TL;DR
- The decision: whether to classify every resignation as regretted or non-regretted at the moment of exit, and on whose authority.
- When to wait: when the volume is too small for any split to mean anything, and a single quarter's noise would dominate the picture.
- What must be true: the call has to be made close to the exit, by someone whose own number doesn't depend on the answer, and the record has to survive the person who made it.
- How options split: manager judgement, a rating pulled from the system, a single rehire question, a panel, or a delayed review. Each carries its own drift.
- The rule of thumb: if you can't describe, in writing, the test you applied, you've not classified the exit. You have gestured at it.
- What to expect: a turnover figure that stops averaging opposite outcomes, and a set of conversations with leadership that finally have a real subject.
The Monday Morning Call
The HR director is on a video call at 8:47 on a Monday. The chief executive has just asked, for the third quarter in a row, why the engineering turnover number is climbing. The director opens the dashboard. The number is climbing. It has been climbing for six quarters. It will probably climb again next quarter, because two of the four leavers this month were senior engineers the director personally fought to keep, and the other two were placed on a performance plan twelve weeks ago and told the company wouldn't be sad to see them go. The dashboard can't tell the difference, and neither can the chief executive, because the two outcomes have been folded into a single figure and a single line on a graph. The director is being asked to explain a number that doesn't describe anything.
The instinct is to defend the number. Add a caveat in the meeting. Say the figure is "broadly stable" and move on. But the caveat is also a confession, because the only reason a caveat is needed is that the underlying figure is hiding a story the company needs to hear. The two senior engineers who left are a failure of retention. The two people on the performance plan are a success of performance management. The single line on the graph treats both as a loss, which is wrong in opposite directions, and the wrongness compounds every quarter that the split isn't made.
The issue isn't the turnover figure. The issue is the missing classification that should sit underneath it, the call about which exits the company tried to stop and which ones it didn't, made close to the moment of the resignation, by someone whose own number isn't changed by the answer. Every other question in this article follows from that one.
Best tools for Employee Retention
When You Genuinely Do Not Need to Do This Yet
You're a team of forty with four leavers a year. A team this small has no statistical object called "attrition" in any meaningful sense. The number is too thin to split without the split being louder than the underlying reality. Four departures a year is four conversations, four handovers, four sets of reasons to listen to. If you're reporting a single percentage to anyone outside the company, the figure is so noisy that nobody should be making a decision on it. Your job is to listen to the four exits well, write down what you heard, and resist anyone who wants a trend line from a sample of four. Adding a regretted split on top of a four-leaver year is theatre, not analysis.
You're a team of one hundred and fifty, and your leadership doesn't yet ask about the breakdown. The split is only worth the friction of doing it when somebody is going to act on the difference. If the only consumer of the turnover number is a board slide that nobody questions, you're paying the cost of classification for no return. You will still want the conversation skills. You will still want the notes. You won't yet want the formal category, because formal categories get gamed, and a gamed category in a company that doesn't act on it's worse than no category. Watch for the moment a leader asks a sharper question. That's when the work starts.
You're a company of six hundred and your leavers are concentrated in one function. Concentrated attrition is its own diagnostic. If all your leavers come from one team, the analysis is about that team, and the people closest to it are the right people to make the call. The risk is that a "regretted" label in that one team gets used as a weapon in a way the label wouldn't be used across a balanced population. So the work here's partly the split, and partly protecting the people in the labelled team from the political consequences of the label. Both need to be in place before the number goes up a level.
You're a multinational and the works council has a view. In some countries, the people representative body has a consultation right over how employees are assessed, and a regretted or non-regretted judgement is a personal data record about a living individual that's usually disclosable to that individual on request. If you've not checked the local position, the cost of getting it wrong isn't theoretical. The local rules differ, and you need to confirm them in each country before you formalise a category that someone is going to be asked to write down. This is the stage at which the work looks like a legal review more than a process design. That's the right shape for it.
