What I learned from rolling out AI to an HR team

When organisations talk about introducing AI, the conversation often starts with the technology. Which tool should we buy? What can it do? How quickly will it save us time?

Those are reasonable questions, but they weren’t the questions that determined whether AI was actually used by the HR team.

The reality was much more human.

Giving people access to Microsoft Copilot didn’t automatically mean they knew how to use it, trusted it or could see how it fitted into their working day. Some people were curious and started experimenting immediately. Others were cautious. Some could see dozens of possible uses, while others opened the tool, stared at the empty prompt box and quietly returned to the way they had always worked.

That experience taught me that rolling out AI isn’t primarily a technology project. It is a change, capability and trust project.

Access is not the same as adoption

One of the easiest mistakes is to assume that once people have licences and some introductory training, adoption will naturally follow.

It rarely does.

People are already busy. Learning how to use a new tool takes time, and the benefit isn’t always obvious at the start. Asking someone to “experiment with AI” can feel like giving them another task without explaining which existing task it is meant to improve.

The most useful conversations weren’t about everything Copilot could theoretically do. They were about the work sitting in front of us.

Could it help turn meeting notes into a clear summary? Could it provide a starting structure for a policy? Could it help organise information from several documents? Could it suggest questions for a difficult conversation? Could it make a long piece of communication clearer and easier to read?

Once the team could connect AI to genuine tasks, it became less abstract and more useful.

What worked

• Starting with everyday work

  The strongest use cases were rarely the most dramatic. AI was most useful when it helped with the repetitive thinking and drafting that surrounds HR work. Structuring information, creating a timeline, producing an initial draft, identifying themes and improving clarity all gave people somewhere practical to begin.

  
  

  This also made the learning feel relevant. People weren’t attending generic prompting sessions and then being expected to work out how the content applied to them. They were learning through their own work.

  
  

• Creating space to learn together

  Confidence developed faster when people could share what they had tried. A prompt that worked well for one colleague often helped someone else see a completely different opportunity. Equally, sharing examples where the output had been poor helped everyone understand the limitations.

  
  

  This mattered because people needed permission to say, “That didn’t work.” AI adoption becomes risky when people feel pressure to present every experiment as a success. Some outputs will be vague, inaccurate or completely unsuitable. Talking openly about that helps people develop judgement rather than blind confidence.

• Keeping human review visible

In HR, a polished answer can be particularly dangerous.

An AI-generated letter might sound professional while missing an important fact. A summary might quietly remove a piece of context. Suggested questions might appear balanced but contain assumptions or inappropriate language.

The team needed to understand that fluent writing wasn’t evidence of a correct answer.

AI could support the work, but the HR professional remained responsible for checking the facts, applying organisational context, considering employee impact and deciding whether the output was appropriate.

• Putting boundaries in place

Governance can sound like the part that slows innovation down. In practice, clear boundaries made people more comfortable using the technology.

People needed to know what information they could enter, what information they shouldn’t enter, which tools had been approved, when an output needed additional checking, where AI shouldn’t be used and who remained accountable for the final decision.

Without those boundaries, cautious users tended to avoid the tool altogether, while confident users risked going further than the organisation intended. Good governance didn’t prevent experimentation. It gave people a safer space in which to experiment.

What didn’t work as well

• Expecting people to explore independently

  Some people will happily spend an afternoon testing prompts. Many won’t, particularly when they are dealing with employee relations cases, recruitment deadlines, payroll queries and the normal pressures of an HR function.

  
  

  Telling people to “have a play” wasn’t enough. They needed examples, time, reassurance and an opportunity to practice using tasks that made sense to them.

• Focusing too heavily on prompting

Prompting matters, but it can easily dominate AI training.

A detailed prompt can improve an output, but it can’t compensate for poor judgement, missing context or an unsuitable task. Someone can write an excellent prompt and still use AI in a situation where it shouldn’t be making the call.

The more important skill was learning how to assess the result. Is it factually accurate? What has it assumed? What might be missing? Does it reflect our policies and circumstances? Would I be comfortable explaining how this was produced to the employee concerned?

• Assuming everyone would progress at the same pace

The difference in confidence across the team was significant. Some colleagues moved quickly from basic drafting to more sophisticated uses. Others needed more support with the fundamentals. That wasn’t resistance or failure. People were starting from different levels of technical confidence, experience and interest.

A single training session couldn’t meet every need. Successful adoption required repeated opportunities to learn, practical support and recognition that confidence develops through use.

• Treating time saved as the sole measure of success

AI can save time, but speed is not the whole story.

Sometimes the value came from creating a better first draft. Sometimes it helped someone consider an issue from another perspective. Sometimes it reduced the mental load involved in starting a difficult piece of work.

There were also occasions when reviewing and correcting an AI-generated response took longer than completing the task without it. That didn’t mean the rollout had failed. It meant we needed to become better at recognising when AI was the right tool and when it wasn’t.

What I would do differently next time

I would spend less time demonstrating features and more time understanding how it can be applied.

Before issuing licences, I would identify a small number of genuine use cases with the team. I would be clearer about which problems we were trying to solve and how we would know whether the tool had helped.

I would introduce governance and practical experimentation together, rather than treating them as separate stages.

I would give managers more support in leading the change. Employees take their cues from their managers. When a manager is either dismissive of AI or unrealistically enthusiastic about it, the team notices.

I would also build in more time for reflection. Which uses are creating real value? Where are people still lacking confidence? What risks or unexpected behaviours are emerging? Which tasks should remain entirely human?

AI adoption isn’t a launch event. It is an ongoing process of learning, reviewing and adjusting.

The biggest lesson

The biggest lesson was that people don’t adopt AI because they have been told it is transformative.

They adopt it

when they understand why it is relevant, feel safe using it, have the skills to review its work and remain confident that their professional judgement still matters. HR has an important role here, not because it should own every part of AI implementation, but because it understands how people experience change.

IT can provide secure systems. Legal and data protection specialists can guide compliance. Senior leaders can provide sponsorship and direction. HR can help connect all of that to the reality of work, capability, behaviour, trust and culture.

That is where AI adoption succeeds or fails.

These experiences have shaped much of my thinking behind HRnetics and the HEART Framework. The aim isn’t to persuade HR teams to use AI everywhere. It is to help them use it purposefully, responsibly and in ways that keep people and professional judgement at the centre.