting a centralized approach to managing generative AI (Gen AI) within an organization:

• Centralization is advantageous given the limited availability of top-tier Gen AI talent, enabling efficient distribution of skilled professionals across the organization. This approach facilitates the formation of a unified, high-caliber Gen AI team, enhancing teamwork and camaraderie which are crucial for attracting and retaining talent.
• With the Gen AI field rapidly evolving, including frequent updates to large language models and new features, a centralized team is better positioned to keep abreast of advancements than multiple decentralized teams.
• A central leadership model proves especially beneficial in the early stages of a company’s Gen AI initiatives, where critical decisions regarding funding, technological infrastructure, cloud services, language model selection, and strategic partnerships need to be made regularly.
• Managing risks and staying informed of regulatory changes is more straightforward under a centralized system.

However, selecting an operating model for Gen AI isn’t a straightforward choice between centralization and decentralization. Financial institutions should consider the insights presented in this discussion to determine the degree of centralization appropriate for different aspects of their Gen AI operations, customizing their approach to align with their unique organizational structure and culture. For instance, a company might opt for a centralized strategy for managing risk, technological architecture, and partnerships, while adopting a more distributed approach for strategic planning and implementation.

Significance of the Operating Framework

An operating framework illustrates the mechanisms of a company's operation, encompassing its structure (such as roles, governance, and decision-making processes), workflows (including performance oversight, systems, and technology), and human resources (covering skills, culture, and informal networks).

Financial institutions that adeptly harness generative AI (Gen AI) dedicate effort to devising an operating framework specifically tailored to accommodate the unique characteristics and risks associated with this technology, rather than attempting to fit Gen AI into their pre-existing framework. It has been noted that institutions achieving significant benefits from Gen AI commonly adopt a centrally guided approach to managing this technology, notwithstanding the potentially more decentralized nature of other organizational areas. This approach is expected to adapt as the technology progresses.

For a financial services entity embarking on a Gen AI initiative, the optimal operating framework must facilitate scalability and be congruent with the company’s structural and cultural ethos; a universal solution does not exist. An operating framework that is thoughtfully curated and capable of evolving alongside the organization is crucial for the effective scaling of Gen AI.

In summary, an appropriate operating framework empowers a financial institution to efficiently execute three core functions:

• Strategic Navigation: Identify and prioritize Gen AI applications that are in line with the organization’s strategic goals into a value-maximizing, risk-managed roadmap. This includes monitoring the realization of value to ensure judicious resource distribution.
• Standardization: Establish uniform standards (related to technology architecture, data management practices, and risk and control frameworks) to enhance operational efficiency and apply insights from completed initiatives to forthcoming projects.
• Implementation: Develop and assess technical solutions for Gen AI applications, transition those that meet specified performance and safety standards to operational status, and scale them based on their business justification. This also involves monitoring and ensuring the realization of their intended impact.

Archetypal Frameworks for Gen AI Operating Models in Banking

Financial institutions, including banks, adopt various strategies for structuring their generative AI (Gen AI) operating models, which can range from highly centralized to highly decentralized systems.

In a recent analysis of Gen AI applications within 16 of the largest financial entities in Europe and the United States, which together hold close to $26 trillion in assets, it was found that over half of these organizations have chosen a predominantly central-led approach for managing Gen AI. This preference exists even in instances where the organization's typical governance for data and analytics leans towards decentralization. Such centralization is anticipated to be a transitional phase, with a shift towards decentralization expected as the adoption and integration of Gen AI evolve. Over time, it may become advantageous for individual departments within these institutions to independently determine the prioritization of Gen AI initiatives based on their specific needs.

From the financial institutions examined, four distinct organizational archetypes for managing Gen AI have been identified, each presenting its unique set of advantages and drawbacks.

Highly Centralized Approach

Advantages: In this framework, where a singular central team oversees the entire spectrum of gen AI solutions — from their inception to implementation, autonomously from the broader organization — there’s a significant acceleration in the development of skills and capabilities within the gen AI team.

Challenges: The gen AI team may become isolated from the decision-making processes, as well as detached from the business divisions and other organizational functions, potentially hindering its ability to impact decisions.

Centrally Directed, Executed by Business Units

Advantages: This model fosters a greater collaboration between the gen AI team and business units, minimizing resistance and facilitating the adoption of the technology across the entire organization.

Challenges: The necessity for business unit approval and feedback before proceeding may decelerate the gen AI team's application of the technology.

Led by Business Units, Supported Centrally

Advantages: This structure facilitates engagement and support from business units and functions, with gen AI initiatives emerging organically from the ground up.

Challenges: Implementing gen AI applications uniformly across different business units can be challenging, and there may be disparities in the development and application of gen AI capabilities among them.

Highly Decentralized Approach

Advantages: This setup naturally facilitates support and commitment from business units and functions. Additionally, specialized resources within each unit or function can swiftly generate pertinent insights, achieving better integration.

Challenges: Business units operating independently with gen AI may miss out on the expertise and best practices available through a centralized model. Furthermore, there's a risk they may not delve deeply enough into individual gen AI projects to realize substantial breakthroughs.