Five Questions You Are Asking Yourself at 11pm
Why does the number feel wrong to me when leadership seems happy with it? Because the number is hiding a story you can see from inside the building and they can't see from the dashboard. The reason leadership is happy is that the number is a comfort. The reason it feels wrong to you is that you know what is behind it. The risk of doing nothing is that one of those hidden stories breaks into the open on a day you didn't choose, and the conversation you've on that day is the one you've lost the ability to shape.
Who in this company is incentivised to make this call honestly? Almost nobody. The manager whose team lost a person is incentivised to call it non-regretted, because a regretted exit is an implicit accusation against the manager. The HR business partner is incentivised to call it regretted, because a regretted exit is proof the function is doing work. The skip-level is incentivised to match the room. So the question isn't "who has the information" but "who has an interest in the answer that doesn't bend the answer in a known direction". If you can't find such a person, you don't yet have a classification. You have a negotiation.
What is the worst thing that happens if the leaver sees what we wrote about them? This is the right question to ask, because the answer is usually "they can ask for it". Notes taken about a named individual in an exit conversation are personal data about a living individual in most places, and that data is usually disclosable to the individual on request. The wording in the record has to be the wording you would be comfortable showing the leaver across a table, in daylight, with a colleague present. If you wouldn't write the sentence in the file, don't let the sentence be the basis for a classification either.
Do I have any way to check whether the calls are drifting? You don't, yet, because there's no second opinion built into the process. The simplest check is a quarterly pull of every classification, names removed, sorted by the manager who made the call, and a look at the distribution. If one manager has never once recorded a regretted exit across two years, that's information. If the director's office always records non-regretted, that's also information. You're looking for the shape of the drift, not for one bad actor, and the shape of the drift is the answer to whether the system is working.
What is the smallest change I could make on Monday that wouldn't be reversible? This is the practical question, and the answer is usually the written definition. You don't need a system. You don't need a workflow. You need a one-page document that says what regretted and non-regretted mean in your company, who decides, and what evidence the decision has to point to. You can hand that document to your HR business partners on Monday. You can ask them, by Friday, to classify the next three exits against it. That's a change that's not yet a commitment, and the not-yet part is what makes it survivable.
The Three Honest Categories This Splits Into
Manager judgement, written down. The line manager makes the call at the resignation conversation and writes it on a form. This is the lightest possible process, and for many companies it's the right starting point, because the line manager usually has the most information and the most context for the decision. It fails in a known way: the manager's own performance review is affected by the leavers in their team, and that creates a quiet pressure to call exits non-regretted. If you adopt this approach, you need a periodic review of the distribution by manager, not because any one manager is dishonest, but because the drift is structural and the structure is the thing you can fix. The failure mode is a steady softening of the category over two or three years, until "regretted" means "left on bad terms" and nothing else.
A single rehire question. The manager is asked, on a form, "would you rehire this person into a comparable role in the next twelve months, knowing what you know today?". This is a useful discipline because it forces the manager to imagine a counterfactual rather than describe a feeling. It works in organisations where the manager has a clear picture of the future hiring plan. It fails where the manager doesn't, because the question gets answered as a verdict on the person rather than a forecast about the role. The honest weakness is that a manager can answer "no" for a top performer whose team is being restructured, and "yes" for a steady player whose team is expanding, and the answer stops tracking what you actually wanted to know.
A second pair of eyes. A panel, a calibration session, or a quarterly review by a senior leader who was not the line manager. The point is to introduce a counterweight to the structural incentives on the manager. The honest weakness is that the panel needs a definition, and if the definition is loose, the panel is two loose judgements meeting in a room and producing a more confident loose judgement. The other honest weakness is that panels slow the decision, and a regretted or non-regretted judgement is most useful close to the exit, when the handover is still being designed. If the panel meets at quarter end, you've lost three months of usefulness. The right panel meets weekly, sees a small number of cases, and writes a sentence for each.
Five Diagnostic Questions You Can Self-Assess Against
Can you write, in one sentence, what regretted attrition means in your company? If you can't, you don't have a category. You have a word. To answer this, sit down for ten minutes and try. If the sentence has more than two clauses, it's too loose to apply consistently. If it has a hedge in it, it's too loose to be useful. The test of the sentence is that two different managers, reading it cold, would classify the same exit the same way.
