How to get hired as a people analytics analyst in 2026-27

What the job is, and the titles that mean the same thing

A people analytics analyst is the analyst for the workforce itself. The subject matter is employees: how many there are, where they sit, who is leaving and when, who got hired and how long it took, what everyone is paid relative to a band, what the survey said, who got promoted and who did not. The datasets are smaller than a product analyst's and messier than a finance analyst's, the stakeholders are more political than either, and the consequences of a wrong number land on an individual person rather than on a dashboard.

The recurring deliverables are predictable, and knowing them is itself an interview advantage because most candidates talk about tools instead. There is a monthly or quarterly headcount and attrition pack that goes to the executive team. There is a hiring funnel view that talent acquisition lives in. There is an annual workforce plan built with Finance, which is where the role either earns credibility or loses it. There is a compensation cycle, which means benchmarking, band analysis, budget modeling and often a pay equity review. There is an engagement survey, which means driver analysis, reporting thresholds and a manager action format. And there is a steady stream of one-off questions from a CHRO or a business leader, usually phrased as a conclusion looking for support.

The honest version of the first year, which postings never say: a large part of it is recurring reporting and request handling. Someone needs the headcount file on the first working day of the month, the board pack needs refreshing, a leader wants the same cut sliced differently, and the survey vendor's export does not match the HCM. Candidates who expect pure analysis are the ones who leave. The way the job gets better is by turning each repeated request into a definition, a scheduled job or a self-serve view, and a candidate who says that in the interview sounds like someone who has done it.

Three adjacent roles get confused with this one, and the distinction matters when you are choosing what to apply for. An HRIS analyst owns the system of record itself: configuration, security roles, integrations, data integrity and the report writer. It is the most common feeder role into people analytics and it usually pays less, so if you are in it, the move is real and worth making. A compensation analyst owns bands, survey submissions, benchmarking and the merit cycle; it is a specialist track with its own credential path (WorldatWork's CCP) and it overlaps with people analytics on pay equity. A workforce planner owns capacity and forecast, often sits in Finance or in operations rather than HR, and in contact centers and hospitals it is a distinct discipline with its own tooling.

Where the team sits changes the job more than the title does. In HR, under a head of people analytics or a VP of people operations, you are close to the stakeholders and far from the data platform, so you will build more of your own pipeline. In a central data organization with a dotted line to HR, you inherit the warehouse, the standards and the review process, and you fight a different battle to get HR data treated as restricted rather than as another source. In Finance, which happens in headcount-heavy businesses, the headcount number is the product and everything else is secondary. Ask in the interview. It tells you what your first year looks like.

There is a second environment question worth asking, because it decides what skill you will actually use. Does the HR data leave the HCM at all? At plenty of mid-size employers it does not: there is no warehouse, there is no dbt project, and the whole job runs on Workday Report Writer, calculated fields, a Prism dataset if they bought it, an export and Power BI. A candidate who only prepared SQL arrives over-equipped for the test and under-equipped for the work. A candidate who can say which parts of their method survive inside a report writer and which need an extract is answering a question the hiring manager has had to explain to every previous applicant.

Team size is the fact that shapes everything else. One to five people is typical, and a single person covering the whole function is common. The practical consequences: breadth is required, you will build the pipeline and write the memo and present it; the hiring manager is often the entire existing team, so you are being assessed as the person who will sit next to them for forty hours a week; and the answer "I would hand that to data engineering" is usually wrong, because there is no one to hand it to.

The same job is posted under many titles, and searching only for the phrase people analytics will hide most of the market from you. Set alerts on all of these.

• People Analytics Analyst, People Insights Analyst, Talent Analytics Analyst.
• HR Data Analyst, HR Analyst, HR Reporting Analyst, HR Insights Analyst.
• Workforce Analytics Analyst, Workforce Intelligence Analyst, Workforce Planning Analyst.
• HR Business Intelligence Analyst, Human Capital Analyst, Human Capital Data Analyst.
• At the senior end: People Analytics Partner, People Analytics Consultant, People Data Scientist, Manager or Director of People Analytics.

The two doors in: from HR, and from data

Almost everyone in this field arrived through one of two doors, and the interview is designed to find out which one you came through and whether you closed the gap it leaves. Know which is yours and build the evidence for the missing half, because the thing you are already good at will not be what gets tested hardest.

