AI for HR leaders: what CHROs ask us about AI

Because AI changes how work is done, and work design is HR's job. Yet many HR teams are on the sidelines. SHRM's State of AI in HR 2026, based on more than 1,900 HR professionals, found just 39 per cent of organisations have implemented AI in their HR functions, and more than half (52 per cent) do not involve HR in AI strategy or vision. [1]

Microsoft's 2026 Work Trend Index, a survey of 20,000 AI-using knowledge workers in ten markets including Australia and India, found organisational factors such as culture, manager support and talent practices account for 67 per cent of AI's real impact, against 32 per cent for individual mindset and behaviour. [2] Those are HR levers. If HR is not in the AI conversation, the biggest drivers of value have no owner.

This is almost always the first question, and rightly so. HR holds the most sensitive data in the business. Our answer has three parts.

• Pay for business accounts. We tell every client the same thing: if a product is free, you are the product. Free consumer tools have historically used conversations to improve models. Paid business and enterprise plans give you contractual controls and training turned off, and an enterprise agreement is as confidential as these tools get.
• Check the settings anyway. Memory and training toggles exist for a reason. Know what is on, what the tool remembers and who in your organisation can see shared conversations.
• Share less. Newer desktop agents can work from a folder on your computer and read only the sections of files they need, rather than you uploading whole spreadsheets. That is better hygiene than pasting a full employee file into a chat.

For data that must never leave your network, private AI is an option. Quantana runs open models on its own Nvidia-powered pods or on a client's existing infrastructure, so sensitive data stays inside. Our private AI guide covers when that is worth it.

Not blindly, and HR should be the function that says so loudest. The mental model we share in sessions is that there are three personalities inside any large language model: a wise expert, a bright but inexperienced graduate, and a seven-year-old who just wants to give you an answer. The models are improving and the expert shows up more often, but you never know for certain which one you are talking to.

So the rule is accountability. As we put it in one session, you cannot say "the car had an accident" when you were driving. The person who uses the output owns it. Treat AI work like work from a capable new hire: useful, fast and always reviewed.

Two habits make answers more reliable. First, ground the model in authoritative sources and ask for the link behind every claim, then check the links. Second, never let AI make or justify a decision about an individual's pay, performance or employment on its own. The Work Trend Index found 86 per cent of AI users already treat AI output as "a starting point, not a final answer". [2] Write that into your AI policy.

We hear this objection often, sometimes from senior leaders whose top teams have taught themselves. They are partly right. People pick up the basics quickly.

The problem is what the basics look like. When companies hand out AI accounts and say "use AI", most people use it like a search engine: summarise this email, draft a slide. That is useful, but it leaves most of the value untouched. The step change comes from learning to hand a tool a whole task, turn a repeated process into a reusable skill that anyone in the team can run, and schedule it so a report is waiting at 9 am every day.

That is what we teach in practice sessions, working through each department's real data together so ideas cross-pollinate. In one workshop at another company, a live analysis of marketing data surfaced a distributor claiming a coupon 900 times in a day. Nobody learns that from a slide deck.

Managers matter most. The Work Trend Index cites a separate Microsoft study in which employees reported a 17-point lift in AI value when managers actively modelled AI use. [2] Training managers first is usually the fastest route to adoption.

Start with data HR already owns and a cost the business already tracks. In a recent session with a manufacturer, the managing director's HR question was not about chatbots. It was whether years of overtime data could help manage the workforce better and reduce overtime costs. That is a good first use case: the data exists, the cost is visible and the outcome can be measured.

SHRM found AI in HR is concentrated in recruiting (27 per cent), HR technology (21 per cent) and learning and development (17 per cent). [1] Other strong starting points we see:

• Overtime and rostering analysis. Patterns by month, site and team, with flags before costs escalate.
• Policy and handbook questions. An assistant that answers from your own documents and cites the clause.
• Reporting. Daily HR dashboards built automatically from system exports instead of a monthly manual merge.
• Data entry. Letting frontline staff dictate updates, in Hindi or any language, instead of filling in forms. Less friction means more and better data.

Two shifts come up in almost every session.

The first is pace. When reports that used to arrive at month end can be produced every morning, teams have to act on information weekly or daily. The culture of "we will look at it at month end" has to change, and that is a people question before it is a technology one.

The second is the leader's role. AI can now generate hundreds of insights from business data. The job moves from producing insights to choosing the five that matter and acting on them. HR has to help leaders build that judgement, and decide how performance is measured when part of the work is done by AI.

Alignment is the gap. Only 26 per cent of AI users in the Work Trend Index say their leadership is clearly and consistently aligned on AI. [2] HR is well placed to close it.