Who, by name, has the authority to record a regretted exit against a person? If the answer is "the manager", the answer is also "the person whose number the exit affects". That's not a defect, it's a fact, and you need to know it. To answer honestly, look at your last ten exit forms. Who signed them. Whose manager signed them. Whether the two signatures are from people with different incentives.
What is the distribution of regretted calls across your managers in the last twelve months? If you don't have this, you don't have a diagnostic. To build it, pull the forms, sort by the manager who recorded the call, and look at the rate. You're not looking for a single bad actor. You're looking for the shape. A team where every manager has roughly the same rate is a team with a working process. A team where one manager has zero regretted exits in two years and a neighbour has one in three is a team with a structural problem in the category, not in the people.
What changes in the next conversation if you can answer the regretted question with a number? This is the value test. If the answer is "nothing", the split is decoration. If the answer is "we can show the board whether we are losing the people we tried to keep", the split is doing the work. To answer this, write down the next three conversations you expect to have about turnover and ask, of each, whether the split would change the decision or only the rhetoric.
What is the cost of a wrong call? Wrong in which direction. Calling a regretted exit non-regretted means a real failure gets filed as a success. Calling a non-regretted exit regretted means a person the company was glad to see go is, on paper, a loss. Both are bad, and they're bad in different ways, and the shape of the badness should change how you design the form. If you can't describe the cost of a wrong call, the call isn't yet load-bearing.
Five Ways to Make the Call
Manager Designation at the Time of Resignation
The line manager makes the call in the resignation meeting, or within forty-eight hours of it, and writes it on a short form. The form has a one-sentence definition of regretted and non-regretted at the top, a single yes-or-no question, a free-text reason, and a signature. The form goes to HR within five working days. This is the most common starting point, and the reason it's common is that it's the lightest process that still produces a written record.
It earns a place because the manager is the person with the most information, the conversation is fresh, and the cost of doing the work is low. The handover, the reference, the access revocation, and the knowledge transfer are all happening at the same moment, and the classification is part of the same package.
It falls short in a specific way. The manager's own number is affected by the leavers in their team, and that's a structural pressure, not a moral one. A good manager can hold the line for a while. Over a year, across a hundred managers, the pressure shows up as a steady drift toward the flattering answer. The only honest fix is a periodic review of the distribution by manager, with the distribution shown to the manager's manager, and a written expectation that zero regretted exits across a long period is itself a signal worth investigating. Without that, this approach becomes the path of least resistance, which is exactly the wrong path for this decision.
Performance Rating as the Classification
You take the most recent performance rating on file and use it as the classifier. A high rating plus a resignation is regretted. A low rating plus a resignation is non-regretted. The data already exists, the system can do the join, and the result is reproducible.
It earns a place because it removes the judgement call from the moment of exit. The manager isn't being asked to make a decision in a room with a person in front of them. The system is making the call, and the system isn't subject to the same structural pressure. For a company that has a trusted performance process, this can be the cleanest of the five.
It falls short because performance ratings aren't the same as loss to the business. A high performer in a function the company is shrinking is a different problem from a high performer in a function the company is growing. A high performer on a personal development plan is a different problem again. The rating captures the past. The regretted question is about the future, and the future depends on the role, the plan, and the team, none of which are in the rating. Used alone, this approach will misclassify every exit that happens at the intersection of a strong person and a shrinking team, and that intersection is where the most expensive mistakes are made.
The Would-You-Rehire Question
The manager is asked, on the form, whether they would rehire this person into a comparable role in the next twelve months, knowing what they know today. The answer is the classification. Yes means non-regretted. No means regretted. The form is signed and filed.
It earns a place because it converts a feeling into a counterfactual. The manager is no longer saying "I am sad they left" or "I am not sad they left". They're saying whether, given the choice, they would do the deal again. That's a sharper question, and sharper questions produce more consistent answers.