From HR. Your advantage is domain, and it is a genuine advantage that data candidates cannot fake. You know what a requisition is, what a comp cycle does to a quarter, why a transfer is not a hire, what a works council is, and what happens in the room when a leader disputes a number. Your gap is technical and it is specific. It is not "learn data". It is: SQL beyond a simple select, including window functions and joins on validity ranges; the ability to reconstruct the state of the workforce on an arbitrary past date out of effective-dated records; a real grasp of denominators; and enough statistics to know when to refuse to claim something. Excel skill is not the proof. Many HR candidates lead with advanced Excel and it reads as the ceiling rather than the floor.

The fastest way to close that gap costs you nothing and happens inside your current job. Volunteer to own the headcount report and reconcile it to Finance line by line. Rebuild the attrition metric and write the definition down, including how you treat transfers, rehires, interns, contingent workers and acquisitions. Take the engagement survey analysis back from the vendor and do the driver analysis yourself. Build the hiring funnel conversion view from the ATS rather than accepting the vendor's dashboard. Each of those is portfolio-grade experience you can describe in an interview without a single confidential number attached.

From data or BI. Your advantage is technique, and it is also real. Your gaps are three, and they are not the ones you expect. First, the HR data model: records are effective-dated, a worker and a position are different objects, the org hierarchy moves under you, people have leaves and rehires and sometimes two jobs at once, contingent workers may or may not be in the same system, and the row you see today is not the row that was true in March. A data candidate who assumes a clean employee table is immediately visible. Second, the calendar: the work is seasonal, built around the performance cycle, the comp cycle, the survey and the annual plan, and if you do not know the calendar you cannot plan your year or sound credible about it. Third, the confidentiality instinct, which is the one that cannot be taught quickly and which the interview tests hardest.

The attitude failure that sinks data-path candidates is talking about HR as the shallow end of analytics. It is not. The samples are small enough that most differences are noise, the definitions are contested rather than given, you usually cannot run an experiment on people, the data is entered by humans under time pressure, and the stakeholder who disagrees with your finding outranks everyone in the room. Those are harder conditions than a product funnel with ten million events, and saying so in an interview is one of the quickest ways to be taken seriously.

Put the missing half at the top of your resume, not the half you already have. An HR-path candidate should open with the technical evidence and the portfolio link, then the HR history as the credibility layer underneath. A data-path candidate should open with the HR domain evidence, the comp cycle or the survey or the headcount governance work, and let the technical stack sit below it where nobody doubts it anyway.

One structural fact to plan around: a large share of these seats never reach a job board. The team is small, the hiring manager already knows who in HR has been quietly doing the analysis, and the role gets filled by transfer. If you are outside, that means two things. Apply where teams are being built rather than backfilled, which is most often at companies that just crossed a few thousand employees or just hired their first head of people analytics. And apply to hourly-workforce employers, where the turnover problem is large and already priced, the analytics function is young, and the candidate pool is far thinner than it is in technology.

The quickest self-check before you apply is whether you can produce the proof points for your own door, and whether you can name the one for the other door that you are still missing.

• HR-path proof points, in order of weight: a SQL exercise you can walk through, a headcount reconciliation you ran, a metric definition you authored, a survey analysis you did yourself rather than received.
• Data-path proof points, in order of weight: you can explain effective dating without prompting, you know what a comp cycle involves month by month, you can state a reporting threshold policy and why it exists, you have handled restricted data before.
• Both: one analysis that changed a decision, stated as the decision first.

How these jobs are hired: who screens and what each stage tests

A people analytics posting draws a strange applicant pool: HR people who want out of transactional work and data people who want a smaller, more human problem. That means volume, and it means the screen is looking for a specific signal rather than a general one. The two things that move you through it fastest are a referral into a small team and a portfolio link that a technical hiring manager can open and judge in under two minutes.

The recruiter or HR operations screen comes first, often run by the same talent acquisition team the role will serve, which has an awkward consequence: they will read your resume as a sample of your work on structured documents. They screen on literal nouns, mostly system names. If you have used Workday, the word Workday needs to be on the page. Expect fifteen to thirty minutes on scope, systems, why this role, and compensation expectations.

The hiring manager interview is the real first round. This is the head or director of people analytics, and sometimes it is a VP of people operations with no analytics background, which changes the conversation entirely. With a technical manager, expect to be asked how you would calculate something and then challenged on the definition. With a non-technical manager, expect to be assessed on whether you can explain a method to them without condescension, because that is the job they are hiring for.

