An AI strategy for HR sets out how the HR function will use AI in its own work and how it will lead the people side of AI adoption across the organisation. It starts from an inventory of HR tasks, sorts them by how close they come to decisions about people, and builds in human review, legal checks, and adoption support.
HR leaders are usually handed AI twice. First, someone asks how HR will use it. A few weeks later, someone asks HR to help the rest of the business adopt it. These are different jobs with different risks, and a strategy that only covers the first one tends to stall on the second.
This guide is for the HR director or people lead who has been given the AI brief on top of an already full role. It covers where to start, what the law in South Africa and Europe already asks of you, and how to get from a policy document to people working differently on Monday morning.
The first job is inside the function. HR work is heavy on documents and repeated questions: role profiles, job adverts, onboarding packs, policy queries, and scheduling. AI can take a real share of that effort, which gives the team more time for the conversations that need a person.
The second job is across the organisation. AI changes how almost every role works, which makes it a people change as much as a technology change. Training, communication, new ways of working, and the understandable worry that AI puts jobs at risk all land on HR's desk whether or not anyone planned it that way.
HR professionals increasingly see this. A Culture Amp survey reported by HRreview found that 47% of respondents believe HR should own their organisation's AI strategy, up from 37% a year earlier. The sample was small, 264 members of Culture Amp's own community, so treat it as a signal rather than a benchmark.
My view is that HR should lead the people side of AI strategy, and should not try to own all of it. Decisions about platforms, data, and architecture need people who understand systems, and a strategy owned by a single function tends to become that function's project. The arrangement I recommend is HR and the technical lead holding the strategy together, with the managing director making the final calls.
The fastest way to waste an AI budget in HR is to start with a vendor demo. Every HR platform now has an AI feature, and each one looks useful in isolation. Without a picture of your own work, you have no way to judge which of them matters.
Start instead with an inventory. Ask each person in the team to list what they spent their time on last week. For each task, capture four things:
This takes a few hours and it changes the conversation. Instead of asking "which AI tool should we buy", you are asking "which of these 40 tasks would benefit, and which should stay firmly with people". We follow the same order in every engagement: diagnose first, then map where value sits, then build. If you want a structured version of this exercise for the whole organisation, our guide to running an AI readiness assessment walks through it.
Once you have the inventory, sort each candidate use case by one question: how close does the AI get to a decision about a human being? That single test does more to keep an HR strategy safe than any framework I know.
Drafting role profiles from a hiring manager's notes, first drafts of policies, onboarding plans, and internal communications. A person reviews everything before it goes anywhere. The risk is low, the time saved is visible, and it gives your team daily practice with the tools. Start here.
An internal assistant that answers routine questions about leave, benefits, or policy. This is useful, and it needs more care. Answers should come only from approved policy documents, there must be an obvious route to a person, and employees should know they are talking to AI. If you operate in the EU, that last point is already a legal requirement for chatbots.
CV screening, shortlisting, interview scoring, performance ratings, and promotion recommendations. This is where the risk sits. A model trained on your past hiring data will learn your past hiring patterns, including the ones you would not choose to repeat. The best-known example remains Amazon's experimental recruiting engine, which Reuters reported learned to penalise CVs from women because it was trained on a decade of applications that came mostly from men. Amazon scrapped it.
AI still has a place in recruitment, on firm conditions. A person makes the decision, you test the tool for uneven outcomes across groups before relying on it, and you tell candidates how AI is used.
Productivity scoring, sentiment analysis of internal messages, and similar tools. Some organisations use them. I would advise most mid-sized organisations to stay away, because the damage to trust usually outweighs whatever you learn. Employees can tell what kind of AI programme they are in, and that sense decides whether they adopt anything else you introduce.
I am not a lawyer, and you should take legal advice on any specific use. The point of this section is to show that HR's AI obligations are already in force rather than waiting on future legislation.
Section 71 of POPIA says a person may not be subject to a decision with legal consequences, or one that affects them to a substantial degree, if it is based solely on automated processing intended to build a profile of them. The section names performance at work explicitly. There are exceptions, including where appropriate measures protect the person's legitimate interests. In practice, the safe design is a human who makes the decision and can explain it.
Section 8 of the Employment Equity Act prohibits psychological testing and similar assessments of employees unless the assessment has been scientifically shown to be valid and reliable, can be applied fairly, and is not biased against any employee or group. Whether a particular AI screening or scoring tool counts as a "similar assessment" is a question for your legal advisers. I would plan on the assumption that it might, and ask vendors for their validation evidence before you sign.