It falls short where the manager doesn't have a clear picture of the next twelve months of hiring. In a frozen requisition environment, the question becomes a verdict on the person rather than a forecast about the role. The honest weakness is that managers answer it consistently when the business is hiring and inconsistently when it isn't, and a classification that changes with the hiring plan is a classification that's not really about the leaver at all. The fix is to remind the manager, on the form, that the question is about the role and the team, not the person in isolation.
A Calibration Panel
A small group, usually the HR business partner, the skip-level leader, and one other senior manager, meets weekly to review the last few exits against the written definition. The panel either confirms or revises the manager's call, and the panel's decision is the one that goes on the record. The meeting takes thirty minutes and covers four to six cases.
It earns a place because it's the only one of the five that builds a second opinion into the process by design. The structural pressure on the manager is balanced, in the room, by a countervailing view, and the definition gets sharpened each time the panel has to apply it. Over a year, the panel becomes the institutional memory of what the category means in this company, which is the thing the category most needs.
It falls short on speed. The classification is most useful at the moment of the handover, and the panel meets on a cadence. The fix is to run the panel weekly, to limit the cases per meeting, and to give the manager provisional authority to act on their own call between meetings, with the panel's ratification coming within seven days. Without that, the panel is a quality check on history, not a tool for the present.
After-the-Fact Review at Quarter End
The exits aren't classified at the time. The data is collected. At the end of the quarter, the HR business partner and the relevant senior leader sit down, look at each exit, and assign a classification with the benefit of hindsight and of any information that emerged during the notice period.
It earns a place because some information is only available after the exit. A manager who was ambivalent at the resignation meeting can become clear in the handover, and a person who left gracefully can leave badly. The after-the-fact review captures both.
It falls short on the most important thing, which is influence over the handover. The classification made at quarter end doesn't change the reference, the access, the knowledge transfer, or the project plan. It's a label on a record, not a tool for the moment it could have been useful. The honest weakness is that this approach produces a clean dataset and a useless process, because the dataset is the part that gets reported and the process is the part that was supposed to matter. Used alone, this approach gives you a number you can defend and no ability to do anything with it.
The Decision Table
| Situation | Scale | Setup | Primary Pain | Recommended Starting Point |
|---|---|---|---|---|
| Single site, stable headcount, low leaver volume | Under 150 | One HR generalist, no formal process | A handful of exits a year, no time to build a system | Manager designation with a written one-sentence definition; review the distribution once a year |
| Multi-team company with active hiring | 150 to 1,000 | Several HR business partners, an HRIS in place | Leadership is reading a single number and acting on it | Would-you-rehire question on a short form, with a quarterly distribution review by HR |
| Function-heavy organisation with conflicting manager incentives | 1,000 to 5,000 | Centres of excellence, formal performance process | The split is being made by the people whose numbers it changes | Calibration panel meeting weekly, with the manager's call as the starting point |
| Multinational with works council presence | Any size, multiple countries | Country-level HR leads, local legal advisers | Local rules on personal data and consultation vary, and the record has to be defensible in each | Manager designation, written definition, and a documented local review before the form goes live in any country |
| Company with a strong and trusted performance process | 500 to 5,000 | Calibrated ratings, performance as a system | Managers are unreliable classifiers, and the data is already in the system | Performance-rating-based classification as a starting filter, with a manager override that requires a written reason |
| Restructuring organisation with high exit volume | 1,000+ | A dedicated people analytics function, a transformation office | The same quarter contains both the people you wanted to keep and the people you are losing deliberately | Calibration panel for the ambiguous middle, performance rating for the clear cases, written definition signed off by the transformation lead |
| Public sector or regulated employer with strict documentation rules | Any | Records management, audit, freedom of information in some places | The classification will be visible to people outside the company, and possibly to the leaver | Manager designation with a documented local review, the narrowest possible definition, and an audit trail for every call |
| Early-stage company, no HRIS, no formal performance process | Under 100 | Founders doing HR between other things | Every process is a tax on a small team, and the wrong process is worse than no process | Do not formalise yet. Keep notes from the exit conversation. Review the notes once a year. Wait for the volume. |
Writing a Definition People Will Apply Consistently
The shortest definition that survives across managers and departments is a sentence that names the test. "We classify an exit as regretted if, knowing what we knew on the day of resignation, we would have made a counter-offer the company could afford to retain the person in a comparable role." That sentence isn't elegant, but it's testable. A manager reading it can place their own case against it. Two managers reading it can compare cases. The sentence does the work because it forces the question out of the feeling and into the forecast.
