Employee Attrition: How to Measure and Reduce It

What attrition means, precisely

Attrition is the rate at which people leave an organisation over a defined period, expressed as a share of average headcount. The word is often used loosely, so it is worth separating two things that behave very differently.

Voluntary attrition is when the individual initiates the exit: a resignation to join another employer, a career change, relocation, or retirement. Involuntary attrition is when the organisation ends the role: redundancy, restructure, or dismissal for performance or conduct. The distinction is not academic. A spike in voluntary attrition among recent hires points to a hiring or onboarding problem. A spike in involuntary attrition points to a planning or performance-management problem. Averaging the two together hides both.

A further split is worth tracking. Regretted attrition covers leavers the organisation wanted to keep, usually strong performers. Non-regretted attrition covers exits that were expected or welcome. A 15 per cent attrition rate made up almost entirely of non-regretted exits is a healthy signal. The same 15 per cent concentrated among first-year high performers is a warning.

How to calculate an attrition rate

The standard formula is straightforward:

Attrition rate = (leavers in the period ÷ average headcount) × 100

Average headcount is the headcount at the start of the period plus the headcount at the end, divided by two. Using average headcount rather than a single point avoids distortion when a team grows or shrinks during the period.

A worked example

Suppose a team starts the year with 180 people and ends with 220, after both departures and new hires. Over those twelve months, 24 people left.

• Average headcount = (180 + 220) ÷ 2 = 200
• Attrition rate = (24 ÷ 200) × 100 = 12 per cent

So the annual attrition rate is 12 per cent. Now layer in the splits. If 20 of those 24 leavers resigned, voluntary attrition is 10 per cent and involuntary is 2 per cent. If 9 of the 20 resignations were people the business rated as strong and wanted to retain, regretted attrition is 4.5 per cent. These derived figures are where the real diagnosis sits. The headline 12 per cent is only the entry point.

One practical warning on the arithmetic. Pick a consistent period and a consistent definition of a leaver, and apply them every time. Teams routinely compare a figure that counted contractors against one that did not, or an annualised rate against a quarterly one, and reach conclusions that the numbers do not support.

Why attrition matters

Attrition is expensive, and the cost is wider than most budgets capture. There is the direct cost of replacing someone: advertising, agency fees, recruiter time, and the hours hiring managers spend interviewing. There is the productivity gap while a role sits open and while a replacement ramps up, which for a specialist role can run many months. There is the knowledge that walks out the door, some of which is never written down. And there is the effect on the people who stay, who absorb extra work and watch colleagues leave.

Published estimates for the cost of replacing an employee vary widely because they depend on seniority and specialism, but a common planning figure sits between a half and two times the annual salary for the role. The precise multiple matters less than the direction: attrition that looks like a routine HR metric is, in financial terms, a significant and recurring cost line.

What actually drives attrition

Attrition has many causes, and the mix differs by organisation, so treat the following as a checklist to investigate rather than a ranked universal truth. Common drivers include compensation that has drifted below market, limited progression, a manager relationship that has broken down, workload and burnout, and a mismatch between the job as advertised and the job as experienced.

That last driver deserves attention because it is decided early, often before the person even starts. When a candidate is hired against a vague or inflated role description, or selected on keyword overlap rather than real capability, the gap tends to surface within the first few months. First-year attrition is frequently a hiring-stage problem wearing an engagement-stage disguise. If you only look at exit interviews, you will attribute it to the manager or the workload. If you trace it back, a meaningful share of it was set in motion at selection.

How AI and better hiring can predict and reduce attrition

AI contributes in two distinct ways. The first is prediction: estimating which people or cohorts carry a higher probability of leaving, so a human can act early. The second, and in practice the more actionable, is prevention at the hiring stage, where fit is decided.

Predicting attrition signals

A model trained on historical data can learn associations between leaving and signals such as tenure band, time since last role change, engagement-survey trends, and patterns observed during hiring. The output is a probability for a cohort or an individual, not a verdict. Used well, it tells a manager where to have a conversation this quarter rather than waiting for a resignation letter. Used badly, it becomes a label that follows a person around and shapes decisions no one can explain.

Skills-first matching for genuine fit

Because so much early attrition is a fit problem, the highest-return intervention is often better selection. Matching candidates on demonstrated skills and capability, rather than on keyword overlap with a job advert, reduces the mismatch that drives first-year exits. GoVerse scores each candidate against the role criteria on a 0 to 100 scale and attaches a written rationale for each criterion, so a recruiter can see why a candidate fits before extending an offer. A policy engine checks the criteria for bias before a screening run executes, which keeps the selection evidence-based rather than a proxy for something it should not be. Bulk screening costs about $0.01 per CV, so testing fit across a large applicant pool is a few dollars rather than a budget line.

Candidate engagement to reduce early drop-off

Attrition starts before day one. Candidates who go quiet during a slow or opaque process disengage, and some of that disengagement carries into the first weeks of employment or ends the relationship before it begins. Keeping candidates informed across the channels they actually read, with timely and accurate updates, lowers the drop-off that quietly feeds early attrition. Consistent communication during hiring sets a realistic expectation of the role, which is itself a retention measure.

Talent pools for rehiring and continuity

Not all attrition is permanent, and not all of it is bad. Strong leavers who left on good terms, and strong candidates who were not selected this time, are both worth keeping warm. A maintained talent pool lets a team re-approach a known, pre-assessed person when a role reopens, which shortens the time a vacancy stays open and reduces the productivity cost that makes attrition expensive in the first place.

An honest limitation

AI predicts signals. Humans decide. That boundary is not a disclaimer to tuck away; it is the design principle that keeps this work defensible and fair. A model can tell you that a cohort has a higher probability of leaving. It cannot tell you why a specific person is unhappy, and it should never be the thing that quietly moves someone off a promotion list or out of a role. Those are human judgements, made with context the model does not have.

Two cautions follow. First, data quality sets the ceiling. A model trained on sparse, inconsistent, or biased historical records will reproduce those flaws confidently, and a confident wrong answer is more dangerous than an obvious gap. Before trusting any attrition prediction, check what the model was trained on and whether the labels mean what you think they mean. Second, prediction can become self-fulfilling. If a person is flagged as a flight risk and then treated as one, the flag helps cause the outcome it claimed to foresee. This is why the EU AI Act treats employment-related AI as high-risk and why explainability and human oversight are not optional extras. A score you cannot explain is a score you cannot defend, to a candidate, to an employee, or to a regulator.

A practical starting point

If you take one thing from this guide, make it the move from a single headline number to a segmented view. Calculate attrition consistently, split it into voluntary, involuntary, regretted, and non-regretted, and look at when in the lifecycle people leave. That segmentation usually points to a cause before any model does. Then use AI where it is strongest: making the hiring decision more evidence-based so fewer people arrive into a role that was never going to work, and keeping candidates and alumni engaged so the pipeline is ready when a role reopens. Keep the human in the loop for every decision about a named person, and insist on an explanation for every score. Done this way, the technology reduces attrition by improving the decisions underneath it, rather than by handing those decisions to a black box.

Frequently asked questions