SQL + Python | IBM HR Analytics Dataset
Finding why employees leave — and what it's actually costing the company
237 employees left. That is 16% of the workforce.
Using a standard 6-month replacement cost, this company is spending roughly $6.8 million to replace people who left — and the exits are not random. They are concentrated in the same roles, under the same conditions, for the same reasons.
This analysis finds exactly where the bleeding is happening — and shows that fixing overtime, pay, and early-career retention can cut attrition significantly and save millions per cycle.
Attrition here is not a company-wide crisis. It is a concentrated problem.
A small group — primarily Sales Representatives working overtime and earning below market rate — is responsible for a disproportionate share of the total cost. The rest of the company is comparatively stable. This means broad, company-wide policies will not work. The fix needs to be targeted at specific roles, under specific conditions.
If the company ignores this, the $6.8M replacement cost repeats every single cycle. If the targeted fixes are applied, the data shows attrition in the highest-risk groups can drop significantly — with measurable financial impact.
Sales Representatives working overtime leave at 66.7%.
Two out of every three Sales Reps working overtime will quit. This is not a small pattern. It is the loudest signal in the entire dataset.
| Area | Number | What It Means |
|---|---|---|
| Overall attrition | 16.12% | 237 out of 1,470 employees left |
| Estimated replacement cost | ~$6.8M | 237 × $4,787 avg salary × 6 months |
| Overtime vs no overtime | 30.5% vs 10.4% | Nearly 3x higher attrition for overtime workers |
| Sales Reps on overtime | 66.7% | Highest risk group in the entire dataset |
| Lab Technicians on overtime | ~50% | 3x their no-overtime rate |
| Pay gap | $4,787 vs $6,833/month | Leavers earned $2,046 less — confirmed by t-test (p < 0.05) |
| Youngest employees (18–25) | 35.8% attrition | More than double the company average |
| Early leavers avg tenure | 5.1 years | vs 7.4 years for employees who stayed |
The same role behaves completely differently depending on overtime load. Lab Technicians without overtime leave at ~16%. With overtime, that number jumps to ~50%. The role did not change. The workload did.
Overtime is the single strongest factor in this dataset. Nothing else creates a gap this large.
Employees who left were earning $2,046 less per month than those who stayed. This gap exists in every department — it is not a Sales problem or an HR problem. It is a company-wide pattern.
A t-test confirms this is not a coincidence (p < 0.05). The income difference is real.
Sales Representatives leave at 39.8% — nearly 10x higher than Research Directors at 4%. The gap between roles is not small. Attrition is heavily concentrated in a few specific positions.
Sales as a department has the highest attrition at 20.6%, followed by HR at 19.1%. R&D is the most stable at 13.8%.
Most people who leave do it in the first five years. Employees who left had an average tenure of 5.1 years, compared to 7.4 years for those who stayed.
The 18–25 age group leaves at 35.8% — the highest of any age bracket. This group is not leaving because they hate the company. They are leaving because they cannot see where they are going.
The correlation heatmap shows that no individual column strongly predicts attrition on its own. Attrition happens when multiple problems stack together — low pay, overtime, early career stage, and role type all compound each other.
Other patterns worth noting:
- Single employees leave at 25.5% vs 12.5% for married employees — likely because they are younger, not because of marital status itself
- Job satisfaction level 1 has 22.8% attrition vs 11.3% at level 4 — satisfaction matters, but it is not the root cause
To confirm the income gap is real and not just a coincidence, I ran an independent-samples t-test comparing the monthly income of employees who left vs those who stayed.
- T-statistic: -6.20
- P-value: < 0.05
The result is statistically significant. Employees who left were genuinely earning less. This is not random variation in the data.
I scored each employee group by combining attrition rate, group size, and statistical significance — so the focus goes to groups that are both high-risk and large enough to actually matter, not just small groups with extreme but unreliable numbers.
The top three risk groups:
- Sales Representatives on overtime → 66.7% attrition
- Laboratory Technicians on overtime → ~50% attrition
- Research Scientists on overtime → ~34% attrition
| Problem | Action | Expected Impact |
|---|---|---|
| Sales Reps on overtime leaving at 66.7% | Cap overtime in Sales, redistribute workload | Brings attrition toward company average — retains ~50 Sales Reps per cycle, saves ~$1.4M |
| Pay below market for high-risk roles | Salary review for Lab Technicians, Research Scientists, Sales Reps | Replacing one employee costs ~$28K. A $700/month raise costs $8,400/year and pays back in under 4 months |
| Early-career employees leaving fast | Structured check-ins at 6, 12, 24 months. Clear growth path for employees under 25 | 20% reduction in early-tenure attrition retains ~47 additional employees — saves ~$1.35M |
| Over-reliance on satisfaction surveys | Fix overtime and pay first, then measure satisfaction | Satisfaction matters but is not the root cause — a satisfied employee working overtime and earning below market will still leave |
Fix overtime first. Then pay. Then, early-career support. In that order.