The drift is in the words "comparable role" and "could afford". "Comparable role" drifts toward the role the company happens to be hiring for, which can change with the quarter. "Could afford" drifts toward what the manager wishes the budget allowed, which is a different number from what the budget actually allows. Both drifts push the classification toward the answer the manager wants to give, which is the structural problem the definition is supposed to fix.
| Approach | Who decides | Where it drifts |
|---|---|---|
| Manager judgement at resignation | Line manager | Toward the flattering answer, over time, as the manager's own number absorbs the leavers |
| Performance-rating-based | System, with manager override | Toward whatever the rating captured, which is the past, not the future of the role |
| Would-you-rehire question | Line manager, on a form | Toward the verdict on the person when the hiring plan is unclear |
| Calibration panel | Panel, with the manager's call as the starting point | Toward the dominant view in the room, if the panel is not diverse enough |
| After-the-fact review | HR business partner and senior leader | Toward the cleanest dataset and the most useful present, with a loss of influence over the handover |
The definition belongs on the form, not in a separate document. A separate document is read once and forgotten. A definition on the form is read every time the form is filled in, which is the moment the decision is made. The form is the training.
What Changes Once You Split It
The single number becomes two numbers, and the two numbers describe different things. The regretted number is a measure of how well the company holds the people it wanted to hold. The non-regretted number is a measure of how well the company manages out the people it needed to manage out. Reporting them separately isn't a small change. It changes the conversation in the room, because a leader can't wave at a single line on a graph and say "turnover is up" without also looking at which line moved. The two numbers move for different reasons and the leader has to think about which one moved before deciding what to do.
| Reporting change | What it makes possible | What it makes harder |
|---|---|---|
| Regretted and non-regretted reported on separate lines | A retention conversation that is about retention, not performance management in disguise | Dodging the conversation, because the split is visible |
| Regretted rate by team, with the distribution shown to senior leaders | An honest look at where the company is losing the people it tried to keep | Hiding a bad team behind an average |
| A trailing twelve-month view of the regretted rate, by function | A signal that survives quarter-to-quarter noise | Optimising for a single number on a dashboard |
| Non-regretted rate reported alongside time-to-competency of the replacement | A view of whether the exits the company wanted are actually improving the team | A neat story when the answer is that the replacement is no better |
| Regretted rate of new hires within twelve months | A signal about the hiring process that no single number can show | Hiring for the metric, by adjusting the bar in the wrong direction |
The decisions that become possible are mostly decisions that were already being made on instinct, in corridors, without a number to point to. The split puts the number under the decision, which is what numbers are for. The decisions that get harder are the ones that depended on the single number being a comfort. Those are the decisions that need to get harder.
What to Put in Writing
A classification that's not in writing is a conversation, and conversations don't survive the people who had them. The minimum set of artefacts is small and the work to produce it isn't, which is why most companies skip it and then discover, two years later, that they can't answer a basic question about their own process.
