AI centres of excellence are taking shape. Is HR in the room?

In many organisations AI has spread faster than any formal strategy around it. A tool appears in marketing. A team in operations starts experimenting. Customer service gets the first serious investment. IT draws up guardrails. Legal starts asking questions. Somewhere in the middle of all this HR is often still trying to work out where it fits.

AI adoption does not stay neatly inside technology teams. It reaches into roles, workflows, skills, communication and trust. Once organisations begin setting up AI centres of excellence, steering groups or similar forums to coordinate activity the people questions are already on the table whether anyone has invited HR into the room or not. 

Amazon Web Services (AWS) describes an AI CoE as a dedicated unit that coordinates and oversees AI initiatives, bridging business strategy to value delivery. It presents the model as a response to common challenges, including weak business cases, governance gaps, scaling problems and lack of the right talent. 

This definition helps explain why the model is attractive, but it also opens up the question of what kind of structure can help organisations coordinate AI well and what happens when HR arrives late to that discussion?

David Liddle, CEO of TCM Group and author of People and Culture: A Practical Guide for HR Professionals and Leaders, describes AI adoption inside organisations as growing “like little mushrooms”, popping up in pockets across the business. He sees a multidisciplinary CoE as one way to bring those threads together and put people into the conversation early enough to matter. 

Emily Rose McRae, director analyst in Gartner’s HR practice, is less attached to the label and more interested in the function. In her view some organisations use a CoE and others a steering committee. Either way they need a cross-functional mechanism that can set direction, coordinate activity and keep people aligned around a goal. “Without it, these strategies fail,” she says. 

This is especially relevant now because many HR teams remain some way down the adoption curve. In plenty of businesses the early AI energy has gone into software engineering, operations, marketing and customer-facing functions where the commercial payoff is easier to see. 

Liddle is blunt about the state of play inside HR, calling adoption “lamentable” and warning that the function has a real problem if it still sees AI as someone else’s agenda. He recounts being at an HR event where attendees appeared to view him as some sort of magician-like Gandalf as he shared some of the GPTs he has created: “If this is the adoption of AI within the HR community, and with AI being a people agenda and not a technical agenda, then we have real problems, because AI is coming at us so fast.”

What should an AI centre of excellence do?

An AI centre of excellence should define what the organisation wants AI to achieve and connect that goal to decisions about tools, governance, skills and workflow. It sounds straightforward but is still one of the areas where many organisations fall short.

McRae returns to this point repeatedly. “The first question I always ask a client when they come to me to talk about AI is, what is your organisation’s goal with AI? What are you trying to do?” she says. “If you don’t have the goal or business outcome it’s really hard to evaluate if you’ve been successful.”

Plenty of organisations feel they need to “do AI” and set up a group to prove momentum. Fewer can explain clearly whether they are trying to improve service, redesign a process, build internal capability, increase speed in a specific workflow or cut cost in a defined area. McRae’s warning is that a steering group formed in a fog of AI FOMO is likely to spend its time circling around tools and approvals without building much coherence. 

AWS makes a similar point in more formal language. It recommends a mission aligned to strategic objectives and a clear value proposition, quantified through metrics such as cost savings, revenue gains, user satisfaction, time savings and time-to-market. A CoE earns its keep by linking AI activity to business outcomes rather than simply managing a portfolio of experiments. 

McKinsey’s 2025 Global Survey on AI suggests this kind of structure is already becoming part of how organisations are trying to capture value from gen AI. Its research finds that organisations are selectively centralising elements of AI deployment, with risk, compliance and data governance often handled through a fully centralised model such as a CoE. Other areas, including technology talent and adoption of AI solutions, are more likely to sit in a hybrid model, with some resources held centrally and others distributed across functions or business units.

Where AI centres of excellence lose momentum

AI CoEs lose momentum when they drift too far towards policy, permissions or narrow technical ownership. McRae sees this often. Some teams become “bogged down in permission and policy” which, while important, can prevent the level of innovation that organisations are hoping to achieve. 

Others are so dominated by IT or AI enthusiasts that they misread low adoption as a skills issue and prescribe more training. McRae has also seen legal- or compliance-led groups that move more slowly, which can be fine in some contexts, though it still requires alignment with the wider business. Her broader point is that overly technical groups often miss the change management reality of AI at work. Employees may already be overloaded, sceptical or unclear on why a tool helps. “There’s this change management process that doesn’t always get picked up when it’s too tech centric,” she says.

