For a long time, many human resources decisions rested mainly on the experience of whoever was making them. People were hired "on instinct", the most eager-looking person got promoted, and a good employee leaving was explained away with a shrug. Intuition has value, but it does not scale, cannot be audited, and rarely tells luck apart from skill.
People analytics is the practice of applying data analysis to an organisation's people in order to decide in a fairer, faster and more defensible way. It does not replace human judgement — it gives it a factual basis. Instead of "it feels like we are losing people", you can say "we are losing 18% of salespeople in their first year, mostly those who had fewer than three onboarding check-ins".
This guide is for anyone who wants to start without a large investment. We will look at what people analytics is (and is not), what data you already have on hand, which metrics to track first, and which mistakes to avoid. The goal is not to build a data science department, but to make better decisions with what already exists.
What people analytics is, and is not
People analytics — also called HR analytics — is the systematic use of data about employees to answer management questions. Who tends to stay longer? Which factors precede someone leaving? Is training translating into performance? These are old questions; the difference is answering them with evidence rather than anecdote.
It is worth saying what it is not. It is not surveillance: measuring by the minute what each person does destroys trust and rarely improves results. It is not replacing managers with algorithms. And it is not a dashboard full of charts that nobody opens. If a number changes no decision, it is probably not worth tracking.
Why intuition is no longer enough
Intuition works well in stable environments with fast feedback. HR is rarely like that: months can pass between a hiring decision and its outcome, and dozens of factors interfere. It is fertile ground for bias — we hire people who resemble us, we remember recent cases more vividly, and we mistake confidence for competence.
Data does not remove these biases, but it exposes them. When you look at the numbers, you often discover that the "rule" you followed for years has no support. That is how many organisations realised that requiring certain formal criteria was turning away good candidates without improving performance.
The four levels of maturity
It helps to think of people analytics as a four-step ladder, climbed in order:
- Descriptive — what happened? Last year's turnover rate, absenteeism by team, average time to hire.
- Diagnostic — why did it happen? Cross turnover with salary, tenure or direct manager to find patterns.
- Predictive — what is likely to happen? Estimate which employees are most at risk of leaving in the coming months.
- Prescriptive — what should we do? Suggest concrete actions, such as adjusting workloads or reviewing career progressions.
Most teams gain a great deal simply from mastering the first two steps well. Jumping to predictive models without a solid descriptive base is a recipe for wrong conclusions.
Where to start: data you already have
You do not need to buy anything to get going. Most organisations already have the essential data, scattered around: join and leave dates, role, department, salary, training history, appraisal results and attendance records. The first job, almost always, is to bring this together and clean it — not to model it.
Start by choosing a question that matters to someone with decision-making power. "Why do salespeople leave in their first year?" is a better starting point than "let's analyse everything". A concrete question defines the data you need and avoids months spent building reports nobody asked for.
Metrics worth tracking first
There are dozens of possible indicators, but few teams need more than a well-chosen handful:
- Turnover rate, separating voluntary from involuntary departures — mixing them hides what matters.
- Early turnover (departures in the first year), a signal almost always tied to recruitment or onboarding.
- Time to hire and cost per hire, to understand recruitment efficiency.
- Absenteeism by team, useful as an early warning of workload or leadership problems.
- Offer acceptance rate, which reveals whether your proposition is competitive.
Note that none of these requires sophisticated technology. It requires consistency in definition: if everyone calculates turnover differently, the numbers stop being useful for deciding.
Tools: from Excel to Power BI
Do not let the tool set the pace. A lot of useful HR analysis starts in a well-organised spreadsheet. When the questions recur and the sources grow, it makes sense to move to a Business Intelligence tool such as Power BI, which refreshes reports automatically and lets you combine sources without copying and pasting.
If you later need predictive models, languages such as Python or R come into play. But starting there is a common mistake. The sensible order is: first reliable data and clear questions, then visualisation, and only then modelling — once the value is already proven.
Common mistakes, and how to avoid them
The first mistake is measuring what is easy instead of what is important. Counting hours to the minute is simple; understanding why the best people leave is hard — and it is what matters. The second is confusing correlation with cause: teams with more training may perform better because they were already better, not because of the training.
The third mistake is ignoring privacy and ethics. People data demands extra care: anonymise wherever possible, limit access, and be transparent about what is measured and why. A people analytics programme that employees experience as spying is doomed, however good the technique.
Mini case study: a services company
A services company with about 200 employees saw turnover rise with no clear explanation. Instead of launching a new retention programme blindly, it brought together in a single file the join and leave dates, role, direct manager and training history of the previous three years.
Descriptive analysis revealed that almost half of the departures happened in the first nine months and were concentrated in two teams. Cross-referencing the data, it turned out that those teams had had shorter onboarding and less early support. The company extended the onboarding period and created monthly check-ins during the first year.
A year later, early turnover in those teams had fallen from 34% to around 20%. No complex algorithms: it was enough to bring together data that already existed, ask the right question and act on the answer. This is the typical payoff of the first steps in people analytics.
In practice
People analytics does not start with technology, it starts with a good question and data you almost certainly already have. Choose a problem that matters to decision-makers, define the metrics rigorously, present the results simply, and act on them. Climb the steps in order — descriptive before predictive — and treat people data with the ethical care it deserves. Done this way, even a small team starts deciding with evidence, not just intuition.