| Artefact | Who owns it | When it is written | What it prevents |
|---|---|---|---|
| Written definition of regretted and non-regretted | HR lead | Before the first classification, reviewed annually | Two managers classifying the same exit differently and both being right by their own lights |
| The classification form | HR operations | At the resignation conversation, within five working days | A classification made on instinct, by memory, three months after the fact |
| The manager's written reason | Line manager | At the form, in one or two sentences | A label on a record with no basis that can be defended |
| The HR business partner's counter-signature | HR business partner | Within five working days of the form | A classification that drifts because the structural pressure on the manager was not balanced |
| The quarterly distribution review | HR analytics or HR lead | At quarter end | A steady softening of the category that nobody notices until the number is meaningless |
| The local legal review, by country | Country HR lead, with local advice | Before the form is used in a new country | A record that is not disclosable in the way local rules require |
| The retention conversation log | HR business partner | At the same time as the form, if a counter-offer was considered | A regretted exit that the company could have kept, with no record that the company knew it |
| The reference decision | Line manager, with HR sign-off | Before the reference is given | A non-regretted exit whose reference is written as if the company was sorry to lose them |
The artefacts are small. The discipline of producing them, every time, is the work. A team that does the work for a year will have a dataset that's honest. A team that doesn't will have a number on a slide, and the slide will be wrong, and nobody will be able to say how they know.
Questions to Ask Before You Commit
The definition. "In one sentence, what is regretted attrition in this company?" A bad answer is a paragraph. A bad answer is also a sentence with three clauses and a hedge. The right answer is a test the manager can apply to a specific person on a specific day.
The authority. "Who, by name, has the authority to record a regretted exit against a person?" A bad answer is "the manager" with no second pair of eyes. A bad answer is also "the panel" with no provisional authority between meetings. The right answer names the role and the cadence.
The record. "What is on the form, in what order, and who signs it?" A bad answer is "we've a process" with no form. A bad answer is also a form that asks for ten data fields and the manager's reason is a single line at the bottom. The right answer is a short form, read top to bottom, that produces a record the company can show.
The data. "Where does the classification live, who can change it, and what is the audit trail?" A bad answer is "in the HRIS" with no audit trail. A bad answer is also "in a spreadsheet" with no owner. The right answer names the system, the owner, and the rule about changes.
The drift. "How will you know if the classification is drifting toward the flattering answer?" A bad answer is "we will check it". A bad answer is also "we've a dashboard". The right answer is a quarterly distribution by manager, looked at by name, with an expectation that the shape of the distribution is a conversation.
The employee. "What does the leaver see, and what can they ask for?" A bad answer is "nothing". A bad answer is also "we will deal with it if they ask". The right answer is a written statement of what is on the record, in language the leaver could read, and a process for the question they may ask.
The local rule. "What is the position in each country where this form will be used?" A bad answer is "we are standardising globally". A bad answer is also "we will check when we get there". The right answer is a documented review, by country, before the form is used in that country. Confirm the position locally before you commit.
The political shape. "Who in the leadership team benefits from a particular answer, and have you accounted for that?" A bad answer is "we are neutral". A bad answer is also "the CEO wants to see this number go down". The right answer names the pressure and the design choice that responds to it.
The exit. "What is the smallest change you can make on Monday that's not yet a commitment?" A bad answer is a six-month programme. The right answer is a one-page document, three exits classified against it, and a decision in eight weeks about whether to keep going.
The Cost of Getting This Wrong
The first cost is the conversation you've with leadership about a number that's hiding a story. You will defend the number. You will add caveats. You won't say, out loud, that the number is the average of a failure and a success, because saying that out loud is admitting the company can't tell them apart. So you'll have the conversation in code, and the code won't survive the next resignation, and the next resignation will be the one the chief executive noticed because it was a senior person. The conversation you've then is the conversation you lost the ability to shape by not doing the work earlier.
The second cost is the manager who quietly stops classifying exits as regretted because the label is a quiet accusation against them. The manager doesn't stop losing people. The manager stops writing down that the people they lost were the people they wanted to keep. So the company keeps losing the people the manager wanted to keep, and the only record is a single line on a graph that says "attrition is stable". The drift takes a year to show up. It takes another year to notice. It takes a third year to undo, if anyone is still paying attention.
So the question isn't whether you can afford to make the call. The question is whether you can afford to keep reporting a number that's the average of two opposite outcomes, in a company where the difference between the two outcomes is the only thing the leadership team actually needs to know.
When You Are Ready to Go Further
If the work above has surfaced a question you can't answer with what is already in the building, the next step is to look at how other companies in your situation have approached it. HROpsLab is a review publication, not a vendor. We don't sell software, payroll services, or advice. What we do is independent comparison work across the people operations stack, written for HR and people analytics leads who are tired of borrowed numbers and vendor pitches.
