TL;DR
- Core decision: You're choosing between depth (what you didn't know to ask) and comparability (what you can count across hundreds of leavers). One instrument can't deliver both.
- When doing nothing is right: When fewer than around twenty people a year leave and you already talk to each one, an extra process adds friction without insight.
- What has to be true: Your leavers must trust that honest answers are safe to give, the person reading the data must have permission to act on it, and the format you pick must match how many people you're trying to reach.
- How the options split: Conversations surface the unexpected; forms produce a record you can compare; combined approaches buy reach but cost time; delayed surveys catch what people were too polite to say on the day; manager-only notes catch nothing systemic.
- Decision rule: Pick the lowest-friction instrument that still gets you the answer you would actually change something in response to. Then commit to acting on what comes back.
- Outcome to expect: A defensible record, fewer repeat leavers from the same root cause, and a process your managers don't quietly stop doing.
When the Resourcing Question Becomes a Data Question
A head of people operations at a mid-sized SaaS company sits down on a Tuesday morning with a problem that isn't quite what she thought it was. Her leaver count has crept up to roughly one a week, sometimes two. Last quarter she ran twelve exit interviews. This quarter she's already booked seven and has eleven more names on the list and no slots left before quarter end. She looks at the calendar, then at the form a previous HR manager built two years ago and nobody updated. She has half an hour to decide whether she switches to a survey, hires a temp to do the conversations, or just lets the queue pile up. Her instinct is that this is a resourcing problem: she doesn't have enough hours.
But it isn't. The real issue isn't whether she has more time. It's whether the thing she's collecting is worth the time it costs to collect, and whether the thing she's currently skipping is the thing that would have told her something she didn't already know.
The exit process is the one conversation in HR where the person you most want honest answers from has the least reason to give them. They've already mentally resigned. They've often started a new job. They may want a reference. The format you choose sets the temperature of the room before anyone has opened their mouth or filled in a field, and a badly chosen format gives you either a high completion rate on a form nobody took seriously or a small set of conversations that happen too late to be useful. Both outcomes feel productive. Neither is.
Best tools for Employee Retention
When You Can Honestly Do Nothing Yet
There are stages here, and pretending otherwise will only make the rest of this piece read like a pitch. Some setups are genuinely fine as they're.
Stage one: genuinely fine. Fewer than around twenty leavers a year, a People team small enough that the HR lead knows each departure personally, and managers who already have a working relationship with their direct reports. In a forty-person agency with three or four leavers a year, the head of people has probably had the conversation already, in the corridor, in a one-to-one, in the resignation letter itself. Writing a formal process on top of that's process theatre. The risk is that someone in audit reads about best practice and imposes one anyway.
Stage two: friction. The leaver count has roughly doubled and the HR lead is now missing about a third of conversations, not because they're rude but because calendars won't align. The manager is doing them instead, which is fine until it isn't, because managers vary wildly in what they ask and what they write down. At this point the choice is whether to formalise what the manager is already doing or to bring it back in-house with a tighter template. Both are defensible. Doing nothing quietly means that a year from now you've no record you can compare one quarter against the next.
Stage three: real risk. The volume is high enough that you can't reasonably talk to everyone, and the consequences of a single bad departure are high enough that you can't afford to miss one. A two-hundred-person operations team with seasonal churn, a regulated employer where each leaver takes process knowledge with them, an engineering org where one departure stalls a project. The cost of missing the systemic reason people are leaving is now higher than the cost of running a real process.
Stage four: the edge case. An organisation in active restructuring, with severance discussions already underway, where leavers are partly leaving because of decisions the company has made and partly because of decisions that haven't been communicated yet. In this setup, anything that looks like a survey is going to read as tone-deaf, and anything that looks like an interview is going to be treated as evidence in a dispute. The honest move here's often to pause formal data collection entirely and route what would have been the exit conversation through the legal channel that's already running.
If you're not in any of those four stages yet, this article is useful background. If you're in stage three and pretending you're in stage two, you'll know by the end of the next section.
The Five Questions You're Asking Yourself at 11pm
These are the ones that actually keep an HR lead up, not the abstract version.
