How to analyze exit interview data: from a stack of forms to a decision

Turning a stack of exit interviews into a decision takes four steps: code each response into a theme using a fixed list of tags, tally themes by frequency and by where they cluster (department, manager, tenure), separate a real pattern from a handful of unrelated comments using a threshold you set in advance, and write the action against the pattern, not against any single comment. Below is the method with a worked example using twelve invented exit interviews, so you can see exactly how raw answers become a decision rather than a vague sense that "people mention growth a lot."

Step 1: Standardize what you are coding

This method assumes every exit interview used the same question set, ideally the form in exit interview form template. If interviewers ask different questions in a different order, coding produces noise, because you cannot tell whether a theme is missing from a response or simply was never asked. If your organization has been running exit interviews without a fixed form, standardize going forward before trying to code the historical mix — coding inconsistent data reliably is not usually possible, and pretending otherwise produces a false sense of precision.

Step 2: Build a short list of themes before you start reading

Read five or six exit interviews first and draft a starting list of eight to twelve themes. Resist the urge to make a new theme for every distinct phrase — "unclear promotion criteria" and "no path to senior level" are the same theme worded two ways. A workable starting list for most organizations:

THEME CODES
COMP    — Compensation or benefits
GROWTH  — Promotion path, skill growth, career development
MGR-FB  — Manager feedback or communication quality
MGR-REL — Manager relationship, trust, respect
WORKLD  — Workload, staffing, burnout
TOOLS   — Tools, systems, resources to do the job
FLEX    — Flexibility, remote/hybrid, schedule
CULTURE — Team culture, belonging, recognition
OFFER   — Left for a specific competing offer, no dissatisfaction cited
OTHER   — Doesn't fit above; note verbatim for the next review

Tag each response with the theme(s) it raises — most responses get one
or two tags, some get none if the answer names no specific issue.

Step 3: Code each response

Go through each exit interview's open-ended answers (mainly "why are you leaving" and "what would have changed this") and assign one or two theme codes based on what is actually said, not what you assume based on the role or department. Keep a short verbatim next to each code so a reviewer can check your coding later without re-reading the whole form.

Worked example: twelve exit interviews from one quarter

Step 4: Tally by theme, then by where it clusters

The raw tally across twelve exits:

GROWTH   3 of 12 (25%) — all three in Sales, all under R. Kim
MGR-FB   3 of 12 (25%) — all three in Support, all under T. Alvarez
WORKLD   2 of 12 (17%) — both in Engineering, both under J. Osei
COMP     2 of 12 (17%) — one Sales, one Engineering, different managers
OFFER    1 of 12 (8%)  — no dissatisfaction cited
OTHER    1 of 12 (8%)  — personal relocation, not actionable

The raw percentage across the whole company understates what matters here. GROWTH looks like a quarter of exits company-wide, but it is really three of three Sales exits under the same manager — 100 percent of that specific group. The same is true for MGR-FB in Support under one manager. This is why step 4 is clustering, not just counting: a theme spread evenly across every team is a different problem, and a different fix, than a theme concentrated under one manager.

Step 5: Apply your action threshold

Using the threshold from exit interview report template — three or more exits naming the same specific issue, or a serious disclosure regardless of count — this quarter produces two themes that clear the bar:

Notice WORKLD is treated as action-worthy despite being under the numeric threshold. A rigid rule applied without judgment would file it as "watch and wait," but the content of the comments — a specific, named operational problem, with one departing employee reporting that a third person already raised it outside the exit interview — is strong enough evidence to act now. A threshold is a floor for routine themes, not a ceiling on judgment for a specific, credible, repeated complaint.

Step 6: Write the decision, not just the finding

"Growth came up in three exits" is a finding. "Review R. Kim's team's promotion conversations with them directly, owner: their manager, by [date]" is a decision. Every theme that clears the threshold should leave this process with an owner and a date attached, in the format used in the report template. A finding with no owner and no date is how the same theme reappears, unchanged, next quarter.

Tagging sentiment alongside theme

A theme code alone can mislead. "FLEX" (flexibility) might come up because someone is angry a hybrid policy was tightened, or because someone mentions, neutrally, that they liked the flexibility and will miss it. Add a second tag for sentiment — negative, neutral or positive — next to each theme code. In the worked example above, all three GROWTH mentions were negative (a gap the employee wanted fixed), which is why the theme is actionable. If one of those three had instead said "growth was fine, I'm leaving for a completely unrelated reason, just answering the question," it should not count toward the same cluster, even though it mentions the same topic. Sentiment is what separates "this came up" from "this is why they left."

Reviewing the OTHER bucket before you close the quarter

Do not treat OTHER as a bin to ignore. Read every response tagged OTHER once at the end of the quarter, specifically looking for two things: an answer that actually fits an existing theme but was worded unusually, and an answer that reveals a new theme you had not anticipated and should add to the code list next quarter. In the worked example, exit 7 ("spouse relocated, no complaints") is genuinely not actionable and stays in OTHER. A response like "just didn't feel like I belonged here, hard to explain" might get missed by the fixed list and deserves a new CULTURE-FIT tag rather than being left uncoded.

Common mistakes in analysis

• Coding to confirm what you already believed. If you expect "growth" to be the big theme, you will read ambiguous answers as growth-related more often than a neutral coder would. Coding blind to which manager or department a form belongs to, where practical, reduces this.
• Averaging away a cluster. A 25 percent company-wide theme that is really 100 percent of one team's exits needs a completely different fix than a theme evenly spread across the company. Always check the cluster, not just the total.
• Treating a small sample like a large one. Three exits under one manager is a real signal for a team of eight; it is much weaker evidence for a team of eighty. Note team size next to any cluster claim.
• No second coder, ever. Even an occasional spot-check by a second person catches drift in how one person applies the tags over time.
• Reporting themes with no decision attached. Turns the whole exercise into documentation instead of action.

Not every quarter produces a clean pattern, and forcing one where the data does not support it is worse than reporting "no clear theme this period, details below in case something develops." Twelve exits with twelve different, unrelated reasons is a real result. Say so, keep the coded data for next quarter's comparison, and move on rather than manufacturing a theme to justify the time spent analyzing it.