Anonymize and Theme Exit-Interview Notes into a Quarterly Attrition Brief

The task

HR business partners and People Ops leads collect exit-interview notes all quarter, then have to turn them into something leadership will actually read — usually a one-pager naming the real reasons people left. With layoff news rippling through tech (Elastic just cut 7% of staff citing AI-driven restructuring), execs are paying closer attention to voluntary attrition signals too. This workflow turns a messy folder of interview notes into an anonymized, themed brief in one sitting.

Before AI

The manual version: open each interview write-up, redact names and identifying details by hand, tag each comment with a theme in a spreadsheet, count frequencies, then draft a narrative summary. For 15-25 interviews in a quarter, that's a 3-4 hour block — and the redaction step is the one most likely to get skipped under deadline pressure, which is exactly the step that protects departing employees.

The bar for a good brief is well-established: aggregate themes across multiple interviews to identify systemic issues rather than one-off complaints, and share findings with relevant department heads without identifying individual employees. The work is straightforward — it's just slow.

The workflow

Paste raw notes into Step 1. The output of each step feeds the next. See AIHR's seven-step exit interview data analysis guide for the underlying method this workflow automates.

Step 1 — Anonymize the raw notes

You are an HR analyst preparing exit-interview notes for thematic analysis. The notes below contain identifying details that must be removed before any analysis.

Produce an anonymized version of the notes with these rules:
- Replace every employee name with [EMPLOYEE_##] where ## is a stable two-digit ID per person (EMPLOYEE_01, EMPLOYEE_02, etc.)
- Replace manager names with [MANAGER_A], [MANAGER_B], etc.
- Replace specific team or product names with [TEAM_1], [TEAM_2], etc.
- Replace specific client/customer names with [CLIENT_X]
- Remove or generalize any detail that could identify a single person: unusual job titles, specific tenure in months, relocation cities, named medical conditions, named family members. Round tenure to nearest year. Replace cities with region (e.g., "US West Coast", "EMEA").
- Keep department (e.g., Engineering, Sales, CX) at the department level only — do not preserve sub-team names.
- Preserve the substance of every complaint, reason for leaving, and suggestion verbatim where possible.

Output the anonymized notes only, separated by "---" between interviews. Do not summarize yet. Do not add commentary.

Exit-interview notes to anonymize:

Interview 1 — Priya Sharma, Senior Engineer, Platform Infrastructure team, 3.5 years, leaving for Stripe. Manager: David Chen. Based in Seattle.
Reason given: "Burned out. On-call rotation got worse after Marcus left in Q1 and we never backfilled. I raised it with David four times. Comp also fell behind — got a 3% bump while a friend at Datadog got 18% to join." Would she recommend the company? "To a junior, yes. To anyone senior, no — the ladder above L5 is broken."

Interview 2 — Tomás Rivera, Account Executive, Mid-Market SMB segment, 1 year 2 months, going to Gong. Manager: Sarah Whitfield. Austin office.
Reason: Quota was raised 40% mid-year after the GTM reorg. "I hit 92% of the old number and was put on a PIP for missing the new one. Felt like a setup." Praised onboarding. Felt CLIENT Acme Corp account was pulled from him unfairly and given to a rep closer to Sarah.

Interview 3 — Aisha Okonkwo, Product Designer II, Growth pod, 2 years, going to a startup. Manager: David Chen (dotted line). London.
"I love the work. I do not love that the only two women on my pod both left this year. The promo committee feedback I got said I needed to be 'more assertive' — same word my 2023 review used. Meanwhile two male peers at my level got promoted with similar output." Mentioned she has a toddler and the return-to-office push to 4 days made London commute untenable.

Interview 4 — Greg Halloran, Staff Engineer, Platform Infrastructure, 5 years, retiring early. Manager: David Chen. Seattle.
"Honestly, it's time. But I'll say this: we have lost three senior ICs from Platform in nine months. David is a good person and a poor manager — he avoids conflict and lets staffing problems fester. On-call is brutal. Comp is below market for senior folks."

Interview 5 — Lina Park, Customer Success Manager, Enterprise segment, 2.5 years, going to Notion. Manager: Sarah Whitfield. Remote, Toronto.
"The RTO mandate was the trigger. I was hired remote. Changing the deal 18 months in felt like a breach." Also flagged tooling: "We're still doing QBRs in Google Slides because the CS platform purchase has been 'next quarter' for six quarters."