"Do I trust the data I'll get?" Probably not, in the form you'd like to. The honest version: people are nicer on the way out than they were on the way in, and anything they write will be filtered through who might read it later. Reasoning: trust is set by who is in the room and what is in writing. If the manager who is the problem is also the person reading the survey, the answers will be shorter and sweeter than reality. Getting it wrong: you make a retention decision on data that flatters the company.
"Should I just hire someone to do the conversations?" Maybe. If your bottleneck is calendar slots and the new person's value is opening those slots, yes. If you're hiring them so that the conversations still happen but you personally never see the answers, you've bought yourself plausible deniability rather than insight. Getting it wrong: you spend money and the systemic signal still doesn't reach anyone with authority to act.
"What if I just send a form and call it done?" You'll get a respectable completion rate and a thin pile of data. The form will tell you what people were willing to type into boxes you already built. It won't tell you what you didn't know to ask. If your goal is to be able to say "we ran an exit process this quarter" to an auditor or a board, the form is enough. If your goal is to learn something, it isn't.
"Is it unfair to ask after they've left?" Possibly, depending on the data protection rules where you operate, and depending on whether they agreed to be contacted. The exposure shape: a survey sent after employment ends is a fresh data processing activity that needs its own lawful basis in most jurisdictions. Getting it wrong: you ask, they ask a regulator, and you discover the answer was a privacy question you hadn't considered.
"Can I just do both?" Yes, and many organisations do, but the combination reliably delivers less than either alone if it isn't designed deliberately. A long form that ends with an optional conversation gets the form and not the conversation. A short form followed by a selected interview gets both, but only for the people you've selected, and the selection criterion is itself a judgment you have to defend.
The Three Honest Categories
There are really three categories here, and most of the format names you'll read about are variations on them.
The conversation. A live, semi-structured exchange, usually forty-five to ninety minutes, with a person the leaver has not worked for day to day. What it's good at: surfacing the thing you didn't know to ask about. The interviewer can follow a thread. They can ask why, then why again. They can hear what someone almost says. Where it fails: it doesn't scale, it depends entirely on the skill of the interviewer, and it produces anecdotes, not data you can roll up across the company. A real example of it failing: a People director conducts every exit conversation personally, writes long notes in a private document, and never aggregates them. Six months later, the same theme has appeared in seven conversations and the company has no record of having heard it once.
The written form. A structured set of questions, usually a mix of scale and free text, sent before or on the last day. What it's good at: producing comparable data across a large number of leavers, and producing it in a form that doesn't require an interviewer to be present. Where it fails: it gets a higher completion rate than the conversation gets a booking rate, but a completed form with three words in every free-text box isn't data. It tells you what you asked. It tells you nothing you didn't ask. The honest version of when it fails: the answers to the open-ended questions collapse into polite generalities, because the leaver is on their way out and wants a clean reference, not a confession.
The combined approach. Either a form that flags certain answers for follow-up conversation, or a short conversation that ends with a written survey. What it's good at: getting reach from the form and depth from the conversation, for the cases that warrant it. Where it fails: the selection rule. If "warrants a conversation" means "scored low on satisfaction", you'll end up talking only to the people who were already going to tell you something. If it means "managers flagged as high risk", you'll end up confirming what your managers already believed. The combined approach works only when the trigger for the second instrument is set deliberately and written down before anyone starts collecting.
Five Diagnostic Questions You Can Answer Today
These are the ones to put to yourself before you put them to anyone else, and the point is that the answer matters more than the question.
One: what would you actually change in response to what you receive? If you can't point to a specific decision that a piece of exit data would feed into, you don't need more exit data. The honest answer is usually one of three: a manager has to hear it, a policy has to be reviewed, a compensation band has to be checked. If none of those would change, the survey is decoration.
Two: who has authority to act on what comes back? A People team that gathers signal nobody else can act on is a people team that will eventually be told to stop gathering signal. To answer this, name the person, not the team. The HR lead, the line manager, the chief of staff, the CEO. If the answer is "everyone", it's no one.
Three: how many leavers do you expect in the next twelve months, by quarter? Not how many you had last year. Forward, with a range. To actually answer this, ask the hiring and the restructuring plan what it implies for attrition. The number drives the format, not the other way round.