The technical exercise takes one of two shapes. A live SQL screen on a shared editor, usually forty-five minutes, with tables that look like an HR system: a worker table, a job history table with effective start and end dates, a position or org table, sometimes an ATS application table. Or a timeboxed take-home, usually three to five hours of actual work, with a small dataset and a vague question. Both are testing the same thing, which is whether you reconstruct state correctly in time and whether you notice that the obvious join gives a confidently wrong answer.

On the take-home, the single highest-leverage decision is format. Most candidates send a notebook with twenty cells and a conclusion at the bottom. Send a one-page or two-page memo: the answer and the recommended decision in the first paragraph, the two or three pieces of evidence behind it, an explicit section listing what you could not conclude and what data would have let you, and the notebook or the repo as the appendix. Converting analysis into a decision for someone with ninety seconds is the scarce skill in this field, and the take-home is the only place you can demonstrate it unprompted.

The case presentation is usually the final and deciding stage. You present to a panel: HR business partners, total rewards, talent acquisition, sometimes a Finance partner or someone from the central data team. The brief is normally either your take-home or a scenario such as attrition rising in one function. It is ten to fifteen minutes with interruptions, and it is built so that someone will challenge your first claim before you reach slide three. The failure mode is giving a tour of the analysis in the order you did it. Lead with the answer, hold the method in reserve for the question that will come, and know in advance which number in your deck is weakest, because someone in the room will find it.

Expect a confidentiality question in at least one round, usually phrased as a scenario rather than a policy question. It is not a trick. It is the thing the team cannot risk being wrong about, because a single leak ends the function's ability to run a survey or hold comp data.

Timeline is typically three to six weeks, and small teams move slowly for an unglamorous reason: the hiring manager is also doing the whole job, so scheduling drags. A polite check-in at two weeks is normal and does not count against you.

Two routes run differently. Consulting (Mercer, Deloitte Human Capital, Aon, WTW, Korn Ferry, and the people and organization practices at the strategy firms) runs a consulting loop: a structured case, estimation, and a strong emphasis on client presence, with less depth on SQL. Vendors (Visier, Workday, One Model, Culture Amp, ChartHop, Crunchr, OrgVue) hire analysts into customer-facing and product-adjacent roles, and the test is whether you can explain a workforce metric to a buyer who is skeptical, plus a product sense round. Government, healthcare systems and large hourly employers run more structured, slower processes, sometimes with a formal scoring rubric, and often with a hard requirement for a named HRIS.

The loop in summary, stage by stage, so you can work out which one you are actually in when a recruiter calls it a chat.

• Recruiter or HR ops screen: scope, systems by name, compensation expectations. Fifteen to thirty minutes.
• Hiring manager: how you define a metric, how you handle a disputed number, what you have owned end to end.
• Technical: live SQL on effective-dated tables, or a three to five hour take-home with a deliberately vague question.
• Case presentation to a cross-functional panel: the deciding stage, scored on the decision rather than the analysis.
• Stakeholder or executive round: can you be trusted in front of a CHRO, and will you say no when you need to.
• Reference and background checks: standard, and in this field references are often called informally within a small community, so assume your reputation precedes you.

The portfolio: real analysis without touching employee data

This is the problem specific to the role. Every analytics field says build a portfolio. In people analytics the data you work with is the one category you can never show, and the standard advice to anonymize it is wrong twice over.

The rule is absolute: never use your employer's data. Not aggregated, not anonymized, not "it was only headcount", not a screenshot with the numbers blurred, not a dashboard image left in a resume PDF. The first reason is legal and contractual. Your employment agreement, and the data protection agreements your employer signed covering workforce data, make copying it out a terminable act, and in some jurisdictions a reportable breach. The second reason is the one that actually costs you jobs. The person watching your demo is imagining their own employees' salaries, performance ratings and exit reasons in your hands, and you have just shown them what you do with an employer's data when you leave. Candidates lose offers for this and are almost never told why.

Anonymization does not save you either, and knowing why is itself a signal worth showing. Workforce data is small and highly dimensional. Department, level, tenure band, location and gender together identify most people in most organizations, which is exactly why reporting thresholds exist. If you can explain re-identification risk in a sentence, you have answered one of the standard interview questions before it is asked.

There are three legitimate sources, and a strong portfolio uses more than one.

