Accelerating Banking Innovation: Why a Central Gen AI Blueprint Works Best

Generative AI is reshaping financial services by creating new opportunities for efficiency and innovation. Banks around the globe have taken notice, using gen AI to boost customer-facing chatbots, fight fraud, and handle labor-intensive tasks such as coding, drafting pitch books, and summarizing regulatory reports.

According to the Santiago & Company research and MGI research, Gen AI could generate between $200 billion and $340 billion in annual value across the global banking sector—equivalent to 2.8 to 4.7 percent of total industry revenues—primarily through gains in productivity.¹ In their haste to tap into these opportunities, however, financial institutions have also encountered obstacles. Getting gen AI right can yield outsized rewards, but mistakes may bring complications. Risks include the creation of inaccurate or unreasonable outputs, intellectual property conflicts, opacity around how models work, bias, fairness, security threats, and more.

In an earlier discussion, we examined strategies that could help banks capture the full value of gen AI. Beyond simple proofs of concept, truly harnessing gen AI requires strong capabilities across seven interlinked dimensions:

1. Strategic Road Map
2. Talent
3. Operating Model
4. Technology
5. Data
6. Risk and Controls
7. Adoption and Change Management.

These dimensions function as a tightly woven network. For instance, even the most robust operating model rarely succeeds unless the right talent and reliable data support it.

This article focuses on one of those seven dimensions: the operating model. "Operating model" means a business's blueprint for transforming strategy into action. Future articles will explore other dimensions. Here, we explain what an operating model is, why it matters, and how banks organize around gen AI—highlighting the model that has produced the most substantial results so far. We also outline the decisions financial institutions must consider as they design their gen AI operating models.
‍
Our findings suggest that a high degree of centralization works best for gen AI operating models across various industries. When pilot use cases are launched in disconnected silos, it becomes challenging to scale the most promising ideas. In banking specifically, organizations that centrally oversee gen AI often achieve more tremendous success. Over time, banks may lean more toward a federated approach as the technology matures. However, financial institutions have found that centralized operating models deliver the most tangible gains for now.

Why a centrally led gen AI model delivers results

Central oversight brings significant benefits in several ways. First, banks face intense competition in the hunt for top-tier gen AI talent. By pooling scarce specialists and data scientists under a single umbrella, the organization can assign them where they will have the most significant impact. A unified, expert core also fosters a sense of camaraderie among these professionals, which encourages both recruitment and retention.

Second, gen AI evolves with startling speed. New large language models and features appear regularly. A central team can track these developments far more effectively than multiple small groups scattered across the institution. This capability is especially valuable during the early stages of gen AI adoption when leadership must make weighty decisions about funding, partnerships, technology architecture, and choosing a large language model or cloud provider.

Third, a centralized function allows for better management of risk and regulatory demands. It becomes easier to adopt consistent controls and frameworks, reducing the likelihood of violations or oversights. In financial services, where stringent rules govern data privacy and security, the ability to enforce uniform standards is often critical to success.

However, operating models need not be strictly "all or nothing" regarding centralization. A financial institution might decide to keep certain elements—for example, technology architecture, risk protocols, or partnership agreements—under centralized stewardship while granting business units a more independent hand in shaping their own strategic priorities and execution plans.

Operating-model archetypes for Gen AI in banking

Banks take different approaches when structuring their gen AI activities. Some centralize responsibilities completely, while others distribute them across business units. A recent review of gen AI initiatives at 18 major European and US financial institutions (covering nearly $26 trillion in assets). Over half embraced a more centralized structure for gen AI, even if their existing data and analytics frameworks had been relatively decentralized. For many, this central design will last only until gen AI achieves greater maturity—at this point, a federation might make more sense, allowing individual business lines to prioritize projects as they see fit.

In these studies, four organizational archetypes emerged: highly centralized, centrally led (with execution in business units), business unit–led (with central support), and highly decentralized. Each has clear benefits and drawbacks.

Highly centralized.

A single team independently manages everything from concept to execution, which accelerates skill-building but risks isolating the gen AI function from everyday business decisions.