The important point is that AI adoption needs to be in the lived experience of work. People are not using it in the abstract. They are using it inside already crowded days, fragmented processes and imperfect systems. A CoE that stays remote from those conditions is unlikely to drive much value.

Why the CPO needs a voice in AI governance

The CPO does not need to own every technical choice, but the HR function does need a serious role in shaping how AI changes work.

As Liddle says, AI is a people agenda, not a technical agenda. He sees the value of an AI CoE in its ability to “put people at the centre of the AI conversation”, bring in diverse voices and anchor AI as part of a people strategy rather than a narrow efficiency push.

In McRae’s view HR and people teams are critical because somebody must help business leaders think through workflow redesign, role changes, handoffs, skills and communication. “How will roles and workflows need to change to get the most out of this?” is the question she recommends CHROs ask on any steering committee. She is also clear that IT should not be trying to redesign work on its own.

This gives the CPO a practical mandate. HR’s role is in redesign, enablement, capability-building and employee confidence. The risk of leaving these discussions to technology or compliance teams is obvious. Decisions will get made but without the people lens they require.

How AI literacy changes the debate

The European Commission’s guidance on Article 4 of the AI Act says that providers and deployers of AI systems must take measures to ensure a sufficient level of AI literacy among staff and others dealing with AI systems on their behalf. The guidance defines AI literacy in terms of skills, knowledge and understanding that allow people to make informed deployment decisions and understand opportunities, risks and possible harms. The obligation has applied since 2 February 2025, while supervision and enforcement rules apply from 3 August 2026 onwards. The Commission also says there is flexibility in how organisations respond and that no specific governance structure is mandated. 

This flexibility makes the CoE debate more relevant. Organisations still need a mechanism for deciding who requires what level of AI understanding, how guidance is tailored by role and risk and how literacy links to responsible use, human oversight and day-to-day practice. For HR leaders this brings a familiar mix of questions around learning, communication and role-specific support into the core of the AI agenda. 

Why access to tools still falls short of value

McKinsey’s analysis found that, among 25 organisational attributes tested, workflow redesign had the biggest effect on an organisation’s ability to see EBIT impact from gen AI. Yet only 21% of respondents whose organisations use gen AI said their organisation had fundamentally redesigned at least some workflows. The same research also found that CEO oversight of AI governance was one of the elements most correlated with higher self-reported bottom-line impact, particularly in larger organisations.

As we have seen in many automation programmes, tool access on its own rarely creates much value. The gains become meaningful when workflows and responsibilities change around the technology. 

Saving a few minutes here and there through a general-use AI tool may feel productive but those fragments of time usually dissolve into the rest of the day. Without redesign these savings rarely add up to much. As McRae says, this gained time can just as easily be taken up with “checking email and a cup of coffee.”

She recounts the experience of a professional services firm that built a tool to help draft responses to requests for proposals and assumed proposal writers should use it directly. She advised a different workflow where proposal experts should spend their time reviewing and refining the output of the tool. She then pushed the client to think about onboarding and delivery teams, because winning more work without the capacity to deliver it would simply move the bottleneck downstream. Organisations can, as she put it, “address one problem and create a new one”. 

This is one of the strongest arguments for a cross-functional CoE or steering group. It creates a forum where adjacent teams, downstream effects and real operating constraints can be surfaced before AI use cases turn into expensive detours.

What HR leaders should do next

HR leaders should find out whether a formal AI group already exists in their organisation, what it is trying to achieve and whether HR has a voice in it. They should ask how the organisation is approaching workflow redesign and AI literacy. Critically, they should also stop waiting for perfect confidence before stepping in.

As McRae says: “If someone can’t get to the table for various reasons internally, then it’s time to build your own HR steering committee that includes cross-functional expertise from within HR and externally and to demonstrate why you need a cross-functional team doing this.”

Liddle stresses the urgency of getting involved now. HR cannot assume its traditional language around empathy and people will protect it from AI. He argues that a great deal of HR work remains administrative and transactional, which makes it highly exposed if the function stays passive while other leaders pursue efficiency and cost reduction. His advice? Engage now.

AI centres of excellence can shape the conditions in which AI becomes embedded across the organisation. Some will work well while others may stay procedural or use a different label entirely. What they are called doesn’t matter. What does is that organisations need a way to connect AI strategy, governance, literacy and real work. HR must be in that conversation early enough to influence it.