Voluntary Turnover Analysis: 5 Root Causes (2026)

The average voluntary turnover analysis stops at two answers: people left for more money, or people left for a better growth opportunity. Both are usually surface explanations that hide the actual root cause sitting one or two layers underneath. Work Institute's longitudinal data on 250,000+ exits shows 75% of voluntary turnover is preventable, yet most companies intervene on the wrong lever because their analysis never reached the real driver.

This guide is a working framework for voluntary turnover analysis that goes past the comp-and-growth defaults. It covers the 5 root cause categories every analysis must distinguish, the 4 data sources you need, the 5-step workflow from raw exits to retention intervention, the cohort segmentation that exposes manager-level patterns, and the common analytical mistakes that produce false-positive root causes. For the broader offboarding context, start with our employee offboarding process complete guide.

Every framework here is grounded in research from Work Institute, Gallup, McKinsey, AIHR, and SHRM, and aligned with EU AI Act and GDPR requirements for people analytics. The cost of getting this wrong is not abstract: regrettable turnover runs 50% to 200% of annual salary per departure. The cost of a wrong root cause is paying that bill twice, because the next person in the same seat leaves for the same unaddressed reason.

Why Most Voluntary Turnover Analysis Fails

The dominant pattern in mid-market HR teams: pull a quarterly turnover number, segment by department, read the exit interview comments, conclude the issue is compensation or career growth, and propose a salary band review. Six months later the rate has not moved, and the next round produces the same conclusions. The cycle repeats because the analysis confuses the stated reason for leaving with the root cause of leaving.

The stated-reason trap

Employees rarely lie in exit interviews. They also rarely tell the full truth. "I got a better offer" is true and incomplete. The honest version is closer to "I started looking because my manager stopped giving me feedback 4 months ago, and once I was looking the better offer was easy to find." Two very different interventions follow from those two framings.

The aggregation and "we already know" traps

Company-wide rates are almost useless for root cause work. A 14% rate evenly distributed is a different problem from a 14% rate running 35% in one team and 4% elsewhere. The first signals a structural issue, the second signals a manager. Most analyses also cut by one dimension at a time (department, tenure, gender), but the actionable patterns sit at intersections like "women in engineering with 18 to 36 months of tenure under 3 specific managers". Single-axis analysis cannot see those intersections. Then there is the "we already know" pattern: HR teams who have run a few exit cycles develop a narrative about why people leave here, and the analysis effectively confirms what was already believed. Disconfirming evidence gets explained away.

Voluntary, Involuntary, Regrettable, Preventable

Before any root cause work, the categorization has to be clean. Mixing categories produces analytic noise that no amount of slicing can correct.

Voluntary vs involuntary

Voluntary turnover is employee-initiated separation: resignations, retirements, fixed-term contracts not renewed at the employee's request. The voluntary turnover rate equals voluntary exits divided by average headcount, expressed as a percentage. SHRM's 2024 European mid-market benchmark sits between 11% and 17% annually. Involuntary turnover is employer-initiated: performance dismissals, role eliminations, RIFs, end of probation. Different root causes, different legal context. Mixing involuntary into voluntary inflates the rate and confuses the diagnostic.

Regrettable and preventable

Regrettable turnover is the subset of voluntary exits the company would have preferred to retain, defined by performance band, role criticality, and replacement difficulty. It is the only category where the cost calculation makes sense: 50% to 200% of annual salary per departure for individual contributors, scaling to 400% for senior or leadership roles per AIHR's 2025 meta-analysis. Preventable turnover is the subset that better management, fit, or recognition would have prevented, around 75% of voluntary exits per Work Institute. The cell that gets the analytical attention is regrettable + preventable. That is where root cause analysis turns into ROI.

The 5 Root Cause Categories Every Analysis Must Distinguish

Most exit reasons collapse into one of 5 root causes. Treating them as a single bucket called "engagement" is the most common analytical error. Each category has a different signature in the data and a different intervention. A clean turnover root cause analysis tags every regrettable, preventable exit into exactly one primary category, optionally with a secondary.

