How AI Changes the HR Operating Model and HR Roles

By Kevin Kim · Updated Oct 11, 2026

The HR operating model is how an HR function organizes its work: who answers employee questions, who advises managers, and who designs policies and programs. AI changes that model by taking over a large share of routine requests and drafting work, which moves the human effort toward judgment, design and oversight. The shift doesn't remove the need for HR. It changes what each team inside HR is for, and it creates a handful of new roles.

This guide is for CHROs, HR directors and HR managers who are deciding how to restructure HR as AI tools arrive. It's the strategy-level companion to AI in HR: the full guide, which covers the use cases themselves.

Key Facts

• Dave Ulrich's Human Resource Champions (Harvard Business Press, listed as October 1996, widely cited as 1997) described four HR roles: strategic partner, administrative expert, employee champion and change agent (Google Books listing).
• The familiar "three-box" structure of shared services, centres of expertise and business partners grew out of that role thinking. In CIPD's operating-models collection, Allan Boroughs says Ulrich "never claimed to have invented it" (CIPD, Changing HR Operating Models).
• In SHRM's State of AI in HR 2026 research (more than 1,900 HR professionals), only 28% said HR is leading upskilling efforts, while 39% reported AI had shifted job responsibilities, 24% reported new roles and 7% reported job displacement (SHRM, 5 critical insights for CHROs).
• In Deloitte's 2025 Global Human Capital Trends, managers spend nearly 40% of their time on today's problems and administrative tasks and only 13% on developing their people (Deloitte press release).
• Under the EU AI Act, AI systems used to recruit or select people, or to make decisions on promotion, termination, task allocation and performance monitoring, are listed as high-risk in Annex III (EU AI Act, Annex III).

The Model Everyone Starts From: Ulrich's Roles and the Three-Box Structure

Most HR functions of any size are organized around some version of three teams. Shared services (sometimes called HR operations or the service center) handles transactions and employee questions: payroll queries, leave, contracts, records. HR business partners (HRBPs) sit with a business unit and advise its leaders on people issues. Centers of excellence or expertise (COEs) are small specialist teams for areas like compensation, talent acquisition, learning and organization design.

A point of accuracy matters here, because the model is often misattributed. Ulrich's own book proposed four roles that HR professionals need to play at the same time. The publisher's description says HR has to work in all four areas at once to contribute fully (Google Books). The structure that most people call "the Ulrich model" came later, as practitioners turned those roles into teams. CIPD's collection says the book "proposed a differentiation of HR activity based on the nature of HR roles," and that Ulrich "never claimed to have invented" the three-legged structure that carries his name (CIPD). In the same collection, Nick Holley calls it "fundamentally a sound model" but points to "a big gap between intention and reality," especially for business partners.

That gap is the opening AI walks into. If your HRBPs spend their week answering manager questions and chasing approvals, they're acting as shared services with a better title. AI doesn't fix a muddled model. It makes the muddle more expensive, because the work that should have been automated is now automatable.

How AI Changes Each Pillar

The table below is a working view of what shifts in each box. It reflects how the technology fits each team's work, not a forecast for any one company.

Notice that the biggest structural pressure lands on shared services and the biggest skill pressure lands on HRBPs.

Shared services: from answering to managing a knowledge system

Shared services is the box where AI has the clearest early fit. An AI assistant grounded in your policies can answer "how many days of bereavement leave do I get in Germany?" in seconds, by retrieving the relevant policy text before it answers. Ticket classification and routing, and drafting of standard letters, follow the same logic. For a deeper look at assistants that act as well as answer, see AI agents for HR.

What this changes in practice is the job mix. Fewer people answer the same question forty times. More people own the quality of the answers: keeping policies current, fixing the gaps that the assistant exposes, reviewing the conversations it escalates. A shared-services lead's new core skill is closer to running a knowledge base than running a queue.

Josh Bersin, quoted in the CIPD collection, links well-designed self-service systems to a shift toward more specialists and fewer generalists (CIPD). That fits what you'd expect when the simple questions disappear and only the hard ones reach a person.

HR business partners: from doing the analysis to interpreting it

Business partners lose some of the preparation work: pulling attrition by department, summarizing engagement comments, drafting the quarterly people review. They gain time for the part that can't be handed off, which is deciding what the numbers mean for this leader, this team, this quarter.

Managers are the other half of this story. Deloitte's 2025 research found managers spending nearly 40% of their time on today's problems and administrative tasks and only 13% on developing their people (Deloitte). If AI-backed manager self-service handles the "how do I log this?" questions, an HRBP's calendar can shift toward coaching and team design. If it doesn't, the HRBP stays the escalation desk. The tool alone doesn't decide which one you get; the operating model does.

Centers of excellence: from designing programs to governing the machine

CoEs use AI for faster research, first drafts of job descriptions and learning paths, and analysis of survey text. But their larger new responsibility is that AI systems now apply their rules at scale. A compensation team's pay-band logic, embedded in a tool that suggests offers, will be followed thousands of times. A talent team's definition of "qualified," embedded in a screening tool, becomes a decision rule.

That makes CoEs the natural owners of fairness testing, definitions and exception policies. It ties directly into AI governance and the monitoring of bias in AI. Boroughs, in the CIPD collection, noted that most past transformation investment went into shared services and its IT systems, while centres of expertise, especially for talent management, were neglected (CIPD). Repeating that pattern with AI would be a mistake: you'd automate the front desk and leave the rule-makers under-resourced.

