Employee attrition analytics is the practice of combining HR records and employee feedback to explain why people leave, spot who is likely to leave next, and decide what will keep them. It turns a single turnover number into answers about specific teams, tenure bands, and managers.
Most HR teams can report how many people quit last quarter. Far fewer can say which of those exits they could have prevented, or what signal showed up three months earlier.
In this blog, we’ll break down the metrics that matter, the data sources to connect, how attrition prediction works, and a seven-step process with worked examples.
What is employee attrition analytics?
Employee attrition analytics is the process of analyzing workforce data to find the causes and patterns of employee departures, with a focus on voluntary exits an organization wants to prevent. It sits inside the wider field of people analytics. Its focus is one narrow business question: why are we losing people, and what can we do about it?
The work pulls together turnover rates, exit interview answers, employee demographics, job satisfaction surveys, performance ratings, and pay data. Patterns in that combined data point to likely causes. Common ones include low job satisfaction, below-market pay, poor management, stalled careers, or a toxic culture.
The output is not a report for its own sake. Good attrition analysis ends with a short list of decisions. Which managers need support? What roles need a pay review? Which hiring channels bring in people who stay?
What questions does attrition analytics answer?
A useful attrition analysis answers five questions in order. Each one builds on the last. Skipping ahead to prediction without the first three usually produces a model nobody trusts.
- Who is leaving: Which departments, locations, roles, tenure bands, and performance levels lose the most people?
- When they leave: Do exits cluster in the first 90 days, after year two, or after a missed promotion cycle?
- Why they leave: What do exit surveys, stay interviews, and engagement scores say about the drivers?
- Who might leave next: Which current employees share the profile and signals of past leavers?
- What worked: Did the retention actions you took actually lower attrition in the target group?
How is attrition analytics different from turnover reporting?
Turnover reporting counts departures, while attrition analytics explains them and points to action. The two use some of the same formulas, but they serve different purposes and audiences.
Strictly speaking, attrition refers to departures where the role is not backfilled, and turnover assumes the seat gets filled again. In practice, most HR analytics teams use “attrition analytics” to cover all departures, with special attention to voluntary ones. If you need the base formula and US benchmarks first, see this guide on how to calculate employee turnover and reduce it.
| Term | What it answers | Typical output | Who uses it |
|---|---|---|---|
| Turnover rate reporting | How many people left in a period? | One percentage per month or year | HR operations, finance |
| Employee attrition analytics | Why did they leave, and who is at risk next? | Driver analysis, risk segments, action list | HR business partners, people leaders |
| People analytics | How does any workforce factor affect business results? | Studies on hiring, performance, pay, engagement | CHRO, executive team |
| Exit interview program | What did individual leavers say? | Themes from departing employees | HR, direct managers |
Why does employee attrition analytics matter for US employers?
Attrition analytics matters because most voluntary exits are both expensive and preventable, and you cannot prevent what you cannot see coming. Even in a cooler labor market, US workers keep moving.
The latest federal data shows about 3.4 million quits in a single month, a quits rate of 2.1%. At that monthly pace, a 500-person company could lose roughly 10 people every month to resignations alone.
Many of those exits did not have to happen. Workplace research found that 52% of voluntary leavers said their manager or organization could have done something to keep them. The same research puts the cost of replacing one employee at one-half to two times their annual salary.
Each preventable exit carries four kinds of cost:
- Hiring costs: Recruiting, interviewing, and onboarding spend rises with every replacement.
- Lost productivity: New hires take months to reach full output, and teams cover the gap.
- Lower morale: Colleagues of leavers feel the extra workload and start asking whether they should go too.
- Lost expertise: Departing employees take client relationships, process knowledge, and context with them.
Which attrition metrics should you track?
Track a small set of attrition metrics that separate harmful exits from healthy ones, and always calculate them by segment, not just company-wide. An overall rate hides the team that is losing half its people.