Category 1: Manager relationship

The boss-as-root-cause category. Gallup's State of the Workplace data attributes 52% of voluntary exits to manager-influenced factors: lack of feedback, micromanagement, favoritism, no career advocacy, broken trust after a specific incident. The largest single category in most datasets and the highest-leverage one, because manager behavior is changeable and concentrated.

Category 2: Compensation and equity

Three sub-causes hide inside this category and need separate analysis: market gap (external comp materially higher), internal-peer gap (a coworker doing similar work earns more), and personal comp shock (cost of living, family change, equity vesting cliff). Each sub-cause has a different fix. Treating them as one is why blanket band adjustments often fail to move retention.

Category 3: Growth path

Not training opportunities in the abstract. The specific question is whether the employee saw a credible 18 to 24 month forward path inside the company. Two failure modes: no path exists in the role family (a structural job architecture problem), or a path exists but the employee did not see it (a manager-conversation problem).

Category 4: Role-fit and meaning

The work itself is wrong for the person. Skill-task mismatch, values mismatch, or meaning erosion (the role used to feel important and stopped). Often surfaces as "I want a new challenge" in exit interviews. The challenge framing hides the underlying mismatch.

Category 5: Life events and external pull

Genuinely external causes: relocation following a partner, family caregiving demands, health, return to study, voluntary career pivot into a different industry. Mostly non-preventable and should not be force-fitted into preventable categories. Every exit gets tagged into 1 primary plus 0 to 1 secondary categories within 30 days, by a human reviewer with access to the exit transcript, manager debrief, stay interview history, and the last 2 performance review cycles.

The Data You Actually Need: A 4-Axis View

A serious voluntary turnover analysis needs 4 data axes. Most companies have 1 or 2 and try to compensate with deeper analysis on insufficient data. The result is high-confidence wrong answers. Build the missing axes before you build the analysis.

Axes 1 and 2: Structured exit data and stay interview signals

Axis 1: separation date, last working day, role, level, department, manager, tenure in role, tenure in company, performance band for the last 2 cycles, regrettable flag, preventable flag, and the 5-category root cause tag. Plus structured exit interview answers on a fixed instrument so you can compare across exits, not just read them as anecdote (see our exit interview questions guide). Axis 2: the strongest predictor of who is about to leave is what they said in their last stay interview, not their exit interview. Stay interviews run quarterly with key talent. The asymmetry: stay signals predict, exit interviews explain. Both are needed. See our stay vs exit interview comparison.

Axes 3 and 4: Engagement slope and manager-level cohorts

Axis 3 is the slope, not the absolute score. An employee at 7/10 trending down for 3 consecutive pulses is at higher risk than one at 5/10 stable. Most engagement platforms surface absolute numbers; the slope is what predicts. If you do not have pulse data, the manager's qualitative read on team energy in 1:1 notes is a usable proxy (our 1-on-1 meeting template guide shows how to capture this). Axis 4: roll up turnover, exit reasons, engagement slope, and stay interview signals to the manager level, including past teams that manager has run. Without this view you cannot see the 80/20 manager concentration that explains most preventable turnover.

The 5-Step Voluntary Turnover Analysis Workflow

A repeatable cycle, run quarterly at minimum, monthly if turnover is elevated. Each step has a defined output that feeds the next. Skipping steps produces narrative analysis that feels insightful and changes nothing.

Step 1: Collect and clean

Pull all voluntary exits for the period. Apply the regrettable and preventable flags using criteria leadership agreed in advance. Tag each exit into the 5 root cause categories from the 4-axis data. Output: a structured dataset with one row per exit. Usually 2 to 4 hours per quarter once the pipeline is built.