What Moves to AI and What Stays Human

A useful test for any task is two questions: is it repetitive and rule-based, and is the cost of a wrong answer low or recoverable? If both are yes, it's a candidate to automate with review. If the answer touches someone's pay, job, health or legal standing, keep a person accountable. This is the idea behind human-in-the-loop design.

Good candidates to hand to AI, with checks:

• Answering routine policy and benefits questions
• Summarizing survey comments, exit interviews and performance review text
• Drafting job descriptions, offer letters and communications
• Scheduling and routing requests
• Assembling the first version of a workforce report

Keep with a human decision-maker:

• Hiring, promotion, discipline and termination decisions
• Investigations and employee relations cases
• Pay exceptions and accommodation requests
• Anything where an affected person would reasonably expect to talk to someone

The regulatory picture backs this split. The EU AI Act's Annex III lists AI systems used to recruit or select people, to make decisions on terms of work, promotion or termination, to allocate tasks based on individual behaviour or traits, or to monitor and evaluate performance, as high-risk (Annex III). In New York City, Local Law 144 requires that an automated employment decision tool be subject to a bias audit within one year of use, with information about the audit made publicly available (NYC Department of Consumer and Worker Protection). These rules apply to hiring and employment decisions specifically, which is where the human-judgment line should already sit.

New Roles Taking Shape

The SHRM research gives the best picture of direction without overselling it. Among more than 1,900 HR professionals, 39% said AI had shifted job responsibilities, 24% said it had created new roles, and 7% reported job displacement (SHRM). The pattern is redesign far more than removal. The roles below are design suggestions for larger HR functions, not job titles that SHRM reported; smaller teams usually fold two or three of them into existing jobs.

An "HR AI product owner" deserves a note because it's the least familiar. The idea is borrowed from software teams: someone is personally accountable for whether the thing works for its users, rather than the tool being "owned" by IT and "used" by HR. Without that owner, assistants go stale, answers drift from policy, and no one fixes it.

Skills HR Professionals Now Need

SHRM's CHRO guidance recommends upskilling that redefines HR roles around AI-enabled work, with training in data literacy, prompt engineering and human-AI collaboration (SHRM). Translated into a practical list:

1. AI literacy. Know what these systems do well, where they fail, and how to read their output skeptically. See AI literacy.
2. Data literacy. Read a dashboard critically, ask where a number came from, and spot a sample that is too small or too biased to trust.
3. Prompting and review habits. Give a model clear context, then check its output against the source. Generative AI for HR teams sets out review rules for everyday drafting.
4. Process design. Break an HR process into steps, decide which steps are automated, and define the handoff to a person.
5. Employment law and ethics fluency. Know which uses trigger audit, notice or oversight duties.
6. Change leadership. Help managers and employees adopt new ways of working.
7. Consultative skills. As the routine work shrinks, the advisory work is what's left, so the ability to challenge a leader respectfully becomes the core of the HRBP job.

One uncomfortable finding sits in the same research: only 28% said HR is leading upskilling efforts (SHRM). HR is meant to be the function that builds the organization's capability, yet in most of those organizations someone else is doing it. Start by reskilling your own team in public, so you have a credible story and real experience when you advise others.

A Practical Sequence for Redesigning the Model

1. Map the work. For each team, list the top 15 recurring tasks and tag each as transactional, advisory or design. Don't start with tools.
2. Pick the first pillar. Shared services is usually the safest starting point because errors are visible and recoverable, and volume is high.
3. Name an owner before you buy. Assign a product owner and a content owner. Decide who reviews escalations.
4. Set decision rights. Write down what AI may do alone, what needs review, and what it may never do. Involve legal and IT.
5. Redeploy before you reduce. Move capacity to higher-value work (manager coaching, analytics, workforce planning) and keep an honest record of what changed.
6. Reshape HRBP expectations. Update role profiles, objectives and development plans so the extra time goes to advice rather than becoming more admin.
7. Review after a quarter. Look at containment rate, escalation quality, employee satisfaction and error reports, then adjust.

Pitfalls to Avoid

Automating a broken process. If requests bounce between three teams today, an assistant will bounce them faster. Fix the process first.

Cutting tier-1 staff before the knowledge base is ready. The people who answer questions today also know where the policies are unclear. Their knowledge is what feeds the assistant.

Leaving CoEs out. A model that applies rules nobody in the expert teams has reviewed is a risk, not a saving.

HRBPs becoming data couriers. If business partners just forward AI-generated reports, leaders will ask why they need one. The value is the interpretation.

No owner after launch. Assistants and workflows degrade as policies change. Without a named owner, quality drops quietly.

Overlooking equity of access. Frontline and non-desk employees may not use the self-service channel. Keep a human route.

Ignoring the manager. Deloitte reports that 73% of organizations recognize the need to reinvent the manager role but only 7% are making great progress (Deloitte). If managers aren't equipped, the work AI frees up in HR just lands back on HR as escalations.

Where to Go Next

If you haven't yet, read the AI in HR overview for the use cases that sit inside each pillar. Then pick one team, run the task map from the sequence above, and decide what the first AI-enabled service, owner and review process will be.