| Metric | How to calculate it | What it tells you |
|---|---|---|
| Overall attrition rate | Leavers in period ÷ average headcount × 100 | The baseline you compare everything against |
| Voluntary attrition rate | Resignations ÷ average headcount × 100 | Exits driven by employee choice, the part analytics can influence |
| Regrettable attrition rate | Voluntary leavers you wanted to keep ÷ all voluntary leavers × 100 | Whether you are losing the right or wrong people |
| First-year attrition | Hires who left within 12 months ÷ hires in that cohort × 100 | Quality of hiring, job previews, and onboarding |
| 90-day attrition | Hires who left within 90 days ÷ hires in that cohort × 100 | Early mismatch between the job promised and the job delivered |
| High-performer attrition | Top-rated leavers ÷ top-rated headcount × 100 | Risk to critical talent and future leaders |
| Manager-level attrition | A team’s voluntary leavers ÷ that team’s average headcount × 100 | Where management quality is a likely driver |
| Cohort retention rate | Hires still employed after N months ÷ cohort size × 100 | How each hiring class holds up over time |
A quick example shows why segments matter. A 1,200-person company with 144 voluntary exits a year has 12% voluntary attrition. Now suppose 40 of those exits come from one 150-person customer support group. That group runs at 27%, more than double the company figure.
What data feeds an attrition analysis?
An attrition analysis needs three kinds of data: records of who people are and what happened to them, feedback on how they feel, and behavioral signals that change before someone resigns. Each source alone gives a partial picture.
The strongest analyses join these sources on a common employee ID, then report results only in groups large enough to protect anonymity.
HR system data
Your HRIS (human resources information system, the database that holds employee records) supplies the backbone of any attrition analysis. Pull at least these fields for current and former employees:
- Job facts: Department, location, role, level, hire date, and manager.
- Exit facts: Termination date, voluntary or involuntary, and a reason code.
- Pay facts: Base salary, last raise date, and compa-ratio (salary divided by the midpoint of its pay range).
- Career facts: Last promotion date, performance ratings, and internal moves.
Employee feedback data
Survey data explains the “why” that HR records cannot. Exit surveys capture reasons after the decision is made, so pair them with sources that catch problems earlier.
Stay interviews ask current employees what keeps them and what might pull them away. Regular employee pulse surveys track engagement, workload, and manager support every month or quarter. Employee Net Promoter Score (eNPS) asks how likely employees are to recommend the company as a place to work. That single question gives a fast trend line by team. Onboarding surveys flag new hires who feel misled about the role.
Behavioral and operational signals
Some of the earliest warning signs never appear in a survey. They show up in how work patterns change.
- Absence patterns: A rise in unplanned sick days or late arrivals.
- Time in role: Long stretches since the last promotion, raise, or new assignment.
- Schedule strain: Repeated overtime, weekend shifts, or on-call load above the team norm.
- Manager changes: A new manager or a reorganization in the past six months.
What are the four levels of attrition analytics?
Attrition analytics matures through four levels: descriptive, diagnostic, predictive, and prescriptive. Most organizations should get the first two right before investing in the third.
| Level | Question it answers | Common methods | Example output |
|---|---|---|---|
| Descriptive | What happened? | Rates by segment, trend charts, cohort tables | “Voluntary attrition in sales rose from 14% to 21% this year.” |
| Diagnostic | Why did it happen? | Exit and stay interview themes, key driver analysis, comparisons of leavers vs. stayers | “Leavers rated career growth 1.2 points lower than stayers.” |
| Predictive | Who is likely to leave next? | Logistic regression, survival analysis, tree-based models | “38 sales reps share the profile of last year’s leavers.” |
| Prescriptive | What should we do, and did it work? | Pilot programs, before-and-after tests, control groups | “Teams that ran stay interviews cut regrettable attrition by a third.” |
Jumping straight to prediction is the most common shortcut. Without solid descriptive and diagnostic work, a model can flag risk but cannot tell managers what to change.
How to analyze employee attrition in 7 steps
To analyze employee attrition, define the business question, clean and join your data, segment it, find the drivers, and test fixes on the groups that matter most. The seven steps below work for a 200-person company or a 20,000-person one.
- Define the question and the time window.
Pick one decision the analysis must inform, such as “Why are we losing first-year nurses?” Set a clear period, for example the last 24 months of exits. - Agree on definitions.
Decide what counts as voluntary, regrettable, and early attrition. Write the definitions down so every report uses the same rules. - Collect and join the data.
Combine HRIS records with exit, pulse, and stay interview data using an employee ID. Remove duplicates and fix missing exit reasons. - Calculate baseline metrics by segment.
Run the metrics table above by department, manager, tenure band, location, and performance level. Flag any segment that runs well above the company rate. - Compare leavers with stayers.
Look at how the two groups differ on pay position, promotion timing, engagement scores, and survey comments. The biggest gaps are your candidate drivers. - Confirm the drivers with employees.
Test your hypotheses in stay interviews or focus groups. Data shows where to look, and conversations confirm what is really happening. - Act, then measure again.