Step 2: Segment and concentrate

Cut the dataset by tenure band, role family, manager, performance band, and engagement slope. Look for cells with rates materially above the company average. Output: 3 to 7 high-concentration cohorts that account for the majority of regrettable and preventable exits.

Step 3: Root cause hypothesis

For each high-concentration cohort, generate a root cause hypothesis grounded in the 5 categories. Test the hypothesis against the qualitative data: exit transcripts, stay interview history, manager 1:1 notes. Disconfirm or refine. Output: 3 to 7 hypotheses with evidence strength rated low, medium, or high. Low-evidence hypotheses go back to data collection, not to intervention.

Step 4: Intervene

Each medium or high evidence hypothesis gets a paired intervention with a measurable target and a 60 to 90 day timebox. Manager-rooted hypotheses get coaching, role change, or structured manager development. Compensation-rooted hypotheses get targeted band work or specific case adjustments. Growth-rooted hypotheses get job architecture work or manager conversation enablement. One owner per intervention, named in writing.

Step 5: Measure and iterate

Track the same cohort 60 and 120 days post-intervention. Did the engagement slope flatten? Did regrettable exits drop? Did stay interview sentiment shift? If the intervention did not move the needle, the hypothesis was wrong or the intervention was insufficient. Output: a quarterly retention scorecard that tracks intervention ROI, not just turnover rate.

Cohort Analysis: Tenure x Role x Manager x Performance Band

Actionable insights live at the intersection of dimensions. A clean cohort analysis is a 4-way cube, not a stack of 1-way reports. Mid-market companies between 200 and 2,000 employees can run this in a spreadsheet. Above 2,000, dedicated people analytics tooling pays back fast.

Tenure bands, role families, manager rollup

Standard tenure cuts: 0-6 months (onboarding fail), 6-18 months (role-fit reveal), 18-36 months (growth path test), 36-60 months (mid-career inflection), 60+ months (deep tenure). Each band has a typical root cause profile. Onboarding-fail exits in the first 6 months almost always trace to hiring or expectation-setting failures, not engagement. Cut by role family (engineering, sales, customer success, ops), not by job title, which fragments the data. Then aggregate every exit under each manager, including past teams, with three years of history as the working horizon. Managers with consistently elevated rates across multiple teams are a different intervention target from managers with one bad cohort.

Performance band and the intersection that pays back

Layer performance band: high-performer exits and low-performer exits look identical in raw numbers and have opposite implications. Regrettable turnover is concentrated in high-performer exits. The single most useful query: high-performer exits in the 18-36 months tenure band, by manager. This intersection isolates regrettable, preventable, manager-influenced turnover with high precision. Most companies that run it for the first time discover that 2 or 3 managers account for 30% to 50% of their regrettable losses.

Manager-Level Patterns: The 80/20 of Preventable Turnover

Preventable turnover is not evenly distributed across managers. Across multiple AIHR and Gallup datasets, roughly 20% of managers account for 60% of preventable turnover. The shape varies by company size and industry, but the concentration is consistent. This is the highest-leverage finding most turnover analysis efforts produce, and most companies fail to act on it cleanly.

Three profiles in the high-loss tail

The 20% who lose people are not all the same. Profile 1: the conflict-avoidant manager who never gives feedback and lets quiet disengagement build until the resignation. Profile 2: the high-pressure manager who drives short-term performance through intensity that high-performers tolerate for 18 months and then leave. Profile 3: the absent manager who is technically excellent but spends no time on people, so reports drift toward better-managed teams or external offers.

Why training alone fails, and the hard call

Generic manager training does not move the needle for the high-loss tail. The conflict-avoidant manager already knows they should give feedback. The high-pressure manager already knows people need recovery. The absent manager already knows people need 1:1s. The gap is not knowledge, it is behavior. Behavior change requires structured coaching with measurable behavioral targets, not a workshop. Some managers in the high-loss tail will not change. The honest analysis includes role change as one of the interventions: move them out of people management into IC or specialist tracks where their strengths apply without the people-cost. Our hidden performance networks article covers how to spot these patterns at the org chart level.

Compensation and Growth-Path Root-Causes In Depth

Compensation and growth path are the two most over-attributed root causes. "They left for more money" and "no growth opportunity" both close the analysis prematurely. Each one deserves a clean sub-category split before any intervention is named.

Compensation: market, internal-peer, personal shock

Market gap: total comp materially below external market. Tested against current market data (Mercer, Radford, industry benchmarks) at the time of resignation, not at last band review. An in-band employee can still be 15% behind market. Internal-peer gap: the employee discovered a coworker doing similar work earns materially more. More corrosive than market gap because it triggers fairness violation, not just opportunity cost. Personal comp shock: the employee's needs changed (cost of living spike, family change, equity vesting cliff). Each sub-cause has a different fix.

The disqualifier test

Before tagging any exit as comp-rooted, run the disqualifier: would this employee have stayed at +15% comp if everything else was unchanged? If the honest answer is "probably not, they were already disengaged", the root cause is the underlying engagement issue, not comp. Comp was the trigger, not the cause. This test cuts false-positive comp tagging by roughly 30%.

Growth path: did one exist, did the employee see it

Path exists, employee did not see it: the role architecture has a credible 18 to 24 month next step, but the employee never had the conversation that mapped them to it. Fix: structured career conversations every 6 months, individual development plans linked to role-family progression, internal mobility programs. Path does not exist for this role family: common in specialized IC roles and flat organizations past their growth phase. Fix: job architecture work for lateral or specialist paths, honest conversations about the cap, deliberate alumni programs. The most common operational failure across both sub-causes is the missing career conversation. Stay interviews and structured 1:1 cadence catch this 6 to 12 months earlier than exit interviews. Our 1-on-1 meeting template includes the cadence by default.

Common Voluntary Turnover Analysis Mistakes

Patterns that show up consistently in HR teams running their first or second analytical cycle. Each one produces a confident wrong answer.

Wrong headline, exit comments as ground truth, ignoring stayers

Total voluntary turnover rate is a vanity metric. The headline metric should be regrettable + preventable turnover rate, broken down by manager and tenure band. The total rate makes leadership feel something is being measured; the segmented rate tells them where to act. Exit interview free-text is signal, not evidence. It needs cross-checking against stay interview history, performance review notes, and manager debrief. A pattern across 5 exits is evidence; a single emotional comment is not. And the control group is the people in the same cohort who stayed: if the high-loss cohort is also a high-stay cohort, the manager or role might not be the root cause. Our data-driven burnout prevention guide covers the stayer-side analysis.

Annual cycle, no measurement, GDPR and AI Act blind spots

Annual cycles cannot intervene in time. The decision to leave is typically made 2 to 4 months before the resignation, and the disengagement that produced it started 6 to 12 months earlier. Quarterly is the minimum useful cadence. Most analyses also end with recommendations and never measure whether they worked: the 60 and 120-day cohort follow-up is what makes the next cycle smarter than this one. Finally, people analytics in the EU sits inside GDPR (lawful basis, purpose limitation, data minimization) and the EU AI Act (transparency on automated decisions, high-risk classification for some HR uses). Aggregate cohort analysis is fine. Individual-level "flight risk" scoring without governance is a compliance issue and needs DPIA, clear lawful basis, and human review.

Quantitative vs Qualitative, and the AI Layer

Both modes are required, in roughly an 80/20 split: quantitative for cohort identification, qualitative for root cause confirmation. Most HR analyses get the split inverted, which feels insightful and rarely changes outcomes. Run quantitative first to identify the 3 to 7 cohorts that matter, then qualitative deep-dives on those cohorts. LLM-based analysis can read 50 exit transcripts in minutes, tag them against the 5 categories, and link them back to cohort data, compressing 2 weeks of work into 2 hours while keeping a human in the loop for tag confirmation. Our exit interview automation guide covers the AI workflow with EU AI Act compliance.