If nothing changes, the $6.8M replacement cost repeats every cycle. The data shows that targeting these three areas specifically — not broad company-wide policies — is where the highest return is.
This analysis shows that attrition is not a vague HR issue — it is a measurable, high-cost problem with clear, data-backed fixes.
If the company fixes the main problems — overtime, low pay, and early employee exits — the impact is clear and directly tied to money.
Right now, Sales Representatives who work overtime are leaving at 66.7%. That means 2 out of 3 people quit.
If overtime is reduced and their attrition comes closer to the company average (~16%):
- Around 90 Sales Reps are in this group
- Right now, about 60 of them leave
- After fixing overtime, this can drop to around 15
- So, ~45 employees can be retained
Each employee costs about $28,700 to replace.
- Total savings: ~45 × $28,700 = ~$1.29M per cycle
This is the easiest win because it doesn’t need extra salary spend — just better workload handling.
Employees who left were earning $2,046 less per month compared to those who stayed.
If the company increases salary for high-risk roles by even $700/month:
- Yearly cost per employee: $8,400
- Replacement cost per employee: ~$28,700
So the company recovers this cost in less than 4 months.
If this prevents even 30 people from leaving:
- Savings: ~30 × $28,700 = ~$861K
Even after a salary increase, the company still comes out ahead.
Employees aged 18–25 are leaving at 35.8%, which is more than double the company average.
If better onboarding and career guidance reduce this by just 20%:
- Around 45–50 employees can be retained
- Savings: ~47 × $28,700 = ~$1.35M
This works because small improvements across a large group add up fast.
If all three actions are taken:
- Overtime fix: ~$1.29M
- Salary fix: ~$0.86M
- Early retention: ~$1.35M
Total savings: ~ $3.5M per cycle
The company is already losing about $6.8M per cycle.
If this continues:
- In 3 years → $20M+ loss
- Constant hiring pressure
- Loss of trained employees
- Lower team performance
This problem doesn’t go away on its own — it keeps repeating.
- Fix overtime → fastest and biggest impact
- Fix salary for risky roles → quick return
- Improve early employee experience → long-term stability
Lab Technicians and Sales Executives lose the most employees in raw numbers, partly because they are larger teams. High total exits do not always mean the highest risk — but they do mean the highest replacement cost in absolute terms.
- This is pattern analysis, not a prediction model. I identified risk factors — not which specific employee will leave next.
- Correlation is not causation. Single employees leaving more is likely because they are younger, not because of their marital status.
- The dataset is fictional (created by IBM). Dollar figures use industry-standard benchmarks, not real company financials.
- Source: IBM HR Analytics Employee Attrition & Performance
- Size: 1,470 employees, 35 columns
- Data quality: No missing values, no duplicates, all values within valid ranges
| Tool | Purpose |
|---|---|
| MySQL | Core analysis — attrition by department, role, income, overtime, age, tenure |
| Python (Pandas) | Reshaping SQL results for deeper analysis |
| Matplotlib & Seaborn | All visualizations |
| SciPy | Independent-samples t-test for income validation |
| Jupyter Notebook | Full analysis in one place, start to finish |
- Clone this repo
- Make sure you have Python, MySQL, and Jupyter Notebook installed
- Import the CSV into MySQL (table name:
hr_attrition, database:hr) - Open
notebooks/hr_attrition_analysis.ipynband update the MySQL connection credentials - Run all cells
Libraries needed:
pandas
matplotlib
seaborn
mysql-connector-python==8.0.33
scipy
numpy
hr-attrition-analysis/
│
├── notebooks/
│ └── hr_attrition_analysis.ipynb ← Full analysis (SQL + Python + Charts)
├── data_setup.sql ← MySQL table setup
├── README.md ← You are reading this
└── images/ ← All chart exports
├── Department_Attrition_Final.png
├── JobRole_Attrition_Final.png
├── Monthly_Income_Attrition.png
├── Overtime_Attrition.png
├── Tenure_Attrition.png
├── Age_Group_Attrition.png
├── Top_5_Attrition.png
├── Satisfaction_Attrition.png
├── Income_department_Attrition.png
├── Marital_Attrition.png
├── Comprehensive_Risk_Profile_Large.png
├── Correlation_Matrix_HighVis.png
└── TopRiskPriorityBars_Premium.png
Ashish Kumar Dongre
🔗 LinkedIn | 💻 GitHub | 📂 Dataset on Kaggle