One exit is a story. Five with the same shape is a finding
Exit interviews are conducted carefully and kept on file. Each one is read once by a single person. The patterns across them are never assembled, because that would mean reading a year's conversations again.
The stated reason is often not the real one. People who leave keep the relationship in case they need it later, and the true reason appears only as a side note rather than as the answer.
Individual attribution is dangerous. If you link an exit theme to a particular manager or team and report it to anyone, then talking about a departure is no longer safe for people who leave later.
The reporting level is a design choice. Organization-wide themes are selected with a minimum threshold. More detailed themes are omitted even when the more detailed finding is the interesting one.
How the ActionFlow runs on the canvas
Exit interview notes are read from a Notion or Airtable node, or they are produced with Document Upload and Transcribe Audio when the interviews were recorded with consent.
A Filter node removes fields that identify someone, such as name, manager, and exact team, before anyone reads the content. Anonymity is structural only because the removal happens first, rather than as suppression applied later.
A Start Scheduled node runs the analysis over a period, not once per interview, because a single exit should never produce a finding.
A Generate Object node groups the interviews by underlying reason and returns the themes, their counts, and the general area, not the specific team.
A Math node tests each theme against the minimum count. A Filter node holds anything below that threshold for the next period rather than publishing a thin theme.
A Generate Object node separates the stated reason from the actual reason the conversation supports, and returns both. The gap itself is the most useful output, and it appears only at the aggregate level.
A Notion node publishes the report. A Slack node notifies people leads. The report is not broken down to a level where anyone who reads it could identify a leaver.
Nodes this ActionFlow uses
- Notion (Integration): holds exit interview notes. Document Upload covers recorded interviews.
- Filter (Data): strips identifying fields before any content is read.
- Start Scheduled: runs over a period so a single exit never produces a finding.
- Generate Object (AI Core): groups by underlying reason and returns general areas, not teams.
- Math (Util): tests each theme against the minimum count.
- Filter (Data): holds thin themes for the next period rather than publishing them.
- Notion (Integration): publishes at a level where no leaver is identifiable.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.