Safiul Azam Data & Information Management
Power BIDAXHR analytics

IBM HR Analytics — Employee Attrition Console

A four-page Power BI console that takes a flat HR file with no dates in it and still answers the three questions an HR director actually asks: who is leaving, why, and who is next.

RoleAnalyst & dashboard developer
OrganisationIndependent project
Year2026
ToolsPower BI · DAX · Power Query

This project can be shown in full. It uses a public dataset with no sensitive content, so every figure, method and result on this page is the real thing — the visuals are screenshots of the working output, not rebuilds.

Context

The IBM HR dataset is one of the most over-used files in analytics, and almost every version of it online makes the same mistake: it reports a headline attrition percentage as if it were a turnover rate. It is not. The file is a point-in-time snapshot with no date columns at all, so “16.1%” is the share of the recorded population that has left — not an annualised rate.

I decided that constraint was the interesting part. If you cannot build a trend line, you have to earn insight from structure instead: department, role, tenure band, pay. The design brief became “say only what this data can actually support, and say it clearly”.

Approach

  • Label the limits on the canvas. Every KPI carries a plain-English qualifier — “share of population, not annualised”, “currency not stated in source”, “at 50% of salary — an assumption, not data”. A reader should not have to guess which numbers are solid.
  • A baseline on every comparison. Each attrition bar chart draws the 16.1% population baseline as a dashed reference, so a department is read against the whole rather than against its neighbours.
  • Show the denominator. Small groups produce dramatic percentages. Every bar carries its n, so 19.1% attrition in a 63-person HR department is not mistaken for a bigger problem than 13.8% across 961 people in R&D.
  • Four pages, four questions. Workforce (what is the shape), Why People Leave (what correlates), Who Is At Risk (where to act), Pay & Equity (is compensation part of it).

The four pages

Built on the public IBM Watson Analytics sample HR dataset — 1,470 employee records. These are real screenshots of the working report.

1 Workforce overview

Real output · public dataset Power BI workforce overview page showing headcount, attrition rate, median income, tenure and replacement cost
The opening page. 1,470 employees, 16.1% attrition, an estimated $6.81M replacement cost. Attrition by department and by job role each carry the 16.1% baseline as a dashed line, and the onboarding-cliff chart shows a survival curve rather than a timeline — because the source has no dates.

2 Why people leave

Real output · public dataset Power BI page analysing the drivers of attrition including overtime, job satisfaction and work-life balance
Driver analysis. Overtime, satisfaction and work-life balance are read against the same baseline so the size of each effect is comparable.

3 Who is at risk

Real output · public dataset Power BI risk-tier page identifying the employee segments most likely to leave
The action page — risk tiers that turn the analysis into a shortlist rather than a description.

4 Pay & equity

Real output · public dataset Power BI page comparing compensation across roles, levels and gender
Compensation compared across role, level and gender, to test whether pay explains any of the attrition pattern.

Outcome

39.8% attrition found in one role

The structural read produced sharper findings than a headline rate would have. Sales Representative attrition runs at 39.8% — 2.5× the population baseline, against 2.5% for Research Directors. The onboarding cliff is stark: 36.4% of people in their first year and 34.5% in their second leave, and attrition among the under-25s is 35.8% against 9.2% for 35–44.

Read together, those three findings point at one intervention rather than a general retention programme: the risk is concentrated in early-tenure, younger staff in a specific job family.

What I’d do differently

The replacement-cost figure is an assumption dressed as a metric — 50% of salary is a common rule of thumb, not a measurement, and I labelled it as such on the canvas rather than quietly presenting it as data. With a real cost model per role it would become genuinely decision-grade.

The larger limitation is the dataset itself. Without hire and exit dates you cannot compute a true turnover rate, cohort survival, or seasonality. If I were commissioning this data I would ask for those three columns before anything else.