First, public data that is genuinely real. OPM FedScope publishes the US federal civilian workforce in cube form: employment by agency, occupation, grade, pay plan, length of service and age band, plus separations and accessions over time. It is the closest public thing to an HCM extract that exists, it is real rather than synthetic, and very few candidates in this pool use it, which makes it a differentiator on its own. City and state payroll files on open data portals (New York City citywide payroll, San Francisco employee compensation, many state comptrollers and university systems) give real individual pay with title and department, which is what you need for a pay analysis. BLS JOLTS gives hires, quits and layoffs by industry as a market baseline, which is how you answer "is our attrition bad or is the market bad". ACS PUMS and CPS microdata give pay, occupation, hours and education at person level for labor market questions. O*NET gives the task and skill structure of occupations, which matters for anything about AI and job design. Indeed Hiring Lab and similar published series give postings trends.

Second, simulated data with a published generator. Write the generator yourself: a worker table, a job history table with effective-dated rows including promotions, transfers, leaves and terminations, a position and org hierarchy that changes over time, an application funnel, and a compensation table with bands. Publish the generator and its assumptions alongside the analysis. This is the most underrated artifact in the field, because building a realistic HR data model proves the exact thing the technical screen is trying to test, and it proves it more convincingly than passing the screen does.

Third, the method write-up with no data at all. Describe an analysis you ran at work as a method: the question, how you defined the terms, the design, the pitfalls you handled, who the audience was, what changed as a result, and nothing numeric or identifying. Then say out loud in the interview that you deliberately kept the numbers out. Interviewers notice, and it converts a confidentiality constraint into a point in your favor.

A word on the dataset everybody uses. The IBM HR Analytics Employee Attrition set on Kaggle is fictional data created by IBM staff. A large share of the tutorials in this field use it. Running a logistic regression on it and reporting which factors drive attrition is a mildly negative signal, because the relationships inside it are whatever the generator put there, and presenting them as findings shows you did not ask where the data came from. If you use it, use it to demonstrate mechanics, and say in one line that it is synthetic and what that means for the conclusions.

Four portfolio pieces land harder than anything else, because each one is the actual work of the job rather than a demonstration of a technique. Build at least two.

One: point-in-time headcount and attrition from effective-dated records. Build a date spine, build a slowly changing dimension from job history rows, produce headcount on any arbitrary date, then produce annualized attrition using average headcount over the period as the denominator rather than start or end headcount, split into voluntary and involuntary. Publish the definitions page alongside it: how you treated transfers, rehires, interns, contingent workers, employees on unpaid leave, people holding two positions, and an acquisition. This is literally the first ninety days of the job.

Two: a retention curve rather than an attrition rate. Fit Kaplan-Meier survival curves by hire cohort, handle censoring correctly and say that you did, and show the first-year cliff that an annual attrition percentage hides. The point you make in the write-up is the valuable part: the same annual rate means something completely different if the leaving is concentrated in month seven than if it is flat across tenure, and the intervention is different too.

Three: a pay analysis on public payroll data. Compute the unadjusted gap, then the adjusted gap from a regression adding title, department, tenure and full-time status, and then write the paragraph most candidates cannot write: what the adjusted gap does and does not prove, why controlling for job level can conceal a problem if people were leveled unfairly in the first place, and why the unadjusted gap is a representation story while the adjusted gap is an exposure story. That paragraph is worth more than the model.

Four: a workforce plan. Take a published headcount series, model attrition forward, add a time-to-fill lag so that a hire approved in March arrives in June, and produce a monthly hiring requirement with its cost. Include a sensitivity table across attrition and time-to-fill assumptions. This is the artifact that makes a Finance partner take you seriously, and it is rarely in a candidate's portfolio.

Present all of it in a form that can be judged quickly. A README that states the question, the data source, the definitions, the method, the finding and the limits, in that order. Charts that survive being printed in black and white. Code in a repo behind the write-up rather than in front of it. Not a public dashboard with fourteen filters and no conclusion, which tests nothing except that you can drag fields onto a canvas.

The public sources worth knowing by name, because naming them in an interview is itself evidence that you have thought about where workforce data can legitimately come from.

• OPM FedScope: real federal workforce employment, accessions and separations by agency, occupation, grade, pay plan, length of service and age band.
• City, state and university payroll files on open data portals: real individual pay with job title and department.
• BLS JOLTS: hires, quits, layoffs and openings by industry, for market context behind any attrition number.
• ACS PUMS and CPS microdata: person-level pay, occupation, hours and education.
• O*NET: task and skill statements by occupation, for job design and AI task-exposure work.
• Your own generator: effective-dated worker history, org hierarchy, funnel and comp bands, published with its assumptions.
• Check each source's terms of use before you publish anything built on it, because a portfolio that breaks them raises the exact judgment question you are trying to answer.