If you recruit or employ people in the EU, the AI Act classes AI used in recruitment, performance evaluation, promotion, task allocation, monitoring, and termination as high-risk. The EU's Digital Omnibus, which entered into force on 27 July 2026, moved the deadline for those obligations from 2 August 2026 to 2 December 2027, according to this analysis from law firm Kinstellar.
Two things did not move. Transparency duties have applied since 2 August 2026, so a recruitment chatbot must tell candidates it is AI. And organisations are still expected to support AI literacy among their staff. The extra 16 months are time to prepare, and the sensible use of that time is to find every AI system already touching your employment process, including features switched on inside software you already pay for.
An AI policy, a set of licences, and a training day change very little on their own. The pattern we see most often is that people attend the session and then Monday looks exactly like Friday.
The evidence points the same way. BCG's 2025 AI at Work survey of more than 10,600 workers found regular AI use was markedly higher among employees who received at least five hours of training and had access to in-person coaching. Getting people there takes training built around their own tasks, followed by support while they change how they work. We have written more about this in our piece on AI training for employees.
There is also a gap most leadership teams do not see. McKinsey found that C-suite leaders estimated 4% of employees used generative AI (tools such as ChatGPT or Copilot that produce text and other content on request) for at least 30% of their daily work, while 13% of employees said they already did. Your people are probably using AI already, often through personal accounts. That is a data protection question for HR, and it is also useful information. The people using it quietly are often your best early champions, once you give them approved tools and clear boundaries.
Finally, measure the right thing. Count how many workflows actually changed and what happened to the time they take. If the only numbers in your AI report are licences and attendance, you will struggle to show value, which is one of the reasons we see AI pilots fail to reach the bottom line.
Every AI conversation in an organisation eventually reaches the question of jobs, and HR is usually the function people ask. Answer it plainly. If the leadership team intends AI to take repetitive work off people so they can do more valuable work, say so and show what that looks like in a specific role. If the leadership team has not decided, say that too, and say when it will.
We are open that headcount reduction is not the conversation we are useful for. Whatever your organisation's position, the worst option is vague reassurance that nobody believes. People adopt tools they trust, and they trust organisations that are straight with them.
This is a realistic first quarter for an HR function starting close to zero.
Days 1 to 30: see what you have. Complete the task inventory. List every AI tool already in use across HR, including AI features inside your HR and recruitment software. Publish a short acceptable-use guideline that says which tools are approved, what personal information must never go into them, and who to ask.
Days 31 to 60: start with documents. Pick two low-risk use cases from the documents group and train the team using their own work as the material. Record how long those tasks took before and after. Our zero-to-one workshops are built for exactly this stage.
Days 61 to 90: decide what comes next. Review what changed. Choose whether an employee-facing policy assistant is worth building. Before anything that touches screening or performance, write down how you will test for bias, who makes the final decision, and what candidates and employees will be told, and have that reviewed by your legal advisers. Then write your HR AI strategy on a single page, covering both jobs.
If you want a quick view of where your organisation stands before you begin, our free AI Readiness Diagnostic takes about five minutes. It scores you across strategy, data and systems, people, governance, and execution, and gives you a report you can take to your executive team. If you would rather talk it through, our page on AI consulting for HR teams explains how we work with HR leaders.
Either way, the first step is the inventory. It costs nothing, and it will tell you more about your AI strategy than any vendor demo.
It should cover two jobs: how HR uses AI in its own work, and how HR leads the people side of adoption across the organisation. In practice that means an inventory of HR tasks, a view of which use cases are low risk and which touch decisions about people, an acceptable-use guideline, a plan for training and support, and measures based on how workflows changed.
Yes, within the law. Section 71 of POPIA limits decisions based solely on automated processing that profiles a person, and section 8 of the Employment Equity Act sets conditions for psychological and similar assessments. The safe design keeps a person responsible for shortlisting and hiring decisions, tests tools for bias, and tells candidates how AI is used. Confirm specific uses with your legal advisers.
My view is that HR should lead the people side of it, including training, communication, and change, and share ownership with the technical lead, who is better placed to make platform, data, and architecture decisions. The managing director should make the final calls.
That depends on the choices your leadership team makes. Our view is that AI should take repetitive, administrative work off HR teams so they have more time for work that needs judgement and care. We do not run engagements aimed at reducing headcount.
Yes, if you recruit or employ people in the EU. AI used in recruitment, performance evaluation, promotion, monitoring, and termination is classed as high-risk, with those obligations now applying from 2 December 2027. Transparency duties, such as telling candidates they are talking to a chatbot, have applied since 2 August 2026.
Start with a task inventory, then pick two low-risk uses such as drafting role profiles or onboarding material, and train the team on their own work. Leave screening and performance tools until you have a bias-testing and human-review process in place.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic +