HR analytics is the practice of answering questions about a workforce from the records you already keep. Most of the useful answers are descriptive and unglamorous: how many people left, from which teams, after how long, and what it cost to replace them. The predictive claims that sell the category need far more history and far more leavers than most employers have, and using them anyway is how a confident number ends up steering a decision it cannot support.
Start with four questions you will act on
Headcount by team over time, turnover by tenure, absence by reason, and time to fill a vacancy. Each is answerable from an ordinary record, each has an owner who can act on it, and each exposes whether your data is good enough before anybody promises a model.
History is the binding constraint
Answering what changed requires a record that keeps dated changes rather than current values. If your system overwrites, your analytics begin the day you start snapshotting, and no tool can reconstruct the years before that. This is the single most consequential thing to check.
Where predictive claims get thin
A model that flags who might leave learns from people who did leave. An employer with a few dozen leavers a year does not have the sample to support a model that generalises, and a vendor benchmark trained on other companies is not a model of yours. Ask what it was validated against and what a manager is supposed to do with the output.
Definitions before dashboards
Headcount can mean three things in one meeting, and turnover can be calculated four ways. Write the definitions down, agree them with finance, and put them next to the number. Analytics that nobody trusts are worse than no analytics, and mistrust usually starts with a definition.
Questions people ask about hr analytics
What can HR analytics actually tell us?
Descriptive things, reliably: headcount and movement, turnover by tenure and team, absence patterns, time to hire and cost per hire. Those answer most of the questions employers have.
Do we need a separate analytics tool?
Usually not. If your HR system keeps dated history and exports cleanly, a spreadsheet or your existing reporting tool answers the first questions. Buy a tool when the questions outgrow the export, not before.
Are predictive attrition models reliable?
Rarely at a single employer's scale. They need many leavers and several years of consistent data. Treat them as a prompt for a conversation rather than as evidence, and never as an input to a decision about an individual.