Choose one or two interventions per high-risk segment and track the same metrics for the next two or three quarters. Compare results with similar teams that did not get the intervention.
How do you predict employee attrition?
Employee attrition prediction uses past leavers’ data to score current employees on their likelihood of leaving within a set window, such as the next six months. The goal is earlier, better conversations, not a list of names to manage out.
Here is a well-scoped prediction question. “Using data available on March 31, who is likely to resign between April 1 and September 30?” Fixing the cutoff date stops the model from learning from information it would not have had at the time. For a broader primer on the method, see this overview of predictive analytics models and examples.
Which signals predict employee flight risk?
Flight risk is the likelihood that a specific employee will leave voluntarily in the near future. No single signal predicts it well, but combinations do.
- Falling engagement: A drop of a point or more on engagement or eNPS between two pulse cycles.
- Stalled growth: No promotion or role change for longer than peers at the same level.
- Pay below range: A compa-ratio under about 0.9, especially in roles with active outside demand.
- Manager strain: Low manager support scores or a manager whose team already runs high attrition.
- Life-stage moments: Work anniversaries, vesting dates, or the end of a bonus cycle.
Which models are used for attrition prediction?
Teams usually start with a simple, explainable model and move to complex ones only when the gain justifies the loss of transparency. The table compares common options.
| Model | How it works | Best for | Watch out for |
|---|---|---|---|
| Logistic regression | Estimates the odds of leaving from a weighted mix of factors | First models, small datasets, explaining drivers to leaders | Misses interactions between factors unless you add them |
| Survival analysis | Models how long employees stay before leaving | Understanding when risk peaks, such as month 9 or year 2 | Needs accurate hire and exit dates |
| Tree-based models (random forest, gradient boosting) | Combine many decision trees to catch complex patterns | Large workforces with rich data | Harder to explain, easy to overfit |
| Survey driver analysis | Links survey items to intent-to-stay or actual exits | Finding which experience factors matter most | Depends on honest, high-response surveys |
Regression analysis is the backbone of the first and last options. Most HR analysts can build a first model with a spreadsheet or statistics tool. A common rule of thumb is to have at least 10 past leavers for each factor you include.
How accurate is attrition prediction?
Attrition models are useful but far from certain, and you should read their output as probabilities. A simple worked example shows why.
Picture 1,000 employees with 15% annual voluntary attrition, so about 150 people leave. A model flags 200 employees as high risk, and 90 of them leave during the year. That gives two numbers to report:
- Precision of 45%: Of the 200 flagged employees, 90 actually left, versus about 30 if you had picked 200 people at random.
- Recall of 60%: The model caught 90 of the 150 people who left and missed the other 60.
The model triples your hit rate compared with guessing, yet more than half of flagged employees stay. That is why a flag should trigger a supportive check-in from the manager, never a negative action.
Employee attrition analytics examples
The examples below are composite scenarios based on common patterns across industries. They show how the same process leads to different fixes depending on what the data reveals.
Example 1: Retail chain with high 90-day attrition
A regional retailer saw 90-day attrition above 40% for store associates. Segmenting by store showed that a handful of locations drove most of it.
- Finding: Onboarding survey scores on “the job matches what I was told” were lowest in those stores.
- Driver: Hiring managers promised fixed schedules that the stores could not deliver.
- Action: Honest job previews and a two-week schedule commitment for new hires.
- Measure: Store-level 90-day attrition, tracked monthly for two quarters.
Example 2: Software company losing senior engineers
A 900-person software company had stable overall attrition but a rising regrettable rate among senior engineers. Comparing leavers with stayers showed a clear gap in time since last promotion. Leavers had waited roughly 30 months, against 18 for stayers.
Stay interviews confirmed the data. Engineers did not want management roles but saw no other path upward. The company added a senior individual contributor track, and the analytics team now tracks high-performer attrition in engineering every quarter.
Example 3: Hospital system with night-shift nurse exits
A hospital system found first-year nurse attrition far higher on night shifts. Pulse surveys showed low scores for “I have the support I need to do my job” on those units.
- Finding: New nurses on nights had fewer experienced colleagues to ask for help.
- Action: A buddy program pairing each new night nurse with a senior peer for 90 days.
- Measure: First-year attrition and pulse support scores for night units vs. day units.
How do you turn attrition insights into retention actions?
