Turn your AI center of excellence from sidelines to impact

Challenges to successful AI CoE deployment

Simply establishing an AI CoE doesn't automatically translate to successful deployment or scaling of AI projects. The following are some of the challenges that might cause these groups to underperform:

1. Lack of strategic clarity and business alignment

AI CoEs need a mission that's clearer than simply "doing AI." They must support AI investments in ways that align with broader business initiatives and strategy.

To this end, it's important for the business to first define its primary objectives for investing in AI and then ensure the AI CoE can support them. Those objectives could range from saving money and speeding operations to developing new products. Whatever they are, the CoE must align clearly with them.

2. Weak executive support

AI CoEs need committed executive leadership support to fulfill their missions. Specifically, business leaders must ensure AI CoEs are adequately staffed and funded, and they also should encourage stakeholders to use their resources. Otherwise, an AI CoE might end up being an initiative in name only, without the buy-in needed to ensure their success.

3. Lack of organizational engagement

Alongside a lack of adequate executive support, there's a parallel risk that other members of the organization won't buy into the AI CoE, even if business leaders do. 

This usually happens when nonleadership-level employees view the AI CoE as a gimmick designed to make the business appear to take AI seriously, rather than an actionable resource that will help drive AI success in practical ways. Employee sentiment that views AI as a threat can also contribute to low organizational engagement.

4. Excessive bureaucracy

The goal of AI CoEs is to streamline AI adoption. But they can do the opposite when bureaucratic processes weigh them down, hampering other groups or functions within the business from accessing CoE resources and guidance.

For instance, if those seeking help from the AI CoE with an AI project must fill out a long application and wait in a queue, it probably won't accelerate the project as it should.

5. Failure to evolve alongside AI initiatives

For most businesses, AI investment is an incremental process. It begins with pilot projects, followed by the launch of a small number of high-priority AI tools or systems. Only later do AI initiatives arise within other parts of the business.

The types of resources and expertise the business needs at different stages of the AI journey vary. For instance, during AI experimentation, infrastructure requirements are usually much lower than for full-scale production AI deployments.

A successful AI CoE is one that evolves along with enterprise AI needs. But an AI CoE that fails to adapt -- for example, by not scaling up its resources as the business moves beyond small-scale AI experiments -- is likely to underperform.

6. Narrow approaches to AI support

Effective AI implementation requires guidance and resources spanning multiple domains: data science, IT, cybersecurity, compliance and business strategy. If AI CoEs don't address all these needs effectively, they'll fail to achieve their goals.

This tends to be a problem when businesses use AI CoEs to solve just one type of AI challenge, such as mitigating AI security or compliance risks. A more constructive approach is to treat these centers as a holistic resource that can address whatever issues arise as the business implements and scales AI.

7. Excessive or permanent reliance on the CoE

An effective AI CoE is one that kickstarts AI projects by providing initial guidance and resources. It should teach the rest of the business how to fish, not fish for them, to use an old analogy.

AI CoEs shouldn't become permanent fixtures or dependencies for the business's ongoing AI success. If they do, they're likely to prevent the organization from continuing to innovate and scale in the long term.

Repositioning the AI CoE to maximize business value

Fortunately, the challenges facing AI CoEs have solutions. Business leaders can reboot stalled or underperforming AI CoEs through actionable strategies like the following:

• Hone the CoE's mission. For AI CoEs with vaguely defined goals, it's important for business leaders to develop clear guidelines for what the CoE should do and how its work aligns with business objectives. Where necessary, articulations of goals should be paired with changes to the CoE's structure, tooling and staffing to ensure its operations align with stated objectives.
• Increase CoE resource allocations. A simple fix for AI CoEs lacking adequate resources is to allocate more budget or staff to them. As long as these changes are made strategically -- rather than tossing more dollars or employees at the CoE without defining how to use them -- more generous resource allocations can help reposition lagging CoEs.
• Educate employees. Increasing employee awareness of the AI CoE and how to work with it is an important fix for businesses that have invested in AI CoEs but find that few stakeholders are taking advantage of them.
• Publicize CoE success. Showcasing successful AI CoE engagements can also help boost employee awareness and buy-in. For instance, if a team deploys a new AI tool with CoE help, publicizing the project internally is a fast and low-investment way to encourage other stakeholders to work with the CoE.
• Report to the board on CoE status. Requiring executives to keep enterprise board members informed about what the AI CoE is doing can help increase and sustain executive support for the initiative. It's also a good way to prime the board to approve requests for more CoE resources, should this become necessary.
• Broaden the CoE's scope. For AI CoEs focused on specific aspects of AI adoption, such as cybersecurity or compliance, expanding their functional and capability scope could make the CoE more useful. This might require more budget or hiring additional staff with expertise in areas that the CoE doesn't currently cover. But the investment is worth it if it turns a failing CoE into a successful one.
• Define a sunsetting plan. To reduce the risk of turning an AI CoE into a permanent part of the business, define a plan that lays out how the CoE will scale down and, eventually, close. This can be done by identifying milestones -- such as the successful launch of a certain number of AI projects -- that will prompt downsizing of the CoE.

In addition to adopting best practices for existing AI CoEs, business leaders can help prime future AI CoEs for success by following a comprehensive implementation roadmap from the start, which should include the following key steps:

Chris Tozzi is a freelance writer, research adviser, and professor of IT and society who has previously worked as a journalist and Linux systems administrator.