Four: what did you learn from the last ten leavers, written down somewhere a colleague could read? If you can't produce that artefact, you're not running a process. You're having conversations and losing them. If you can produce it, you already have a baseline, and the question is whether the format you're using now would have surfaced the same things.
Five: if the leaver asked to see their exit record in a year's time, what would you show them? Exit notes are personal data about a living individual in most jurisdictions, and are usually disclosable to that individual on request. The honest test is whether you'd be comfortable handing over the document. If the answer is no, the format you're using is the wrong one.
Five Formats, Reviewed
A Live Conversation
A scheduled, semi-structured exchange between the leaver and a person who has not been their day-to-day manager. Typically forty-five to ninety minutes, on or near the last day, with notes taken during or immediately after.
It earns a place because it's the only format that can follow a thread. If the leaver says something half-formed, the interviewer can ask what they mean. If the leaver hedges, the interviewer can notice. The conversation is also the only format where you can read tone, which sounds soft until you remember that "I'm fine with the manager" said flatly and "I'm fine with the manager" said with relief are different data points. For senior or hard-to-replace leavers, or for situations where something has clearly gone wrong, the conversation is the right tool.
Where it falls short is scale and consistency. One interviewer asks different questions than the next. One writes down what the other paraphrases. The output is anecdotes that have to be re-read, re-coded and re-argued before anyone can act on them. A conversation is also high-friction for the leaver, which means the people who most need to be heard are the most likely to skip it, because booking a slot is harder than ignoring an email. The format also concentrates risk: a single bad interviewer can collect data that damages the leaver, the manager or the company, and there's usually no second pair of eyes on the notes.
A Written Form
A structured set of questions, a mix of scale items and free-text boxes, sent by email or completed in a tool before or on the last day.
It earns a place because it scales. A form gets a higher completion rate than a conversation gets a booking rate, because the friction for the leaver is much lower. It also produces comparable data. You can put a hundred answers side by side and see the same question answered in the same way by each person. For high-volume leaver populations, seasonal operations, large graduate intakes, or any setup where the volume makes conversation impossible, the form is the only format that lets you read across the whole population at all.
Where it falls short is precisely the thing the conversation was good at. A form asks what you thought to ask. If your leavers are leaving for reasons you haven't put into the survey, the form will tell you they left for the reasons you did put in. A long form with many open boxes looks thorough and reliably delivers three words in each one, because the leaver is tired, has somewhere else to be, and isn't going to write a paragraph in a field labelled "any other feedback". A form that arrives on the last day is also read in the room with the manager still visible, which depresses honesty. And a form that's not anonymous is read as a record the leaver will be judged on later, which depresses honesty even further.
A Form Followed by a Conversation for Selected Leavers
A short, structured form sent to every leaver, with a written rule about who gets a follow-up conversation and why. The rule is the instrument, not the conversation.
It earns a place because it gives you reach and depth at the same time. Every leaver has the friction of a short form. A defined subset gets the depth of a conversation. You can route by score, by role, by reason for leaving, or by a flag the manager sets. For mid-volume organisations that can't afford a conversation for everyone but can't afford to miss the conversations that do, this is the format that actually fits.
Where it falls short is the selection rule. If the rule is "anyone who scores low on satisfaction", you'll only catch the people who were willing to score low. If the rule is "anyone flagged by the manager", the manager is selecting which departures get investigated, which they will use, consciously or not, to suppress the ones that reflect on them. The format also tends to bloat. Two years in, the form has thirty questions, the conversation has a checklist, and the selection rule is whatever the latest HR lead remembers to apply. The format works only when the rule is written down, reviewed, and the volume of follow-up conversations is bounded so that the HR lead can actually do them.
A Delayed Survey Sent Weeks After Departure
A survey sent a few weeks after the leaver's last day, usually by email to a personal address, asking about the experience after they have had time to settle.
It earns a place because it changes the temperature of the answers. A person who has already started a new job, who no longer needs a reference from you, who isn't sitting across from their manager, will often say things they wouldn't have said on the last day. For organisations where the last-day conversation is compromised by the relationship still being live, this is the format that gets you the version of the answer the leaver would have given a friend. It also lets you ask questions you couldn't ask on day one, like whether they would consider returning.