Match each confirmed driver to a specific action and a metric that proves whether it worked. Generic retention programs spread budget thin, while targeted fixes go where the data says the problem is.
| Confirmed driver | Retention action | Metric to watch |
|---|---|---|
| Weak management practices | Manager coaching, regular one-on-ones, and support scores in manager reviews | Manager-level attrition and support scores |
| Poor employee experience | Fix workload, flexibility, and recognition issues raised in pulse comments | Engagement and eNPS by team |
| Limited development | Clear career paths, training budgets, and internal job postings | Time since promotion and internal mobility rate |
| Unclear reasons to stay | Stay interviews with high performers twice a year | Regrettable attrition rate |
| Below-market pay and benefits | Market pay review for roles with compa-ratios under range | Voluntary attrition in reviewed roles |
What are the most common attrition analytics mistakes?
The most common mistakes come from measuring too broadly, trusting one data source, or acting on scores without context. Each one makes the analysis less accurate and less trusted.
- Reporting only company-wide rates: Averages hide the teams and roles where attrition is concentrated.
- Mixing voluntary and involuntary exits: Layoffs and terminations distort the drivers you can actually influence.
- Relying on exit surveys alone: By the time someone answers an exit interview, the decision is final and answers often stay polite.
- Treating every exit as bad: Some attrition of low performers is healthy, so track regrettable attrition separately.
- Skipping the follow-up measurement: Without before-and-after data, you never learn which actions worked.
- Sharing individual risk scores widely: Broad access to flight-risk names damages trust and invites misuse.
How do you keep attrition analytics ethical and legal?
Keep attrition analytics ethical by using data to improve conditions for groups of employees, not to single out individuals for negative treatment. Transparency about what you collect and why is the foundation of employee trust.
Start with anonymity. Report survey results only for groups above a minimum size, often five to ten respondents, so no one can be identified. Limit individual-level risk scores to a small group of HR partners who use them to open supportive conversations.
US employers also need to watch for discrimination risk. Do not use protected characteristics such as age, race, sex, disability, or pregnancy as model inputs. Check model outputs for disparate impact as well. A flag that falls far more often on one protected group can raise concerns under federal equal employment law. Review models with legal counsel before using them in any decision about pay, promotion, or termination.
How QuestionPro supports employee attrition analytics
QuestionPro supports employee attrition analytics by linking lifecycle feedback to HR attributes, so survey drivers can be compared across the same segments you use for attrition metrics. That connection is what turns survey scores into an attrition explanation.
With QuestionPro Employee Experience, HR teams can run onboarding, pulse, engagement, 360-degree, and exit surveys from one platform. An FTP or API-based employee roster integration brings in attributes such as department, location, tenure, and manager. A confidential demographics roster keeps responses anonymous while still allowing that segmentation.
Several capabilities map directly to the steps in this guide:
- Key driver analysis: Shows which experience factors move engagement and intent to stay the most.
- eNPS tracking: Gives a running trend line by team to catch early drops.
- Text and sentiment AI: Groups open-ended exit and pulse comments into themes and flags negative sentiment.
- Role-based dashboards: Give each manager a view of their own team’s results.
- Action planning: Assigns owners and deadlines to the fixes your analysis recommends.
Attrition analytics pays off when it changes the next conversation
Employee attrition analytics earns its value when the findings reach the people who can act on them. A segmented dashboard is useful. The real result, though, is a manager who holds a stay interview this month because the data showed a team at risk.
Start small. Track voluntary and regrettable attrition by team, connect one feedback source, and test one fix in your highest-risk segment. Add prediction once the basics are trusted, and keep measuring so every retention dollar goes where it makes a difference.
Frequently Asked Questions (FAQs)
Descriptive analysis works with a single year of HRIS exit records. Prediction needs more, usually two or more years of history and enough leavers to support each factor. Many teams aim for at least 10 past leavers per factor in a model.
Yes. A company with 50 to 500 employees can calculate attrition by team, compare leavers with stayers in a spreadsheet, and run pulse surveys. Formal prediction models matter less at that size, where managers already know most employees personally.
Review attrition metrics monthly for high-turnover teams such as retail, hospitality, and call centers, and quarterly for most other groups. Run a full driver analysis once or twice a year, timed after your main engagement survey closes.
Generally yes, but the model must not rely on protected characteristics or produce discriminatory outcomes. Federal equal employment law applies to how you act on predictions. Some states also have privacy rules on employee data, so involve legal counsel early.
You need an HRIS for employee records and a survey platform for exit, stay, and pulse feedback. Add an analysis tool such as a spreadsheet, BI dashboard, or statistics software. The most important feature is the ability to join these sources reliably.