KS-Agents Torch: Pay-Per-Use Knowledge Extraction

KS-Agents Torch runs the AI-driven knowledge extraction interview when an employee leaves, produces structured records ready for the successor, and ships with EU data residency, GDPR Article 22 compliance, and EU AI Act alignment built in. Every AI-assisted output is confirmable and reversible by a human reviewer. Torch is pay-per-use, with no subscription required: EUR 45 for one extraction interview, EUR 175 for a 5-interview pack. See Torch pricing →

FAQ: Voluntary Turnover Analysis

What is voluntary turnover analysis?

The systematic study of employee-initiated separations to identify root causes, predict future risk, and design retention interventions. It distinguishes voluntary from involuntary, regrettable from non-regrettable, and preventable from non-preventable exits, and uses cohort segmentation to find concentration patterns.

What is a healthy voluntary turnover rate?

SHRM benchmarks for European mid-market sit between 11% and 17% annually depending on industry. The total rate matters less than the regrettable + preventable subset, which should ideally be below 8%.

What is the difference between voluntary and regrettable turnover?

Voluntary turnover is any employee-initiated separation. Regrettable turnover is the subset the company would have preferred to retain, defined by performance band and role criticality. All regrettable turnover is voluntary, but not all voluntary turnover is regrettable.

What percentage of voluntary turnover is preventable?

Work Institute's longitudinal data on 250,000+ exits puts preventable voluntary turnover at approximately 75%. The remaining 25% is driven by genuine life events, career pivots, or other external causes.

What are the main causes of voluntary turnover?

Five root cause categories: manager relationship (the largest single cause, around 52% per Gallup), compensation and equity, growth path, role-fit and meaning, and life events with external pull.

How do I calculate voluntary turnover rate?

Voluntary turnover rate equals voluntary exits in the period divided by average headcount, expressed as a percentage. Annualize sub-year rates carefully; multiplying a quarterly rate by 4 overstates the annual figure if exits cluster in specific months.

How often should we run a voluntary turnover analysis?

Quarterly is the minimum useful cadence. Monthly is better when turnover is elevated or when interventions are in flight. Annual cycles are too slow, since the decision to leave is typically made 2 to 4 months before the resignation.

What data do I need for a serious turnover analysis?

Four axes: structured exit data with the 5-category root cause tag, stay interview signals, engagement trend slope (not just absolute scores), and manager-level cohort rollups including past teams.

How does the EU AI Act affect turnover analysis?

Cohort-level analysis is unrestricted. Individual-level "flight risk" scoring with automated decisions falls under high-risk AI categories and requires DPIA, transparency to data subjects, lawful basis under GDPR, and human-in-the-loop review.

What is the cost of voluntary turnover?

Regrettable turnover costs run 50% to 200% of annual salary per departure for individual contributors, scaling to 400% for senior specialist or leadership roles per AIHR's 2025 meta-analysis. See our strategic analysis of knowledge loss for the full cost model.

From Analysis to Retention Strategy

A voluntary turnover analysis that does not produce 3 to 5 named interventions with owners, targets, and timeboxes is incomplete. The analytical work is the cheap part; the operational work of changing manager behavior, fixing comp bands, and building career paths is where the cost and the payoff sit. Three things to do this quarter: rebuild the last 12 months of voluntary exits with the 5-category tags and the regrettable + preventable flags; build the 4-way cohort cube and look for the high-performer + 18 to 36 months tenure + manager intersection; name 3 to 5 interventions with owners and 60 to 90 day timeboxes, with the 60 and 120-day cohort follow-up in the calendar before the interventions start.

For the broader retention system, the employee offboarding guide covers the operational backbone, the stay vs exit interview comparison covers the predictive layer, the knowledge transfer template covers loss mitigation, the offboarding checklist template closes the operational loop. Or it works, or we fail. The framework gives you the discipline to know which one you are heading toward, in time to change course.