Potential Benefits

• Consistent strategic direction: A single authority sets the vision and roadmap, ensuring that gen AI initiatives align tightly with corporate strategy and risk tolerance.
• Easier to maintain standards and governance: Centralized data governance, security protocols, and model risk management can be uniformly enforced across all projects.
• Efficient resource allocation: Funding, infrastructure, and highly specialized AI talent (e.g., data scientists, data engineers) are pooled in one place, potentially reducing duplication of roles.

Potential Drawbacks

• Potential bottlenecks: A single unit can become overwhelmed if demand for gen AI services grows faster than staffing or infrastructure capacity.
• Slower responsiveness to local needs: Because every request or requirement has to route through the central team, it can be difficult to accommodate the different paces or specific demands of each business line.
• Cultural barriers to adoption: Business units may view the centralized group as “outsiders,” making it harder to secure cooperation or enthusiastic buy-in.

Centrally led, business unit executed.

The enterprise still makes fundamental decisions at the center, but business units handle project execution. This arrangement encourages deeper collaboration, though it may slow implementation if multiple sign-offs are required.

Potential Benefits

• Balanced governance: The central team provides direction and ensures compliance with risk and regulatory requirements, while business units implement solutions tailored to their unique contexts.
• Greater collaboration: Cross-functional project teams combine centralized expertise (e.g., advanced AI skills) with business-specific knowledge, often leading to more robust solutions, and retained decision-making authority, it can promote consistent data standards and model governance across multiple lines of business.

• Accelerated learning: Central oversight helps share best practices; business units can more quickly adopt lessons learned from other parts of the organization.

Potential Drawbacks

• Complex approval processes: If multiple stakeholders or sign-offs are required—especially from both the center and the business unit—projects may slow down.
• Risk of conflicting priorities: The central team’s agenda might not always align perfectly with a business unit’s immediate needs, creating tension or delays.
• Limited empowerment: Business units might feel they lack full ownership over decisions, which can impact motivation and creativity.

‍Business unit led, centrally supported.

Each business unit steers its own gen AI strategy, with the central team setting standards or providing specialized expertise. Buy-in tends to happen quickly, yet broad, cross-functional use cases can be more complex to scale.

Potential Benefits

• High degree of ownership: Business units have autonomy to customize gen AI solutions that align tightly with their customers, products, and operational workflows.
• Faster buy-in and execution: Because the business lines themselves choose which projects to pursue, they are often more invested in seeing them succeed, speeding up adoption. Each business unit can prioritize use cases most relevant to its market segment, customers, or risk profile, leading to more precise outcomes.

• Flexible central support: The central team can focus on enabling success by providing high-level standards, specialized consulting, or advanced technology resources on demand.

Potential Drawbacks

• Complexity in scaling: While local projects may thrive, rolling them out to other units or across the enterprise can be challenging if processes and data standards differ widely.
• Potential fragmentation: If each business unit runs its own tools and processes, it can lead to inconsistencies and hamper data sharing or enterprise-wide analytics. Ensuring uniform controls, governance, and compliance becomes harder when each unit operates more independently.

• Limited influence of central team: If the central function is perceived only as a support center, it may struggle to enforce enterprise-wide best practices or strategic alignment.

‍Highly decentralized.

Every business unit runs its own gen AI programs, which fosters tailored solutions and faster local decisions but can cause fragmentation, duplication of effort, and inconsistencies in risk controls or data practices.

Benefits

• Maximum autonomy: Each business unit can adopt gen AI at its own pace, focusing on the projects that matter most locally.
• Faster local decision-making: Approvals, funding, and resource allocation often happen within the unit, speeding up project kick-offs and development cycles. Decentralized teams can more freely experiment, potentially leading to a broader range of creative use cases.

• Adaptive to local needs: Frontline units can quickly respond to shifts in their markets or customer demands without navigating centralized governance layers.

Drawbacks

• High potential for duplication: Multiple teams might tackle the same problems or re-invent similar solutions without realizing it, wasting resources.
• Inconsistent standards and controls: Security, compliance, and governance frameworks can fracture, which is especially risky in a regulated environment like banking. Siloed data systems across business units can limit the potential of large-scale AI models that need integrated data.
• Limited economies of scale: Each unit invests in its own infrastructure, talent, and development, potentially missing out on bulk purchasing or centralized support that could lower costs.

