What roles a people analytics function needs, which to hire first, the skills that actually matter, and how team size should scale with organisation headcount.
A people analytics function needs four capabilities: data engineering to make HR data usable, analysis to answer questions with it, translation to turn answers into decisions, and product ownership to build the things that get reused. In a small organisation those four sit in one or two people. They only become four separate roles somewhere north of 5,000 employees.
The most common structural mistake is hiring for the third capability first. Organisations hire a data scientist, discover the HR data is unusable, and spend eighteen months doing data engineering with someone who did not want that job and will leave.
The first person should be a strong analyst who is willing to spend most of their first year on data quality and pipeline work. Not a data scientist. Not a PhD. Someone who is fluent in SQL, comfortable in a spreadsheet, and — critically — genuinely interested in HR as a domain.
What they will actually spend time on:
That last point matters more than it sounds. The first year of a people analytics function is spent earning the right to do the interesting work, and the currency is answering questions quickly and correctly.
Hire for: SQL, data modelling instincts, clear written communication, HR domain curiosity, tolerance for messy foundations.
Do not hire for: machine learning, at this stage.
The second hire should be determined by which constraint bit hardest in year one, and there are only two realistic answers.
If the constraint was data — the systems do not integrate, every question requires a manual export, refreshes are hand-run — hire a data engineer or buy a platform that removes the problem. Do not hire a second analyst into a broken pipeline; you will get two people doing extraction work.
If the constraint was influence — the analysis is good and nothing changes as a result — hire an HR business partner with analytical fluency, or move one into the team. This is the translation role, and it is the most underrated hire in the function. Its job is not to produce analysis; it is to make analysis land with people who make decisions.
Only now does a data scientist or an I/O psychologist make sense, and which one depends on the questions the organisation is asking.
Data scientist if the demand is predictive: flight risk modelling, workforce forecasting, hiring funnel optimisation.
Organisational psychologist if the demand is measurement: survey design, engagement instrument validity, psychometric assessment, psychological safety measurement.
The second is more often the right answer than organisations assume. A great deal of people analytics failure is measurement failure — badly constructed survey instruments and undefined constructs — and no amount of modelling rescues a badly measured variable.
Past roughly six people, the function needs someone who owns the recurring outputs as products rather than as projects: the dashboards, the definitions, the standard reports, the refresh cadence. Without this role, senior analysts spend an increasing share of their time maintaining things they built two years ago, and the function's capacity flattens.
These are rough planning figures, not benchmarks, and they assume a reasonably integrated HR system landscape. Fragmented systems push every band up.
| Organisation size | Typical team | Focus | |---|---|---| | Under 500 | 0–1, often part of an HRBP role | Reliable basic reporting | | 500–2,000 | 1–2 | Reporting plus first diagnostic analysis | | 2,000–10,000 | 3–6 | Diagnostics, dashboards, first predictive work | | 10,000+ | 8–20+ | Specialised: engineering, science, product, embedded partners |
Under 500 employees, a dedicated people analytics hire is usually premature. The right investment is a platform that produces reliable metrics without a person, plus an HRBP who is comfortable interpreting them.
Three options, each with a predictable failure mode.
Into HR (most common). Closest to the domain and the data. Failure mode: the function becomes an HR reporting service and never gets business questions.
Into central analytics or data. Better engineering standards and tooling. Failure mode: technically strong work on questions HR did not ask, and a slow erosion of domain understanding.
Hybrid — reporting to HR, dotted line to central data. Most of the benefit of both. Failure mode: two sets of priorities and a manager negotiation every quarter.
The hybrid is generally right for organisations above a few thousand people. Below that, report into HR and borrow engineering support.
Rank-ordered by how often their absence causes failure, which is not the order job adverts use.
1. Definitional discipline. The ability to define a metric precisely, write it down, and hold the line when a stakeholder wants it calculated differently for their deck. Most people analytics arguments are definitional, not analytical.
2. Written communication. Analysis that cannot be explained in a page does not get used. This is the single strongest differentiator between analysts whose work changes decisions and analysts whose work gets filed.
3. SQL and data modelling. Non-negotiable and consistently underweighted relative to statistics.
4. HR domain knowledge. Knowing that a "manager" in the HRIS is not the same as a people manager, that headcount and FTE diverge, and that a Q1 turnover spike might be a vesting cliff rather than a culture problem.
5. Statistical literacy. Enough to know when a difference is noise, when a sample is too small, and why survey participation below 60% makes the results unusable. Deep modelling ability matters far later than most job specs imply.
6. Ethical judgement. Knowing which analyses should not be run. This is a hiring criterion, not a training topic.
A new function has one job in its first quarter: establish that its numbers are correct. Nothing else matters until that is true, because a function whose headcount figure disagrees with finance will spend two years litigating credibility.
Weeks 1–4. Inventory the data. Reconcile headcount against finance and payroll. Document every discrepancy and its cause.
Weeks 5–8. Publish a metric dictionary: definitions and formulas for the core measures, agreed with HR leadership and finance. Boring, and the single highest-return artefact the function will produce.
Weeks 9–12. Ship one reliable recurring report — usually turnover segmented by function, tenure and manager — and one piece of analysis that answers a question a business leader actually asked.
Resist the pressure to build a predictive model in the first year. A retention model built on data nobody trusts produces a number nobody acts on, and the failure gets attributed to analytics rather than to the foundations.
The function is working when business leaders bring it questions unprompted, when the metric definitions stop being argued about, and when at least one decision per quarter can be traced to something the team produced.
It is not working when the main output is a monthly pack that nobody has questions about. Silence is not approval; it usually means the pack is not being read.
What is the first people analytics hire? A strong analyst with SQL fluency and genuine HR domain interest, who accepts that most of year one is data quality and definition work. Hiring a data scientist first is the most common and most expensive sequencing error, because the interesting modelling work is not possible until the foundations exist.
How big should a people analytics team be? Roughly one to two people at 500–2,000 employees, three to six at 2,000–10,000, and specialised sub-teams above that. Fragmented HR systems push every band higher, because more capacity goes to extraction and reconciliation.
Should people analytics report to HR or to the data team? Into HR for domain proximity, with a dotted line to central data for engineering standards, is the most reliable arrangement above a few thousand employees. Reporting purely into a central data team tends to produce technically strong answers to questions HR did not ask.
Do we need a data scientist for people analytics? Not initially, and many organisations never do. The highest-value early work is reliable measurement and clear diagnosis, both of which are analyst work. A data scientist becomes worthwhile once the data foundation is trustworthy and there is real demand for prediction.
Reduce how much of this you have to build. PeoplePilot Analytics handles the integration, definitions and dashboarding layer, so a small team can spend its time on interpretation instead of extraction. Start a free trial, or use the HR Metrics Library as a ready-made metric dictionary.