Optimal Operating Model for Early Gen AI Integration

In the initial phases of integrating generative AI (Gen AI), financial institutions with centralized operating models are showing superior progress. Approximately 70% of entities with highly centralized Gen AI frameworks have advanced to deploying Gen AI use cases in production. This is a stark contrast to around 30% of those adopting a fully decentralized approach. Centralized management allows organizations to concentrate efforts on select use cases, swiftly moving from preliminary trials to addressing the complexities of production and scalability. Conversely, entities employing decentralized methods often find it challenging to advance beyond the pilot phase.

Given Gen AI's emerging status, financial services companies are reevaluating their operating structures to navigate the technology's evolving potential, inherent risks, and significant organizational impact. Over 90% of the participants in a recent McKinsey forum on Gen AI in banking have adopted some form of centralized Gen AI function to ensure efficient resource allocation and risk management.

Survey data reveals about 20% of examined financial institutions employ a highly centralized model, integrating Gen AI strategic direction, standardization, and implementation. Around 30% follow a centrally led but business unit-executed model, centralizing decision-making while delegating implementation. Approximately 30% adopt a model led by business units with central support, focusing centralization on standardization and allowing units to define and pursue their strategic goals. The rest, about 20%, utilize a highly decentralized model, typical of large institutions with the capacity for self-sufficient Gen AI initiatives within their business units.

However, centralization comes with its challenges, including disputes over strategic direction, funding, and talent distribution, with concerns about resource allocation and operational prioritization.

Financial services firms that have adeptly navigated the shift to Gen AI possessed a high degree of organizational flexibility, enabling rapid procedural adjustments and resource reallocation, either centrally or through temporary, centrally coordinated agile teams for specific use cases. Unlike traditional AI teams, Gen AI groups often involve a broader mix of cloud engineers, industry experts, and risk and compliance staff from the outset due to Gen AI development's iterative nature and the early consideration of potential scaling impacts.

As Gen AI technology and its organizational implications mature, we may see a shift towards a more distributed operating model in strategic decision-making and implementation, while aspects like standard setting, particularly in areas like risk management and technology infrastructure, might remain centralized.

Key Decision-Making Checklist for Gen AI Operating Models

As financial institutions embark on the journey of selecting and deploying a generative AI (Gen AI) operating model, leaders must navigate a series of crucial decisions that span various aspects of their organizations. This checklist offers guidance for executives in crafting an optimal operating model for their institutions:

• Strategy and Vision: Determine who among the leadership will spearhead the Gen AI strategy, deciding if this will be an enterprise-wide initiative or segmented by business units. Establish a clear vision for the potential value Gen AI brings and identify the functions or processes most likely to be transformed.
• Domains and Use Cases: Identify who will select the key enterprise domains or clusters for Gen AI application and the specific use cases within those domains.
• Deployment Model: Decide on the approach to deploying these domains and use cases, whether as a “taker” (adopting specific solutions from vendors), a “shaper” (customizing broader vendor solutions), or a “maker” (creating bespoke in-house solutions that could transform the core business).
• Funding: Outline the funding strategy for Gen AI initiatives, which may vary depending on the level of centralization or decentralization of the Gen AI strategy. This typically involves a mix of investments from individual business units and a dedicated central Gen AI team.
• Talent: Define the skills essential for Gen AI projects and strategize on acquiring these through hiring, upskilling, or strategic outsourcing. Additionally, establish the role of “translators” who bridge the gap between business needs and technical implementation of Gen AI.
• Risk Management: Assign responsibility for setting risk parameters (addressing issues like data privacy and intellectual property) and devising mitigation strategies. Consider if existing risk frameworks need adjustments for Gen AI-specific risks and whether certain use cases (e.g., customer-facing ones) require additional governance.
• Change Management: Implement a committee to oversee a change management strategy, ensuring shifts in mindset and behavior necessary for the successful integration of Gen AI across the organization.

An effective Gen AI operating model is crucial for embedding sufficient structure and agility to achieve significant impact across the organization. Financial institutions must tackle key considerations, such as defining the Gen AI team’s role and ensuring the model’s adaptability over time. This adaptability is critical not just at the organizational level but also in specific areas like funding mechanisms.

In the evolving landscape of Gen AI in banking, a strategic approach to operating models is essential. Institutions must strike a balance between innovation and risk management, tailoring their structures to capitalize on Gen AI's capabilities fully. As they navigate this path, the strategies highlighted here can help align Gen AI efforts with overarching strategic objectives for maximum benefit. The journey to scale Gen AI solutions is challenging, and adopting the right operating model is a critical step towards unlocking the technology’s extensive advantages.

In conclusion

The journey of integrating Gen AI into the banking sector is complex and requires a balanced approach that considers the rapid pace of technological advancement, the unique organizational culture of each institution, and the evolving regulatory landscape. By making informed, strategic decisions across these domains, financial institutions can position themselves to leverage Gen AI effectively, unlocking new opportunities for innovation and growth while navigating the challenges that come with such a transformative technology. The path forward is one of continuous learning, adaptation, and strategic foresight, enabling banks and financial institutions to not only achieve short-term gains but also secure long-term resilience and success in the digital age.