The resume: what a people analytics lead reads, and what gets skipped

Two different people read it. A recruiter scanning for literal system names, and a hiring manager who may be the entire existing team and who is reading it as a sample of your work on structured information. Both facts point the same way: single column, plain headings, real nouns, no skill-level bar charts, no headshot, no two-column template that scrambles when it is parsed. One page under eight years of experience, two above, never three. Put the portfolio link in the header, because in this field a technical hiring manager will actually open it.

Lead with scope, in one line, before any bullet. The shape that works names the population, the geography, the systems and the stakeholders: people analytics for a 4,200-person workforce across fourteen countries on Workday, Greenhouse and Culture Amp, reporting into the CHRO with a standing line to FP&A. A hiring manager is mapping you onto their own environment and they need the shape immediately. Someone who has only ever worked on a two-hundred-person company with BambooHR and someone who has handled multi-entity international headcount are doing different jobs, and pretending otherwise wastes everybody's first call.

The systems line matters more here than in general analytics roles, and it has an order. Name the HCM first, because that is what the recruiter screens on and because it signals the scale and messiness you have handled: Workday, SAP SuccessFactors, Oracle HCM, UKG, Dayforce, ADP, BambooHR. Then the ATS, the survey platform, the compensation tooling if any, then the warehouse and the BI tool, then Python or R if you genuinely use them. If you have used a packaged people analytics product such as Visier, One Model, Crunchr or ChartHop, name it, because the employer either has one or is evaluating one.

What gets skipped, in roughly the order it gets skipped. Certifications stacked at the top of the page, which signal that the credential is the strongest thing you have. The phrase data-driven HR professional, which is common enough on these resumes to carry no information at all. Dashboard counts, because the number of dashboards you built is a measure of unmanaged demand rather than of impact. Excel as a headline skill. A list of chart types. Generic diversity and inclusion language with no metric attached, which reads as a candidate who sat near the work rather than doing it. Storytelling with data, unless there is a decision attached to the story.

For career changers, put a Projects section above Experience, with three entries at most, each one line plus a link, each line stating the finding rather than the technique. The finding is what makes someone click. For an internal mover, put the analytics work above the HR duties even if it was a fifth of your time, and be specific about which part you built rather than received.

One detail that is easy to miss and costs interviews: numbers on your resume will be checked against each other. If one bullet says a 4,000-person company and another implies 12,000 hires a year, a person who reconciles numbers for a living will notice, and they will notice in the one interview where that habit is a professional qualification.

Write every bullet decision first, analysis second, system third, and cut any bullet that does not end in a decision or a number you can defend. The shapes that work, written here as patterns to fill with your own true numbers rather than as claims to copy:

• Rebuilt headcount and attrition definitions across Workday and the FP&A model, reconciled the two to an agreed tolerance, and published a definitions page that ended a recurring monthly dispute between HR and Finance.
• Ran the annual pay equity review with outside counsel, controlling for level, function and location, produced the remediation list and its budget, and wrote the manager talking points that went with the adjustments.
• Found that the hiring funnel drop was between recruiter screen and hiring manager screen rather than at offer, which moved where the talent acquisition team spent the quarter and shortened time to hire for the affected function.
• Replaced the vendor's engagement read-out with a team-level driver analysis at a minimum reporting threshold of five, and cut the output from a forty-slide deck to a two-page manager action sheet that managers actually used.
• Built the quarterly workforce plan, including attrition forecast, time-to-fill lag and cost, which became the headcount input to the annual budget.
• Automated the monthly executive pack from Snowflake into Tableau, removing several days of manual work per month and, more importantly, removing the three places the manual version used to disagree with itself.

The interview, question by question

The questions in this loop are unusually consistent across employers, because the failure modes are consistent. Prepare these specific answers rather than general ones.

"How do you calculate attrition?" The wrong answer is a formula. The right answer starts by asking what decision the number is for, then defines the numerator and denominator explicitly: leavers over average headcount across the period, not start headcount and not end headcount, because a growing or shrinking population distorts both. Annualize correctly. Separate voluntary from involuntary, and define regretted with the business rather than assuming it. Then state the inclusion rules out loud: transfers are not leavers, internal promotions are not hires, rehires need a rule, interns and contingent workers need a rule, and an acquisition needs its own treatment or it will swamp the series. Finish by saying the number is meaningless without the definition published next to it. That last sentence is the one that gets you through.

