Havish M Consulting
Customer Case Study · AI Voice Exit Interviews
Housing Finance × Havish M Consulting

Every leaver heard, in their own words

Exit interviews were skipped, rushed or ended with "personal reasons". HMC built an AI voice interviewer for the housing finance company's HR team: it talks to the leaver, asks five questions that adapt to each answer, and turns the conversation into a coded exit record and an attrition insights dashboard.

ClientHousing Finance FunctionHuman Resources · Retention ThemeAI Voice Exit Interviews
Leavers completing an exit interview
0%
▲ from about 35%
Exits with a specific, actionable reason
0%
▲ from about 20%
From last day to insight with the HRBP
0day
▼ from about 30 days
Regrettable attrition in pilot zones
0% lower
▼ actions on what leavers said

Figures indicative; validated metrics available on request.

The Challenge

Exit interviews that told HR nothing

With a large, spread-out workforce of relationship managers, credit and collections staff, HR could not interview every leaver well, and the ones they did rarely got past the polite answer.

Too few interviews

HRBPs covered many branches; most leavers left without a conversation.

Same script for everyone

Fixed forms got fixed answers: "better opportunity", "personal reasons".

Not comfortable with the manager

Leavers held back when the person asking was close to the reason they were leaving.

Insight arrived late

Notes sat in files for weeks; patterns by branch or manager were never seen.

HMC Approach

A good interviewer, available to every leaver

Conversation design first, then voice, then coding the answers into something HR can act on.

Select a step to see what happened.

Discover · What good exit conversations sound like

Listened to HRBPs' best interviews.

  • Reasons taxonomy from two years of exit data.
  • Follow-up questions that unlock the real reason.
  • Sensitive topics and when to hand over to a person.

Deliverable: Conversation playbook and reason codes.

Design · Five questions, all dynamic

Short, warm and specific.

  • One open question, then four that follow the leaver's answers.
  • Coverage rules: reason, manager, growth, workload, what would have kept them.
  • Guardrails: no pressure to stay, confidential route for misconduct.

Deliverable: Interviewer prompt and guardrails.

Build · Voice, transcript, record

Natural voice and accurate coding.

  • ElevenLabs voice for questions, speech-to-text for answers.
  • GenAI chooses each next question and codes the record.
  • Dashboard by zone, role, tenure and manager.

Deliverable: Working interviewer and dashboard.

Adopt · Pilot with consent

Optional, private and short.

  • Leavers invited on their last week; under 10 minutes.
  • HRBPs review every record and flag.
  • Monthly attrition review uses the dashboard.

Deliverable: Pilot results and HR process.

Scale · From exits to stay interviews

Hear people before they resign.

  • All zones and roles.
  • Manager-level patterns shared with leaders.
  • Same engine for stay interviews with high performers.

Deliverable: Roll-out plan.

The Solution

Talk to the exit interviewer

Pick a leaver profile and start. The interviewer speaks, listens, and chooses each next question from what you just said. Five questions, then a coded exit record. Everything runs in your browser.

✦ Live voice & GenAI

Optional. Without keys the demo uses your browser's voice, browser speech recognition (Chrome) or typing, and rule-based follow-up questions. Add keys for an ElevenLabs voice, accurate transcription and GenAI questions and records.

ElevenLabsBrowser voice
OpenAIRule-based
Keys go only from this browser to ElevenLabs and OpenAI. Audio and transcripts stay in this page.
01 · Voice exit interview

Five questions, each one shaped by the last answer

Choose who is leaving, then press Start and answer out loud. The interviewer waits for you to finish (or tap "Done"), then decides what to ask next. Prefer not to speak? Switch to typing.

Who is leaving

Interviewer voice

Your answers are recorded only in this browser. Nothing is stored on a server by this page.

Ready
Press Start to begin the exit interview.

Live transcript

Rule-based

The conversation will appear here.

Synthetic leaver profiles. Microphone access is requested only when you start in voice mode; if it is blocked, typing takes over.

02 · Exit interview record

The conversation, coded for HR

After the fifth answer the interview is summarised into a record: primary reason, categories, whether the exit is regrettable, what would have kept the person, sentiment through the conversation, key quotes and actions. A sample is shown until you finish your own.

03 · Attrition insights

What leavers are telling the company

Sixty synthetic interviews from the pilot, plus every interview you complete above. Filter by zone and role; your interview appears in the charts as soon as it is coded.

Primary reasons · regrettable vs not

Reason by tenure

Latest interviews

04 · HR copilot

Ask across every exit interview

Ask a question in plain English. Template answers come from the coded records; with an OpenAI key the AI reads the records and answers anything.

Templates
✦

Pick a suggested question or type your own.

Synthetic interview records. HR reviews every record; the tool does not make retention or rehire decisions.

Impact Dashboard

More leavers heard, and heard sooner

Almost every leaver now gives a usable reason, and HRBPs see it the next day instead of a month later.

What we learn from 100 leavers

Pictogram: one figure per leaver

No interviewVague reasonSpecific, actionable reason

Days from last day to insight

Strip plot: one dot per exit, by role

Figures indicative; validated metrics available on request.

Before vs After

How exit interviews work now

DimensionBeforeAfter
CoverageAbout a third of leaversAlmost every leaver, any time
QuestionsSame form for everyoneFive, each shaped by the last answer
Who asksOften the managerA neutral AI voice, reviewed by HR
OutputFree-text notesCoded reasons, regrettable flag, quotes, actions
Sensitive issuesRarely raisedFlagged with a confidential route
InsightWeeks later, if at allNext day, by zone, role and tenure
Tools & Tech Used

The voice interview stack

ElevenLabs natural interviewer voice ElevenLabs Scribe speech-to-text OpenAI GPT dynamic questions and coding Web Audio turn-taking and voice activity D3 attrition insights Power BI monthly HR review
Key Takeaways

What any HR team can borrow

01

Follow the answer, not the form

The second question matters most. Asking about what the person just said is what gets past "personal reasons".

02

Neutral, short and optional

Five questions in under ten minutes, without the manager in the room, gets honest answers.

03

Code it, then act on it

A reason only helps when it is counted by branch, role and manager, and someone owns the action.