Interview 6 — Daniel Brooks, Engineering Manager, Payments team, 4 years, going to a competitor. Manager: VP Eng Rachel Liu. NYC.
Reason: "Comp and scope. My team shrank from 9 to 6 over the year through attrition and I was told to 'do more with less.' Bonus was 60% of target despite hitting all my OKRs." Mentioned the bonus formula changed in Q3 without clear communication.

Step 2 — Tag themes and count frequency

Using ONLY the anonymized notes from the previous step, perform thematic coding.

1. Identify 5-8 distinct themes that emerge across the interviews. Themes should be specific enough to act on (e.g., "On-call burden after unbackfilled departures" — not just "burnout"). Avoid themes that map to fewer than two interviews unless the issue is severe (e.g., a single bias complaint still gets named).

2. For each theme, output:
   - Theme name (short, neutral phrasing)
   - One-sentence definition
   - Number of interviews citing it (e.g., 4 of 6)
   - The anonymized employee IDs that raised it
   - One representative quote (anonymized, in quotes)
   - Severity: Low / Medium / High — High = legal/compliance risk, named bias pattern, or 3+ interviews in same theme

3. Flag any theme that touches potential protected-class bias (gender, race, age, caregiving status, disability) as REQUIRES ESCALATION regardless of count.

Output as a structured list. Do not write a narrative yet.

Step 3 — Draft the leadership brief

Using the themed analysis from the previous step, draft a one-page quarterly attrition brief for the executive team. Use this exact structure:

**Quarterly Voluntary Attrition Brief — Q[X] [YEAR]**

1. **Headline** — One sentence naming the single most important pattern this quarter.

2. **Top 3 themes** — For each: theme name, count (e.g., "4 of 6 interviews"), severity, and 2-3 sentences of plain-English explanation. Do NOT include direct quotes here — this is for executives, keep it analytical.

3. **Escalation items** — Any bias, legal, or compliance flags from the prior step. State them plainly. If none, write "None this quarter."

4. **What we are NOT seeing** — One short paragraph on what is conspicuously absent (e.g., if nobody mentioned strategy or product direction, say so — that's a signal too).

5. **Recommended actions** — 3-5 bullets. Each must name an owner role (e.g., "VP Engineering", "Total Rewards", "DEI Council") and a concrete next step. No vague "consider exploring" language.

6. **Method note** — One sentence: sample size, anonymization, and the caveat that exit-interview data is self-selected and skews toward people willing to talk.

Tone: direct, non-defensive, no HR euphemisms. An exec should be able to read it in under three minutes and know what to do.

Gotchas

• Anonymization is not redaction-proof. With a sample of 6 interviews and detailed context, a determined reader who knows the org can still re-identify people. Before sharing externally, do a human pass — especially on tenure, location, and any "only person who…" details. As Metaview notes, the goal is to protect individual anonymity while still surfacing actionable findings — that tension is real and the model won't resolve it for you.
• Theme inflation. LLMs love to find five themes when there are really two. If you have fewer than ~8 interviews, force the model down to 3-4 themes or you'll over-interpret noise.
• Severity scoring is a judgment call. The model will mark things "High" that you might call "Medium," and vice versa. Override it — you know your company's risk posture.
• Bias flags need a human. A single comment about a promo pattern (like the one in the sample) deserves a real investigation conversation with your employment counsel, not a bullet in a brief. The model should surface these — not action them.
• Self-selection bias is real. People who agree to exit interviews are systematically different from those who ghost the process. Keep the method-note caveat in the final brief; do not let execs treat n=6 as the voice of the workforce.
• Don't paste real notes into a consumer chatbot. Use your org's approved enterprise instance with data-retention turned off, or run this locally. Exit interviews are confidential personnel records.

Time saved

Roughly 3-4 hours per quarterly cycle for an HR team handling 15-25 exits — most of it from the anonymization and theme-coding steps, which are mechanical. Drafting the executive brief from a clean themed analysis takes about 15 minutes instead of an hour. This is an estimate from comparing the manual steps to the prompt flow; your mileage depends on note quality and how much editing the draft brief needs before it goes to leadership.

What we got when we ran it

We executed every prompt above, in order, against the sample input (model: claude-sonnet-4-6). Outputs below are real and unedited apart from length truncation.

