Human resource analytics becomes useful when it is pointed at a decision somebody is about to make. A dashboard of metrics nobody acts on is a reporting exercise with better graphics. A short list of questions tied to decisions is worth more, and each question tells you which data has to be clean.
Where is turnover concentrated and what does it cost
Turnover by team, tenure band and manager, with the cost of replacement attached. This is the question most likely to change a decision, because it identifies a specific place to intervene rather than producing an organisation wide figure that nobody owns. It needs accurate leave dates and reasons, which is where most datasets are weakest.
Is pay distributed consistently
Pay by job level, tenure and group, controlled for the factors that legitimately explain difference. This needs a job architecture more than it needs a tool, and organisations without consistent levels cannot answer it however much software they buy. The residual after controls is the finding that matters.
Where is hiring failing
Time to fill by role, offer acceptance rate, and early leaver rate by source. These connect recruitment to retention and are usually answerable from data you already hold. They are also the measures most likely to be improved by a change you can make this quarter.
Questions people ask about human resource analytics
What should we measure first?
Turnover with reasons, and time to fill. Both are actionable and both expose the data quality problems you will have to fix anyway.
Do we need a data scientist?
Not for these questions. Clean data and somebody comfortable with a spreadsheet answers most of them.
How many metrics should a dashboard hold?
Few enough that somebody acts on one. Five with exceptions beats forty that get browsed.