The comparison is most useful when you've a specific question, not a general one. If you've a question, the comparison work is the place to start, and the case studies show how other teams have approached the same shape of problem with different starting constraints. If you would rather talk it through with someone who has read the same research and looked at the same vendors, the expert line is open.
The work is free to read, and the conclusions are written so you can disagree with them in public, which is the only useful kind of independent.
Frequently Asked Questions
What does regretted attrition actually mean?
Regretted attrition is the subset of resignations where, with the information available at the time, the company would have preferred to keep the person in a comparable role. Non-regretted attrition is the rest. The point of the split isn't the words. The point is that a single turnover figure is the average of two opposite outcomes, and a leader looking at the figure can't tell which outcome is moving. Once the split is made, the regretted number is a measure of how well the company holds the people it wanted to hold, and the non-regretted number is a measure of how well the company manages out the people it needed to manage out. The two numbers move for different reasons and call for different responses, and the single figure hides both.
Who should make the call?
The line manager has the most information and should be the starting point for the call, but the call should not end with the line manager, because the manager's own number is affected by the leavers in the team. A second pair of eyes, whether a panel, a calibration session, or a quarterly review by a senior leader who was not the line manager, is the structural counterweight. The exact shape of the second pair of eyes is a design choice. The non-negotiable is that the person whose number the exit affects isn't the only person whose judgement is on the record.
Should the leaver be told how they were classified?
In most organisations, exit interviews are voluntary, and notes taken about a named individual are personal data about a living individual that's usually disclosable to that individual on request. The classification, if it's in the file in a form that identifies the leaver, is on the same footing. The practical answer is to write the record in language you would be comfortable showing the leaver across a table, in daylight, with a colleague present, and to confirm the local position. The instinct to keep the classification off the record entirely is tempting, and it produces a dataset nobody can use and a process nobody can defend.
How do you stop managers from gaming the classification?
You build a quarterly review of the distribution by manager into the process, with the distribution shown to the manager's manager. You're not looking for a single bad actor. You're looking at the shape. A manager who has never recorded a regretted exit across two years is information, and so is a manager whose regretted rate is exactly the company average every quarter. The drift is structural, not moral, and the structural drift is what the quarterly review is designed to surface. The other discipline is to keep the written definition short and on the form, so the manager is applying the same test each time.
Should you use performance ratings as the classification?
Performance ratings are useful as one input, not as the whole answer. A high rating plus a resignation isn't the same as a regretted exit if the function is shrinking. A low rating plus a resignation isn't the same as a non-regretted exit if the person was on a development plan. The rating captures the past. The regretted question is about the future, and the future depends on the role and the team, which aren't in the rating. Used alone, the rating approach will misclassify every exit at the intersection of a strong person and a shrinking team, which is where the most expensive mistakes are made.
What do you do with the cases that could go either way?
You route them to the second pair of eyes, and you write down the reason. The honest answer is that some exits are genuinely ambiguous, and the right response to ambiguity is to record the ambiguity, not to resolve it by force. A classification of "regretted, conditional on the next quarter's hiring plan" is more useful than a clean binary made on instinct. The dataset will be messier, and the mess is the truth, and the leadership conversation is easier with the truth than with a tidy number that doesn't describe anything.
How do you report both numbers without making the conversation more complicated?
You report the regretted rate and the non-regretted rate on separate lines, with the same denominator. You put the regretted rate above the non-regretted rate, because the regretted rate is the one that describes a failure the company can act on. You don't add a third line called "net attrition" that combines the two back into a single number, because the whole point of the split is that the two numbers should not be combined. The conversation is more complicated for the first two quarters and less complicated for every quarter after that, because the leader is finally being asked to think about which line moved before deciding what to do.
HROpsLab is an independent review publication for HR and people operations. We don't sell software, payroll, or advice, and we never will. Our work is free to read, and our conclusions are written so you can disagree with them in public.
