Every exit interview gets read once, on the day it is filed, by whoever ran it. Nobody ever reads them side by side, because they arrive one at a time over a year. This is the brief that hands that job to Claude. Copy it, paste it into a new Claude conversation, and attach the exit files you already keep. It asks what your notes are before it counts anything.
Nine of the year's exit interviews name the same manager. No single one of them said anything remarkable, because each was read once, by a different person, on the day it was filed, and then never read beside the other eight.
The notes were always there, sitting in a folder or a drive or an inbox, and reading them as a set meant somebody holding a year of departures in their head at once while keeping straight who ran each conversation, which questions were asked that quarter, and which departures were not the person's choice. Claude reads them all at once, asks what your files actually are before it counts a single thing, and puts its candidate themes back to you in your own words. The year becomes a short list, and the reasons people gave stop being filed and forgotten one at a time.
It is an ordinary Claude conversation. You paste the brief, send it, and drag your exit files into the same window when it asks. Nothing gets installed and nothing connects to your systems.
What it does not do. It runs on the files you attach by hand, and it reports what those files say. It does not survey anybody, it does not reach the people who left, and it cannot tell you what the people who stayed think.
An HR lead attaches last year's exit files. Before a single theme is counted, the run reports that the question set changed in April and started asking about workload directly, that four of the conversations were run by the departing person's own manager and the rest by HR, and that two departures in the folder were involuntary and would carry a different meaning if counted alongside the rest.
Your AI department, on demand. We plan, build, and deploy your company's AI strategy.