"Headcount does not match Finance. Walk me through finding the difference." The list of usual suspects: the as-of date and the timezone, effective-dated versus payroll-dated records, open positions being counted as headcount, contingent workers, employees on unpaid leave, FTE versus heads, legal entity and intercompany transfer handling, people holding two positions, and terminations entered into the system late. Then the method: reconcile line by line and publish a bridge that both sides sign off. The emotional content of the answer matters as much as the content. Someone who expects this work and treats it as routine is hireable. Someone who frames it as Finance being wrong is not.

The SQL screen. Practice on your own simulated tables until this is automatic: build a date spine, join job history rows on validity ranges to produce a point-in-time snapshot, compute headcount on any date, compute rolling twelve-month attrition with the correct denominator, build a cohort retention table, and compute funnel conversion by stage with denominators that do not double count a candidate who was reactivated. The standard trap is a join to the current worker record, which returns a confident number that is right only for today. Say what you are doing as you do it.

"Engineering attrition is up four points. Is that real?" The expected answer is a sequence, not a yes or no. How many people is four points, because in a hundred-person function that is four humans and well inside noise. Did the composition change, because adding a large junior cohort raises attrition mechanically. Is one team or one reorganization responsible. Is the comparison period seasonal. What does the market series say. And then a confidence statement rather than a verdict. Candidates who answer with a cause lose this question.

"How would you measure quality of hire?" This one is asked constantly and answered badly, usually with a vendor's composite index. The strong answer starts by refusing the single number and asking what decision it serves, then names the components you would measure separately and where each one comes from: early attrition at six and twelve months from the HCM, performance or ramp against a role-specific milestone that the business already tracks, hiring manager satisfaction collected at a fixed interval rather than ad hoc, and internal mobility later on. Then name the problems honestly: the measure arrives a year after the hiring decision it is supposed to inform, performance ratings are a noisy and politically loaded input, and the sample per recruiter or per source is usually too small to rank anybody. Offer what is actually usable: source and channel comparisons pooled over a long enough window, and early attrition as the one component that is clean and timely.

The Simpson's paradox question, usually disguised. A pay gap appears in the aggregate and vanishes inside every job level, or a promotion rate looks equal overall and is unequal in every department. Name the effect, explain that the aggregate is being driven by the distribution across groups rather than by the within-group rates, and say which direction you would investigate: if the gap lives in the distribution across levels, the question is how people got leveled, which is a bigger finding than a pay adjustment.

"A VP wants the engagement scores for a team of four." The answer is no, and the way you say no is the test. Give the reason, which is re-identification and the promise made to respondents at the moment they answered. State the policy, which is commonly a minimum of five respondents. Then give them a path to what they actually need: the roll-up at the next level, qualitative themes with a volume floor, a comparison against the function, or a direct conversation with the team facilitated by someone neutral. A candidate who only refuses looks rigid. A candidate who only complies is disqualifying.

"We have a twelve percent gender pay gap. What do you do next?" Distinguish the two numbers immediately: the unadjusted gap, which is mostly a story about who holds which jobs at which levels, and the adjusted gap after controlling for legitimate factors, which is the legal exposure story. Say that in the US this analysis is routinely run at the direction of counsel so that preliminary results are privileged, and that you do not circulate a draft gap number by email. Then say what each number implies for action, because an unadjusted gap is fixed by hiring, promotion and leveling over years, and an adjusted outlier is fixed by a pay adjustment this cycle.

Adverse impact. Know the four-fifths rule from the Uniform Guidelines on Employee Selection Procedures as the common screening heuristic, know that it is a rule of thumb rather than a legal safe harbor, and know that you may be the person asked to run that analysis on a vendor's screening tool. If you can also say what you would do when the sample in a subgroup is too small to say anything, you are ahead of most candidates.

Causality. Expect a question engineered to see whether you will claim that a program caused an outcome. Good answers name the design that would be required: a staggered rollout across locations, a comparison group, a difference-in-differences, a metric committed to before the rollout, or an honest statement that only an association is available and here is what would change that. Saying "we cannot conclude that, and here is the cheapest study that could" is a strong answer, not a weak one.

Behavioral: a time you told a senior stakeholder their number or their belief was wrong. Prepare one with the actual sentence you used and what happened afterwards, including whether they accepted it. This role exists to be the one source of a contested number, and a candidate with no story here reads as someone who has never been in the room.

