The people leader’s AI playbook: five principles for putting AI to work in HR

At AI in HR 2026, Anita Lettink, Future of Work expert and Managing Partner at HRTechRadar.com, shared a practical playbook for closing that gap. Here’s what HR leaders can take away.

Technology is the easy part

HR has moved from predictive analytics, to generative AI, and now towards agentic AI, where agents can resolve requests, update records and analyse data, with people setting the guardrails.

Ambitions are high. According to Gartner, 82% of HR leaders plan to deploy agentic AI within the next 12 months, while only 31% of CHROs feel prepared to build AI into how HR operates.

Gartner also puts it clearly: successful AI transformation depends 30% on the technology and 70% on how well organisations redesign HR processes, workflows and roles.

Five principles for your AI playbook

1. Experiment on purpose, then scale

Give every pilot a named owner and a decision date before it launches. At that date, the decision is simple: scale, fix or stop. Review it with the same rigour as a budget line. Define what scale means upfront, too: which team, function or population it moves to next.

Watch for: pilots that quietly become permanent because nobody owns the decision to end them.

2. Start change management on day one

Name who owns adoption before you name who owns the tool, and keep that person separate from whoever sponsored the use case. Share the communication plan before the tool goes live: what it is, why now, and what changes day to day. Give people who are hesitant clear guidance and someone to raise concerns with.

Watch for: launch surveys that ask “are you excited?” instead of “do you trust this?”

3. Build on data and trust

Trust in AI starts with clean data. Fragmented data leads to AI that’s confidently wrong. Be clear about what’s collected, why and who sees it. Employees and work councils need to see why an answer is the answer, so build visible reasoning, audit trails and privacy safeguards into the flow from the start.

Watch for: data collected for one purpose, like engagement, being reused for another, like performance, without a conversation.

4. Match the tool to the task

Before you reach for AI, ask what the task really needs. Does it need judgement or pattern-finding, or would a process or workflow fix do the job? AI is great at routing HR requests to the right team. A leave request that goes to a manager for approval works perfectly well as a workflow.

When AI is the right fit, redesign the process around the outcome rather than asking which steps AI can replace. Measure success by fit: a solution people find workarounds for isn’t a win.

Watch for: AI becoming the default answer because it’s the most exciting one in the room.

5. Govern as you go

Set light guardrails rather than a committee. Keep a short list of who approves new AI use cases and how long that takes, plus a simple way to flag a tool going sideways, such as odd output, an employee complaint or a spike in errors. Then check back every quarter on IP, privacy and cost.

Watch for: governance that only exists on paper until the first thing goes wrong.

Two decisions to make on purpose

Frontier or local models? Frontier models offer the broadest capability and are the fastest to pilot and easiest to scale. They suit tasks where data isn’t sensitive and volumes stay manageable. Local or open models are a few months behind, but you own the setup. They’re the right choice for sensitive data, high-volume tasks or anything you wouldn’t want to leave the building. Decide per task, not once for the whole organisation.

Who does what? Agentic AI can classify and route requests, fill in forms, find patterns in large amounts of data and guide managers on how to improve results. HR professionals own the edge cases, risks, exceptions and AI oversight, along with anything that needs emotional intelligence, like grievances, illness, bereavement and conflict. In Anita’s words, the goal is not headless HR.

Your next Monday

• Run the IP, privacy and cost test on every vendor pitch you’re evaluating this quarter, before you sign anything.

• Sort your top pain points into two piles: “a workflow already solves this” and “this genuinely needs AI”. Fund the second pile first.

• Pick one narrow, high-volume, low-emotion process. Baseline it, then pilot AI there before anywhere else.

• Decide frontier or local per use case, in writing, so the choice isn’t made by default at renewal time.

• Name one person accountable for change management on day one.

One last question

Anita left the audience with a question worth asking about every AI project: are you redesigning technology, or redesigning work?

Watch the full session

Missed it? Watch Anita’s session.