The operating model with the best results

In this early phase of gen AI's evolution, those adopting centralization appear to have a head start. Around 70 percent of financial institutions that use a highly centralized gen AI operating model have already brought use cases into production.² By contrast, of those with a highly decentralized approach, only about 30 percent have cleared that hurdle. A central steering body can channel resources into a few high-potential ideas, push beyond initial pilots, and solve the operational challenges needed to scale. Banks often struggle to move beyond the proof-of-concept stage when responsibilities lie too far apart.

This inclination toward centralization is evident in how large institutions describe their gen AI capabilities. In a recent Santiago & Company forum on gen AI in banking, more than 90 percent of participating institutions had developed a centralized gen AI function to align resources while managing operational and regulatory risks. Nearly 20 percent of these institutions chose the "highly centralized" archetype—centralizing decision-making, technical execution, and strategic direction. Another 30 percent used a "centrally led, business unit–executed" approach. Roughly 30 percent preferred "business unit–led, centrally supported," while only 20 percent were "highly decentralized." The latter group mainly included large organizations whose business units command significant resources independently.

Of course, centralization can spark friction. Debates commonly arise over where to allocate funding, how to chart the strategic road map, and whether specific business units risk losing critical talent. Financial institutions that adapt most seamlessly already have a measure of organizational agility, which lets them pool and shift resources more readily. They also form coordinated squads that involve not just data engineers and AI experts but risk, compliance, and cloud professionals right from the beginning. Gen AI development tends to be highly iterative, so early collaboration helps teams anticipate and manage broader implications before a use case scales.

As institutions gain a better grasp of gen AI's potential and limitations, they may tilt toward a more federated model. While standard setting—especially for risk management, tech architecture, and partner selection—will likely stay centralized, the actual decision-making and execution might eventually disperse throughout various business functions.

A checklist of essential decisions

When choosing a gen AI operating model, financial institutions face critical decisions that affect not only the operating model itself but also broader areas across the enterprise:

1. Strategy and vision. Who will shape the bank's gen AI ambitions, and will they do so at an enterprise-wide level or for each business unit? Banks need a clear sense of potential value and an understanding of which processes or functions could benefit most.
2. Domains and use cases. Institutions must decide who identifies the business units where gen AI can have the greatest level of impact. They also need to determine who defines each use case and monitor success.
3. Deployment model. Will the organization buy targeted solutions ("taker"), integrate vendor offerings to create broader solutions ("shaper"), or build in-house technology ("maker") that redefines how the business operates?
4. Funding. Centralized models often pool funds in a gen AI center of excellence, with additional support from individual business units. More decentralized structures may parcel out funds to each unit.
5. Talent. Financial institutions should pinpoint the general AI skills they need and decide how to secure them. A mix of hiring, upskilling, and outsourcing is standard, along with employing "translators" who can bridge business and technology demands.
6. Risk. Banking leaders must determine who sets guardrails for data privacy, intellectual property concerns, and other key risks. Given gen AI's novel challenges, they must also decide if existing frameworks for regulatory and reputational AI need recalibration.
7. Change management. A dedicated group must spearhead the behavioral and cultural changes that will allow gen AI to take hold at scale.

Banks cannot effectively combine speed, ambition, and structure without a purposeful operating model to achieve enterprise-wide impact. To select a model that fits, executives should clarify the gen AI team's mandate and build enough flexibility for the organization to adapt as it learns. Such flexibility matters at a high level, for instance, in governance and budgeting and in more practical issues like funding each use case.

Banks and other financial services companies can maximize Gen AI's value by aligning operating models with strategic objectives while preserving a strong focus on risk management. The experiences discussed here suggest that establishing central oversight pays off in the early stages. As gen AI capabilities deepen, banks may devolve more decision-making power to business units. Regardless of the structure, the ultimate success of gen AI initiatives hinges on building a model that can evolve with the technology, handle its complexities, and consistently deliver tangible results.