By Johannes Sundlo, AI & Future of Work Advisor. I help leaders and HR teams turn AI into adoption that sticks, through keynotes, workshops and change programs.
People analytics has promised a lot for years: better decisions about hiring, retention and development, grounded in evidence rather than gut feel. The hard part was rarely the ambition. It was the work of pulling scattered HR data together, cleaning it and turning it into something a manager could use. This is where AI for people analytics starts to earn its place, not by replacing the analyst, but by removing much of the friction that kept good questions unanswered.
What AI changes in people analytics
HR data is usually messy. It lives in an HRIS, a payroll system, a survey tool, an applicant tracking system and a handful of spreadsheets that someone maintains by hand. Different names for the same job title. Inconsistent date formats. Free-text fields where a manager wrote whatever felt right at the time. Historically, most of the effort went into reconciling all of this before any analysis could begin.
AI helps in three practical ways:
- Turning messy data into something usable. Language models are good at reading inconsistent text, matching records that refer to the same thing and flagging gaps. They can map free-text reasons for leaving into a consistent set of categories, or summarise hundreds of open survey comments into themes you can read in a few minutes.
- Surfacing drivers rather than just numbers. A traditional dashboard tells you attrition rose to nine percent. A model can help you see which factors move alongside that rise: tenure bands, specific managers, commute distance, time since last promotion. It does not prove cause, but it points you toward the questions worth asking.
- Making analysis conversational. Instead of waiting for a report, an HR business partner can ask a plain question and get a draft answer with the underlying figures attached. That lowers the barrier for people who are not analysts but still need evidence.
None of this removes the need for judgement. It shifts where your time goes, from preparing data to interpreting it.
Realistic use cases
It helps to be specific about what is genuinely useful today, rather than what sounds impressive in a vendor deck.
Understanding attrition and engagement
You can feed historical data into a model and ask which patterns are associated with people leaving. The output is a set of signals, not a verdict. Maybe regrettable attrition clusters in one function, or among people who have not had a development conversation in over a year. That gives you a place to investigate and a hypothesis to test with managers who know the context.
The same approach works for engagement survey data. AI can group thousands of comments into recurring themes, show how sentiment differs across teams, and highlight where positive scores hide a vocal minority. You still read the raw comments. The model just helps you decide which ones to read first.
Scenario-based workforce planning
Workforce planning is where this becomes interesting for leaders. Rather than a single static headcount forecast, you can model several futures. What happens to capacity if attrition stays flat versus rising two points? How does a hiring freeze in one region ripple through delivery? What skills will be short if a product line grows as planned?
AI does not predict the future here. It helps you build and compare scenarios quickly, so the conversation with finance and the business is grounded in numbers rather than assertions. The value is in the speed of iteration, letting you ask “what if” ten times instead of once.
Closing skills gaps
Many organisations have no reliable picture of the skills they hold. AI can read job descriptions, project records and learning data to draft a skills inventory, then compare it against what upcoming work will require. Treat the first draft as a starting point to correct, not a finished map.
The cautions that matter
This is the part that deserves more attention than it usually gets. People analytics deals with information about real people, and the cost of getting it wrong is higher than in most other applications of AI.
- Data quality decides everything. A model trained on biased or incomplete history will reproduce that bias with a confident tone. If your past promotion data favoured one group, a model that learns from it will quietly recommend more of the same. Clean inputs and honest labelling are not preparation for the work. They are the work.
- Privacy is not optional. Employee data is personal data. Under GDPR and similar regimes, you need a lawful basis, clear purpose limitation and transparency about what you analyse and why. Aggregate where you can. Avoid building individual risk scores that managers might act on without context.
- Correlation is not cause. A model showing that people who skip a benefit tend to leave does not mean the benefit retains them. Acting on correlation as if it were cause is one of the easier ways to waste money and trust.
- Keep a human in the loop. Decisions that affect someone’s career, pay or job should never rest on a model’s output alone. Use AI to inform the conversation, not to make the call.
If you want a wider view of where the field stands and how peers are handling these questions, the state of AI in HR is a useful reference point.
How to start small
The teams that get value from this rarely begin with a platform purchase. They begin with one question that matters and data they already trust.
- Pick a single, concrete question. For example: why are people in their first year leaving more than expected?
- Use data you can vouch for, even if it is narrow. A clean slice beats a messy ocean.
- Run the analysis, then validate the findings with managers who know the reality on the ground. Their reaction tells you whether the signal is real.
- Document what you did, including the data sources and the limits of the conclusion. This builds the trust you will need for the next, bigger question.
- Only then consider scaling, with proper governance and privacy review in place.
Starting small also protects you. A narrow first project surfaces your data problems and your governance gaps while the stakes are low, which is far better than discovering them in a board-level workforce plan.
People analytics with AI is less about predicting people and more about asking better questions, faster, with the evidence to back them up. The judgement stays human. The grunt work does not have to. If you are building out broader capability, it sits naturally alongside the rest of your AI for HR efforts rather than as a separate project.
If you want a calm, practical way to bring evidence into your workforce decisions without overreaching, here is how I help teams →