Where it falls short is significant. A delayed survey is a fresh data processing activity that needs its own lawful basis in most jurisdictions, and the leaver's consent to be contacted after employment ends isn't automatic. Contactability falls off quickly: people change addresses, lose interest, forget who you're. The data protection shape of the exercise is genuinely different from an in-employment survey, and the rules vary by where you and the leaver are. The format also costs you the timeliness. If a manager is the problem and three other people are about to leave for the same reason, a delayed survey catches it after the next three have already gone. And because the leaver is no longer in the building, you can't follow up on a concerning answer with a real conversation.
No Formal Process, With Manager Notes Only
The leaver has an exit conversation with their line manager, the manager takes notes, and there's no other process.
It earns a place because it's the format that most organisations actually run, whether or not they admit it. For small teams with stable, well-managed groups, where the manager and the leaver have a trusting relationship and the manager has the time and the skill to run the conversation well, this is fine. The cost is zero. The friction is zero. The notes are usually acted on, because the manager is in the room.
Where it falls short is everything else. There's no aggregation, so themes are invisible. There's no second pair of eyes, so the manager's blind spots are the company's blind spots. There's no record the next HR lead can read to see what the last leaver said about the same manager. And because most managers aren't trained to run an exit conversation, the questions get asked badly, the answers get sanitised in the writing-up, and the document that survives is the one the manager would be comfortable showing their own boss. For any organisation with more than around twenty leavers a year, or with managers of varying quality, or with a turnover pattern that's not yet understood, this format produces the illusion of a process without the substance of one.
The Decision Table
| Situation | Scale | Setup | Primary Pain | Recommended Starting Point |
|---|---|---|---|---|
| Trusted small team, low churn | Under twenty leavers a year | HR lead knows each departure personally | Process theatre if formalised | Keep the manager conversation; write a one-page note template |
| Growing org, manager-led today | Twenty to sixty leavers a year | Managers run exits, notes inconsistent | No comparable record, no aggregation | Standardise the manager form; aggregate quarterly |
| High volume, regulated environment | Sixty to a few hundred a year | HR cannot book everyone; stakes per departure high | Missing the systemic signal | Short form for everyone; conversation only on a written trigger |
| Engineering or specialist talent, low volume | Under thirty leavers a year, high replacement cost | Each departure stalls a project | One bad exit costs more than the process | Conversation for everyone, with a structured template |
| Restructuring or severance underway | Any volume | Active legal channel open | Exit data collides with severance evidence | Pause formal data collection; route through legal |
| Post-acquisition integration | High short-term churn | Trust is low, references matter | Leavers cannot afford to be honest on day one | Delayed survey only, with consent; minimal in-employment process |
| Unionised or works-council environment | Any volume | Representative consultation rights | Process design must include consultation | Build the format with representatives before rollout |
Choosing by Volume and Stake
Matching the instrument to the population means staring at two numbers: how many people are leaving, and how much each one costs you when they go. The format that's right for a contact centre losing forty people a quarter is wrong for a research lab losing four. The two axes interact. High volume pushes you toward written formats because conversations don't scale. High stake per departure pushes you back toward conversations because the marginal value of depth is high. The mistake is picking on one axis and ignoring the other.
| Leavers per year | Cost per departure | Right format | Why |
|---|---|---|---|
| Under twenty | Low | Manager conversation with a one-page template | The HR lead can read every note; aggregation is unnecessary |
| Under twenty | High | Live conversation for everyone, structured | Depth matters; volume is small enough to make it possible |
| Twenty to eighty | Low to medium | Short form for everyone, quarterly aggregate review | Reach is the constraint; depth comes from aggregation |
| Twenty to eighty | High | Form plus a written trigger for conversation | Reach plus depth on the cases that matter |
| Over eighty | Low to medium | Form only, with anonymous routing | Conversation is no longer realistic; comparability is the value |
| Over eighty | High | Form with a high-stake trigger; dedicated interviewer | Some departures warrant depth; design the rule in writing |
A useful pressure test: if your leaver count doubled next quarter, would your current format still work? If the answer is no, you don't have a process. You have a habit.