Step 1 output

Interview 1 — [EMPLOYEE_01], Senior Engineer, Engineering, ~4 years, leaving for a fintech company. Manager: [MANAGER_A]. Based in US West Coast. Reason given: "Burned out. On-call rotation got worse after [EMPLOYEE_02] left in Q1 and we never backfilled. I raised it with [MANAGER_A] four times. Comp also fell behind — got a 3% bump while a friend at a competitor got 18% to join." Would she recommend the company? "To a junior, yes. To anyone senior, no — the ladder above L5 is broken." --- Interview 2 — [EMPLOYEE_03], Account Executive, Sales, ~1 year, going to a sales technology company. Manager: [MANAGER_B]. US South Central office. Reason: Quota was raised 40% mid-year after the GTM reorg. "I hit 92% of the old number and was put on a PIP for missing the new one. Felt like a setup." Praised onboarding. Felt [CLIENT_X] account was pulled from him unfairly and given to a rep closer to [MANAGER_B]. --- Interview 3 — [EMPLOYEE_04], Product Designer, Design, ~2 years, going to a startup. Manager: [MANAGER_A] (dotted line). EMEA. "I love the work. I do not love that the only two women on my pod both left this year. The promo committee feedback I got said I needed to be 'more assertive' — same word my 2023 review used. Meanwhile two male peers at my level got promoted with similar output." Mentioned that the return-to-office push to 4 days made her commute untenable due to caregiving responsibilities. --- Interview 4 — [EMPLOYEE_05], Staff Engineer,

…output truncated at 250 words.

Step 2 output

Thematic Coding — Exit Interview Analysis --- ### Theme 1: Below-Market Compensation for Senior Roles **Definition:** Senior and mid-level employees report compensation materially lagging market rates, with raises insufficient to close the gap. **Interviews citing it:** 4 of 6 **Employees:** [EMPLOYEE_01], [EMPLOYEE_03], [EMPLOYEE_05], [EMPLOYEE_07] **Representative quote:** "Comp also fell behind — got a 3% bump while a friend at a competitor got 18% to join." — [EMPLOYEE_01] **Severity:** High --- ### Theme 2: Unbackfilled Departures Creating Operational Burden **Definition:** Departing employees are not replaced, leaving remaining staff with unsustainable workloads, particularly on-call responsibilities in Engineering. **Interviews citing it:** 3 of 6 **Employees:** [EMPLOYEE_01], [EMPLOYEE_04], [EMPLOYEE_05] **Representative quote:** "On-call rotation got worse after [EMPLOYEE_02] left in Q1 and we never backfilled. I raised it with [MANAGER_A] four times." — [EMPLOYEE_01] **Severity:** High --- ### Theme 3: Opaque or Inconsistently Applied Performance and Reward Processes **Definition:** Employees report that promotion criteria, bonus formulas, and performance targets shifted without clear communication, undermining trust in evaluation fairness. **Interviews citing it:** 4 of 6 **Employees:** [EMPLOYEE_03], [EMPLOYEE_04], [EMPLOYEE_07], [EMPLOYEE_01] **Representative quote:** "The bonus formula changed in Q3 without clear communication." — [EMPLOYEE_07] **Severity:** High --- ### Theme 4: Remote/Flexible Work Agreements Revoked Post-Hire **Definition:** Employees hired under remote or flexible arrangements report that RTO mandates changed their working conditions without renegotiation, acting as a direct trigger for departure. **Interviews citing it:** 2 of 6 **Employees:** [EMPLOYEE_04], [EMPLOYEE_06] **Representative quote:** "I was hired remote. Changing the deal 18 months in felt like a breach." —

…output truncated at 250 words.

Step 3 output

# Quarterly Voluntary Attrition Brief — Q2 2025 --- ## 1. Headline The company is losing experienced people primarily because it is underpaying them, breaking operational promises, and leaving a concentrated management problem unaddressed — three issues that compound each other and will accelerate attrition if not sequenced and resolved this quarter. --- ## 2. Top 3 Themes Theme 1: Below-Market Compensation for Senior Roles 4 of 6 interviews — Severity: High Senior employees across Engineering and Sales are receiving compensation materially below what competitors are offering, and they know it. This is not a perception problem — departing employees are citing specific competing offers that dwarf internal raise percentages. The promotion ladder above the senior individual contributor level is also flagged as structurally broken, meaning high performers see neither a pay path nor a growth path inside the company. Theme 2: Opaque and Inconsistently Applied Performance and Reward Processes 4 of 6 interviews — Severity: High Employees are losing trust in the integrity of how the company measures and rewards performance. Bonus formulas changed mid-cycle without communication. Promotion criteria appear to be applied differently across individuals. Sales quotas were raised 40% mid-year and existing performance was then judged against the new bar. When employees cannot predict or understand the rules by which they are evaluated, they stop believing the system is fair — and they leave. Theme 3: A Single Manager Is a Concentrated Attrition Risk 3 of 6 interviews — Severity: High Three departing employees from Engineering independently

…output truncated at 250 words.