The case presentation. Open with the answer and the recommended decision. Two or three pieces of evidence. An explicit limits slide. Appendix for the method. Build it to survive an interruption on slide two, and know which number is weakest before they find it. If the case involves individual employee data, say how you would restrict access to it, unprompted. That single unprompted sentence has decided more of these loops than any chart in the deck.

Have your own questions ready, and make them the ones only an insider would ask. Where does the team sit and who sets its priorities. Does HR data leave the HCM, and into what. Who can see compensation data today. What is the reporting threshold on the survey. What does the executive team currently believe that the data does not support. The answers tell you whether the job is the one being advertised, and asking them is itself a work sample.

Confidentiality, privilege and the law you are expected to know

Workforce data is usually classified at the same level as pre-close financials, and the person holding it is you. Knowing how that is handled operationally separates a candidate who has done the job from one who has read about it.

In practice it means least-privilege access in the warehouse, a separate schema or project for HR data, row-level security so an HR business partner sees only their population, named approvers for compensation and performance data, no exports to personal drives or personal accounts, and a documented rule for what happens when a senior leader asks for something they are not entitled to. Have an answer ready for that last one, because it is the scenario the interview uses.

Reporting thresholds exist for survey and demographic reporting, commonly a minimum of five respondents, sometimes three and sometimes ten. The threshold is not bureaucracy. It protects the promise made to the person who answered the survey, and the day it is broken is the day your response rate collapses and you lose the instrument for years.

Pay equity in the US is a legal process as much as an analysis. Employers routinely run it at the direction of counsel so the preliminary results fall under attorney-client privilege, which has concrete operational consequences for you: the deck is labeled as prepared at the request of counsel, draft numbers are not emailed widely, and the remediation decision is made with legal in the room. A candidate who already knows this does not have to be taught the most expensive lesson in the role.

The reporting obligations that create demand for this job, stated as obligations rather than as dates, because the dates and thresholds in this area change and a candidate who quotes a stale one in an interview has made the only kind of error that matters here. US employers above the covered size file EEO-1 Component 1 workforce demographic data with the EEOC, and federal contractors are covered at a lower headcount than other employers; confirm the current threshold and the filing window for the year you are in, because the window has moved repeatedly. California requires covered employers to file pay data with its Civil Rights Department, including pay band and hours by job category, race, ethnicity and sex, under a regime created by SB 973 and expanded by SB 1162. A growing list of states and cities, including Colorado, California, Washington, New York and Illinois, require a pay scale in the job posting, which is what makes posting data such a good public pay source. Federal contractors have additional obligations administered through the Department of Labor.

Outside the US, three things come up often enough to be worth knowing by name. The GDPR applies to employee data, and consent is a weak lawful basis in the employment context because of the imbalance of power between employer and employee, so processing normally rests on legitimate interests or legal obligation with a documented assessment behind it. In Germany, works councils hold co-determination rights over technical systems capable of monitoring employee behavior or performance, which in practice means a works council agreement before an analytics or collaboration-data tool reaches that population; the Netherlands and France have their own consultation requirements. UK employers at or above the covered headcount publish gender pay gap figures annually. The EU pay transparency directive adds gender pay gap reporting and a duty to conduct a joint pay assessment with worker representatives where a gap in a category exceeds the directive's threshold and is not explained by objective, gender-neutral factors or remedied, but it reaches employers through national transposition and the timing and detail differ by member state, so name the obligation and check the country rather than quoting a single date.

Say all of this the way it is written here: the obligation, then the instruction to check the current date and threshold. Confidently quoting a compliance date that has since moved is a specific and memorable way to lose credibility in an interview with someone who deals with it monthly.

Pay, and the first ninety days you should plan for

There is no clean occupation code for this job, which is both a pay problem and a pay opportunity. The US Bureau of Labor Statistics Occupational Employment and Wage Statistics is still the right starting point, but you have to look at several codes and understand which one your seat is actually priced against: 13-1071 Human Resources Specialists, 13-1141 Compensation, Benefits and Job Analysis Specialists, 13-1111 Management Analysts, 15-2041 Statisticians, 15-2051 Data Scientists, and 11-3121 Human Resources Managers for the lead role. Pull all of them for your metro.

The pattern worth internalizing: the more technical the seat, the more it is priced against the data codes rather than the HR codes, and a people analytics analyst sitting inside a central data organization is typically paid on the data scale while the same job title inside HR is often paid on the HR scale. That single structural fact is worth real money in an offer conversation, and it is a legitimate thing to raise, because you can point at the work rather than at a feeling.