The Delayed Survey
A delayed survey is the format most teams consider and few run well, because the obstacles aren't the questions. The obstacle is whether the leaver will even see the email. By the time a few weeks have passed, a meaningful share of leavers have changed personal email addresses, have moved jobs at speed, or have simply stopped opening mail from a former employer. The response rate on a delayed survey is reliably lower than the response rate on an in-employment form, and the gap widens with time. There's also a data protection issue you can't ignore: a survey sent after employment ends is a fresh processing activity that requires its own lawful basis in most jurisdictions. Consent is the cleanest basis, and consent is something you have to ask for, in writing, before the leaver leaves.
The honest version of what you get in return is candour. A person who has already started a new job, who no longer needs a reference from you, who has had two weeks to decompress, will often say things they wouldn't have said on the last day. The classic case is the manager who is the problem. On day one, the leaver will minimise. Two weeks in, with the next job signed and the reference already given, they will name names. That's the value of the delayed survey. It's also the reason to think hard about whether you want it, because the answers you get may not be comfortable.
| Aspect | In-employment form | Delayed survey |
|---|---|---|
| Friction for leaver | Low | Higher |
| Candour on sensitive topics | Lower | Higher |
| Response rate | Higher | Lower |
| Data protection basis | Existing employment relationship | New, requires fresh basis |
| Timeliness for action | Real-time | Lagged |
| Best use | Operational signal you can act on this quarter | Strategic signal about culture, leadership, structural issues |
The right framing is that the delayed survey isn't a substitute for an in-employment process. It's a second instrument that catches what the first one missed. If you run only the delayed survey, you lose timeliness. If you run only the in-employment form, you lose candour on the topics that matter most.
What to Put in Writing
A good decision isn't defensible because it was right. It's defensible because there's a record that shows what was decided, when, by whom, and on what basis. The artefacts below are the ones that turn a conversation in a meeting into a process the next person can run.
| Artefact | Who owns it | When it is written | What it prevents |
|---|---|---|---|
| The chosen format and its scope | HR lead | Before rollout | Scope drift, ad-hoc additions, a format chosen for the wrong reason |
| The question set, in full | HR lead | Before the first leaver | Different questions being asked of different leavers |
| The rule for triggering a follow-up conversation | HR lead, signed off by a manager | Before rollout | Selection bias, inconsistent application, manager self-selection |
| The notes or responses from each leaver | HR lead, restricted access | On the day or shortly after | Loss of memory, inconsistent quality, loss of record |
| The aggregation, by quarter | HR lead or analyst | Quarterly | Themes staying hidden until they become a retention crisis |
| The action taken in response | The person with authority | Within a reasonable window of the data | The process being perceived as a box-ticking exercise |
| The data protection basis for any post-employment contact | HR lead, with legal review | Before the first delayed survey | A privacy complaint, a regulatory query, an uncomfortable surprise |
| The retention rule for the records | HR lead, signed off by data owner | At rollout | Records surviving longer than they should, or being deleted too soon |
The artefact most teams skip is the action. Without it, the survey is a recording with no listener, and the next leaver will quietly know that.
Questions to Ask Before You Commit
These are the questions worth sending to a provider, an adviser, or your own team before you lock the format. They're written to be used, not admired.
On scope. "How many leavers did we've in each of the last four quarters, and what do we expect in the next four?" A bad answer: a guess, or a number pulled from a single quarter. You're matching the format to the volume, and the volume has to be honest before the format can be right.
On ownership. "Who has authority to act on what the data shows, and what is the example of an action we've already taken?" A bad answer: "the HR team". The HR team can flag. They can't usually fix.
On data protection. "What is our lawful basis for contacting leavers after employment ends, and where is that documented?" A bad answer: "we've always done it". The exposure shape varies by jurisdiction, and what you've always done may not be what the current rules allow.
On the trigger. "What is the written rule for deciding who gets a follow-up conversation, and who set it?" A bad answer: "we use our judgment". Judgment is fine. Undocumented judgment isn't.
On aggregation. "Where do the notes or responses live, who can see them, and how often are they read across the population?" A bad answer: "in a folder". A folder isn't aggregation.
On candour. "What do we do to make the format safe for honest answers, and what do we do when the honest answer names a senior manager?" A bad answer: nothing, on either count. The candour mechanism and the handling mechanism are the same question.
On the action loop. "Show me the last time exit data changed a decision in this organisation." A bad answer: silence. The process has not worked if no one can name the change it produced.
