Skip to content

Interview Questions — HR Analytics

Questions an interviewer might ask about this project.


[Beginner] What was the most important finding in your HR attrition analysis?

Show answer

Overtime was the single biggest driver: employees who work overtime leave at 30.5% vs 10.4% for those who don't — a 3× difference.

What made it the most important finding wasn't just the size of the effect — it was that it's actionable. Other strong correlates like age or tenure can't be changed by the company. Overtime can. So I led my recommendations with it: fix overtime, and you address the largest controllable source of attrition.


[Beginner] The dataset has only 16% attrition. Why does this class imbalance matter?

Show answer

With 84% of employees staying, a naive model that predicts "everyone stays" would be 84% accurate — but completely useless, because it never identifies anyone at risk.

This is why accuracy is the wrong metric for imbalanced data. Instead I'd use: - Recall (of the people who actually left, what % did we catch?) - Precision (of the people we flagged, what % actually left?) - F1 score or ROC-AUC to balance them

For retention, recall usually matters most — missing an at-risk employee is more costly than a false alarm.


[Mid-level] You found that single employees and those who live far from the office have higher attrition. Why didn't you recommend acting on these?

Show answer

Two reasons:

  1. Ethics and legality: making employment decisions based on marital status is discriminatory and illegal in most jurisdictions. Even using it to "predict" risk and treat people differently crosses an ethical line.

  2. Non-actionability: the company can't change an employee's marital status or where they live. A finding is only useful if it points to a lever you can pull.

The discipline of HR analytics is separating correlation from actionable, ethical levers. I report these correlations for context but explicitly recommend against acting on them — and I'd focus the company on workload, pay, and development instead.


[Mid-level] How would you estimate the cost of attrition, and what assumptions did you make?

Show answer

I estimated replacement cost as 75% of annual salary per departing employee:

SUM(CASE WHEN Attrition_flag = 1 THEN MonthlyIncome * 12 * 0.75 ELSE 0 END)

Assumptions: - Replacement cost ≈ 75% of annual salary (industry estimates range 50-200% depending on seniority; I used a mid-range figure) - This includes recruitment, onboarding, training, and lost productivity during the gap - It doesn't include harder-to-quantify costs: lost institutional knowledge, team morale impact, customer relationship disruption

I'd flag to stakeholders that this is a conservative estimate and the true cost — especially for senior roles — is likely higher. I'd also offer a sensitivity range (50% to 150%) rather than a single point estimate.


[Mid-level] A department head says "your dashboard shows my team has high attrition, but that's because we hire contractors who always leave." How do you respond?

Show answer

This is a valid data-scoping challenge, and I'd take it seriously.

First, I'd verify: does the dataset distinguish permanent employees from contractors? If contractors are mixed in, their planned departures inflate the "attrition" rate unfairly.

If that's the case, I'd: 1. Add an employment-type field and filter contractors out of voluntary attrition 2. Recalculate the department's permanent-staff attrition 3. Update the dashboard to separate the two

If the data doesn't distinguish them, I'd document this as a known limitation and prioritise getting that field added. This is exactly why analysts must understand the business context — a number without context can unfairly blame a team.


[Senior] HR wants to build a model that flags individual employees as flight risks so managers can intervene. What are your concerns, and how would you advise them?

Show answer

I have significant concerns and would advise caution:

Risks: 1. Self-fulfilling prophecy: if a manager sees an employee flagged as a flight risk, they may unconsciously disengage — fewer growth opportunities, less investment — which accelerates the very departure we're trying to prevent. 2. Privacy and trust: employees discovering they're being "scored" for departure risk damages trust profoundly. 3. Misuse: a flight-risk label could be used to deny promotions or stretch assignments ("why invest in someone who's leaving?"). 4. Accuracy: individual prediction is hard and error-prone. False positives damage real careers.

My recommendation: - Use the model for systemic insight, not individual surveillance: "teams with high overtime have high risk → fix overtime" rather than "employee #4471 is a flight risk." - If individual scoring is pursued, keep it with HR (not line managers), use it only to trigger supportive action (a career conversation, a workload review), never punitive action, and be transparent with employees. - Establish governance: who can see the scores, what actions are permitted, and an audit trail.

The goal of retention analytics should be to improve the employee experience systemically — not to build a watchlist.


← Insights and Recommendations · ← Project Overview