Generative artificial intelligence (AI) has caught fire in the business world. A recent survey by Bain & Company finds that 75% of financial services companies are achieving or exceeding their expected value with generative AI initiatives. That exceeds the success rates of other technology implementations, and it is spreading quickly. Many companies are moving from experimenting to putting generative AI into users’ hands at an impressive pace.

Indeed, speed is essential to thriving with AI, and the primary catalyst for speed is an effective operating model. Therefore, it’s critical for companies to make the right choices about how they organize for AI innovation and deployment. “Right” will not be the same for all financial services companies, given their different circumstances. However, in our work with companies globally, we have recognized emerging patterns that point the way to best practice for most companies.

Specifically, our analysis of about 20 banks, insurers, and payment firms shows how organizations make decisions across four dimensions: program sponsorship, governance, staffing, and funding (see Figure 1). Across these four dimensions, we typically see the most ambitious companies, which aspire to use generative AI at a large scale, centralize their core capabilities and governance, with a plan to decentralize down the road. Let’s review each dimension in turn.

Figure 1

Assigning the ideal sponsor

Given the technical intensity and risks of generative AI, sponsorship typically comes from the senior executives overseeing digital, innovation, or strategy, with 63% of respondents structuring sponsorship that way. Initially, this often comes in close collaboration with the chief technology officer (CTO). In about one-quarter of cases, the CTO or CIO leads the program. The challenge here is to ensure the right level of business unit involvement.

Either way, most of the companies we surveyed involve the CEO to build momentum and bring all C-level executives from the business units and risk, HR, finance, and technology groups on board.

One asset management company we interviewed as part of the analysis created a full-time role to lead the AI hub with a dotted line to the CEO. An executive at this company noted, “You need a senior role to have peer conversations with other C-suite executives.”

By contrast, a heavy emphasis on technology or AI teams for sponsorship might lead to initiatives that are too focused on technological capabilities, potentially neglecting broader business impacts and opportunities across the organization. If a CTO builds and runs a generative AI platform independently while business units build their strategic business plans and roadmaps, it will take substantial time and effort to align on the bank’s priorities and infuse them in the new technology platform.

Arranging governance to span functions

For most companies, governance makes sense at the C-level so that all the functions can work together to address the challenges in using a new technology effectively. For that reason, some 60% of respondents have established a governance council involving C-level executives who make strategic decisions around the generative AI program. They often maintain a separate hub at the corporate center to identify and prioritize use cases.

At one insurance company, the AI council composed of C-level executives sets the guidelines and overall strategy, while an AI hub team prioritizes use cases identified by the business units and makes recommendations to the council. This arrangement is particularly important early on, because companies have limited technical resources capable of building generative AI use cases.

Fewer than one-third of respondents have all governance decentralized, though one or two business units take the lion’s share of governance; where this occurs, the organization tends to be more decentralized overall. The risk here is a fragmentation of effort and resources.

When centralizing, companies should address two potential pitfalls. One is the risk of bottlenecks in decision making, especially in large organizations with many potential generative AI applications. The process of reviewing and prioritizing use cases might slow down innovation and responsiveness to market changes.

A second concern is potentially limited user engagement. Insights and innovative ideas from throughout the organization might not be adequately captured or considered. That could restrict the diversity of applications and innovation, as well as the pace of adoption once the technology is rolled out. One way to mitigate this risk is to embed AI stewards, who can source and validate ideas, in each business unit.

Staffing through the center and business units

The structure of staff resources highly depends on how companies have organized their existing AI and analytics capabilities. For companies with a hub-and-spoke model, staff resources are centralized through the hub or AI factory. The hub takes responsibility for developing the generative AI platform or application, with business units developing specific use cases, building the business case, and validating with end users.

For companies with decentralized capabilities, a more effective structure typically will have teams operating within existing reporting lines. Here, the center focuses on foundational capabilities and making sure the generative AI platform and infrastructure are available across the wider enterprise.

More than half of survey respondents operate with a hybrid staffing model—a dedicated team plus cross-functional, decentralized teams. This model requires balancing business unit autonomy to innovate with the company’s overall generative AI strategy and standards.

In addition to allocating resources to developing use cases, the center plays an important role in ensuring a coordinated approach. This matters especially in the beginning, as much of the underlying technology can be reused in other use cases. In addition, if a company has limited technical resources, the center should support the most complex use cases. An executive at one bank commented, “Business units have tried to build proof of concepts on their own, but don’t currently have the scale. The central function is building blueprints that can be replicated by business units in the future.”

As the platform matures, the role of the center will evolve. The central hub will focus mainly on maintaining the platform, while the business units will have their own resources, such as data scientists, to develop use cases. Over the long term, resourcing will become more federated, with business units leading development, implementation, and delivery of use cases.

To ensure a coordinated approach, most companies are allocating risk, HR, and change management resources from the center. This helps to spur adoption and commitment from all employees and functions. Some companies have chosen to introduce a new generative AI champion role that involves taking responsibility for spurring adoption in the transition phase—a role that might not be needed in the long run.

Shifting the sources of funds over time

Given that generative AI came on the scene in 2023 after the budgets for 2024 had already been established, 80% of our respondents redirected funding from existing initiatives in 2023. But in 2024, companies did make incremental investments. Companies are mostly relying on their center to build the infrastructure and platform early on, as they want to move quickly on this opportunity, rather than wait for a new budget cycle.

Over the long term, however, a common goal is for generative AI to be self-funded. One executive noted, “GenAI is funded by incremental investment that is expected to be recovered by [higher agent productivity, handling a greater debtor workload].”

As companies allocate incremental investments, tracking ROI will be crucial. This involves assessing the impact of initiatives on operational efficiency, customer satisfaction, and new revenue streams.

*     *     * 

With speed of adoption being so important to realizing value from generative AI, the C-suite must govern early efforts. At the same time, initiatives should span functions across the enterprise because there are many moving parts to coordinate, and lessons from successful experiments will travel more quickly to teams starting on a new experiment. Finding that balance, with continual adjustments, will allow companies to raise their level of value, and see more types of value, through this powerful new technology.