On exit. "If the leaver asked to see their record in a year's time, what would we hand over, and is anyone reviewing the documents before we do?" A bad answer: "we'd have to check". The format you're using is the wrong one if you don't already know what the document would look like.
The Cost of Getting This Wrong
The bill never arrives as a single line item. It arrives as a pattern that someone, two quarters later, traces back to a departure that was missed, or a manager who should have been flagged, or a compensation band that should have been reviewed and wasn't. The second-order costs are the ones that compound. A pattern of exits for the same reason, repeated, is the cost of a format that didn't surface the theme. A reference dispute where the leaver can show that concerns were raised in an exit survey and never acted on is the cost of a process that collected data with no loop back. A regulator asking why a former employee's data was processed after their employment ended, without a clear basis, is the cost of a delayed survey run on autopilot.
There's also a quieter cost inside the company. Managers who know the exit survey is theatre stop taking it seriously, and the signal you would have got from them, the honest answer about why someone really left, evaporates. So does the signal from leavers who see that the last twenty surveys produced no change. By the time the next leaver opens the form, they have already decided how much of themselves to put into it. The format you choose today sets the ceiling on the answer you'll get next year.
So the question isn't which format is cheapest to run. It's which format produces a record you would still be willing to defend in eighteen months, when the reason it mattered has become obvious.
When You Are Ready to Go Further
If your volume has outgrown informal practice and you're weighing a real change, the next step is comparison work that doesn't start from a vendor list. HROpsLab is a review publication, not a vendor. We don't sell software, payroll, or advisory services. What we do is independent comparison of the tools and approaches HR operations teams actually run, written for the person who has to make the call. Our retention and exit-process research is built around the same trade-off this article is built around: depth against comparability, action against theatre. If you want to see how the formats stack up against each other in organisations of your size and shape, our comparison work is a starting point that's not a sales call.
Frequently Asked Questions
Which format gets better answers, a survey or an interview?
A conversation gets better answers on the topics you didn't know to ask about, because the interviewer can follow a thread. A survey gets better answers on the topics you did ask about, because it produces comparable data across many leavers. If you have to pick one, pick on the basis of which kind of answer you would actually change something in response to.
Should the exit process be anonymous?
The honest answer is that you have to separate the response from the identifying details, and write down how. A leaver will be more candid if they know the document handed back to them on request won't contain their name attached to a critical comment about their manager. Anonymous-by-design is the safe default; identifiable-by-design needs a written reason.
What response rate should we expect?
A written form sent on the last day typically gets a higher completion rate than a conversation gets a booking rate. A delayed survey sent weeks after departure typically gets less than either. None of these are guarantees, and the rate you actually get is the rate your specific format, your specific leaver population and your specific relationship with them produces. The question isn't whether the rate is high in the abstract; it's whether it's high enough to act on.
Should we survey people after they have left?
Sometimes. A delayed survey catches candour that the in-employment process can't, particularly on sensitive topics like the line manager. The trade is timeliness, response rate, and a fresh data protection question. Confirm the lawful basis for contacting former employees locally before you send the first one.
How long should the form be?
Short enough that a tired leaver on their last day will actually finish it. A long form with many open boxes reliably delivers three words in each box, because the leaver isn't going to write a paragraph in a field they suspect no one will read. If the form is longer than you would fill in yourself on your last day at a job, it's too long.
Should we do both a survey and an interview?
Yes, if the selection rule is written down in advance and reviewed. No, if "do both" means a long form with an optional conversation at the end, which reliably delivers the form and not the conversation. The combined approach works only when the trigger for the second instrument is set deliberately, before anyone starts collecting.
What do we do when nobody completes it?
Treat the non-completion as data. It tells you that the format didn't feel safe, the timing was wrong, or the population doesn't believe the answers will be acted on. The fix is rarely a reminder email. The fix is usually a change to who reads it, when it's sent, and what the leaver believes will happen as a result.
How long should we keep the records?
Long enough to be useful, short enough to be defensible. The exact retention period depends on the jurisdiction and on what the records are used for. Define the retention rule at the point of rollout, not when someone asks for the data to be deleted.
HROpsLab is independent HR operations research for teams who have to make the call.