For an actual number rather than a national median, use the sources that reflect live demand. Collect twenty current postings for your target level and metro from pay-transparency states, which publish a real range for a real job at a named employer; that is the strongest public evidence available and it is free. Public sector payroll files and federal GS tables give exact figures for government seats. Levels.fyi is useful for technology employers. If your current employer buys Aon's Radford, Mercer or WTW survey data, your own total rewards team already holds the benchmark for your job, and asking how your role is benchmarked is a normal internal conversation.

Then do the thing the job is: benchmark yourself with the method you would use at work, write the one-page version, and bring it to the negotiation. Few candidates in any field do this. In this one it doubles as a work sample.

Plan your first ninety days before you are asked about them, because you will be asked. The sequence that experienced people describe is almost always the same. Find out what the executive team currently believes and where those numbers come from. Reconcile headcount with Finance and publish the bridge. Write the metric definitions down and get them agreed, because every later argument traces back to a definition nobody wrote. Find out who can see what, and fix the access model if it is loose. Only then build anything new. A candidate who answers the ninety-day question with "build a dashboard" has told the panel which half of the job they have not done.

Where to look for the jobs, concretely. Large technology and financial services employers have the most mature teams and the most competition. Healthcare systems carry an expensive and well-understood workforce problem in nurse turnover and contract labor spend, and they are hiring analysts to work on it. Retail, logistics, hospitality and contact centers run enormous hourly workforces where a point of turnover is a cost the finance team has already priced, and their analytics functions are younger and far less crowded than in technology, which makes them the most realistic door into a first people analytics seat. Consulting firms hire at volume into human capital practices. Vendors hire analysts who can talk to buyers. Government, universities and health systems hire steadily and slowly.

Because the field is small, visibility does disproportionate work. Hiring managers with one open seat fill it from people they have seen talk about the work: in the people analytics communities and Slack groups, at regional meetups, at conferences such as Wharton People Analytics and the HR technology circuit, and in public write-ups. A single well-built public analysis on federal or city workforce data, posted with a clear write-up, reaches more of the people who do this hiring than a long run of cold applications does.

What a screen is looking for

These are the terms that a resume screen, human or automated, is matching against for this role. Use the ones that are true of you, in the words the posting uses.

• People analytics
• Workforce analytics
• HR analytics
• People insights
• Talent analytics
• Workforce intelligence
• HR reporting
• Workforce planning
• Headcount planning
• Headcount reporting
• Attrition analysis
• Turnover analysis
• Retention analysis
• Cohort retention
• Survival analysis
• Kaplan-Meier
• Regretted attrition
• Voluntary and involuntary turnover
• New hire turnover
• Internal mobility
• Promotion rate analysis
• Quality of hire
• Span of control
• Organizational network analysis
• Compensation analysis
• Pay equity analysis
• Pay gap reporting
• Compa-ratio
• Range penetration
• Salary benchmarking
• Engagement survey analysis
• Driver analysis
• eNPS
• Recruiting funnel analysis
• Time to fill
• Time to hire
• Offer acceptance rate
• Adverse impact analysis
• Four-fifths rule
• Bias audit
• EEO-1 reporting
• California pay data reporting
• SQL
• Window functions
• Point-in-time reporting
• Effective dating
• Slowly changing dimension
• Semantic layer
• dbt
• Snowflake
• BigQuery
• Databricks
• Python
• R
• Regression analysis
• Experiment design
• Difference-in-differences
• Tableau
• Power BI
• Looker
• Executive reporting
• Workday
• Workday Prism Analytics
• Workday Report Writer
• SAP SuccessFactors
• Oracle HCM
• UKG
• Dayforce
• ADP
• BambooHR
• HRIS
• Greenhouse
• Ashby
• Lever
• iCIMS
• SmartRecruiters
• Applicant tracking system
• Culture Amp
• Qualtrics
• Workday Peakon
• Microsoft Viva Glint
• Lattice
• Visier
• One Model
• Crunchr
• ChartHop
• OrgVue
• Worklytics
• Skills taxonomy
• Skills inference
• Talent marketplace
• GDPR
• Works council
• Data privacy
• Data governance
• Confidential data handling
• Row-level security
• Stakeholder management
• HR business partnering
• FP&A partnership
• AI adoption measurement
• Automated employment decision tools

Mistakes that cost people this job

Questions people ask