Knowledge management

Recovering institutional knowledge from staff mailboxes

A property management company · August to September 2026

When people leave a property management company, the vendor relationships and building history they carried around in their heads usually leave with them. The fix meant reading through several old mailboxes, some running past a hundred thousand messages, and being just as careful about what got left out of the result as what went in.

Category

Knowledge recovery from unstructured communications.

What I was trying to accomplish

Rebuild the working knowledge that lived only in departed employees' inboxes, who the vendors are, which entity owns which building, what's still pending, into a structured reference successors could actually use.

The data challenges

Several mailboxes, some running past a hundred thousand messages, read through delegated access, reconciled against the one authoritative system of record, while filtering out sensitive topics that don't belong in a shared reference.

The result

Per-mailbox knowledge workbooks and a reconciled reference that corrected vendor names the system of record had gotten wrong, with every finding confidence-rated and anything borderline flagged for a person to decide rather than guessed at.

Technical details

Tools: Claude, Microsoft 365 mailbox connector with delegated access, Excel

In a property management company, much of the working knowledge lives in individual inboxes. When experienced people move on, that knowledge often goes with them. The goal was to rebuild it from email and give successors a structured reference instead of a pile of messages.

The inputs were four mailboxes: three from former staff, read through delegated access, and the successor's own. The property management system's active vendor list served as the authoritative baseline. Across the builds, the workbooks and matching CSV files covered vendors, owners and client entities, properties, projects, pending items, internal contacts, and recurring processes. Findings carry a high, medium, or low confidence rating based on how much evidence supports them.

The challenges were mostly about keeping the record honest. An owner-to-portfolio list built from one mailbox did not survive contact with the next one, so the contradictions were documented and both lists are marked provisional until checked against the system of record. Where only part of a search result was read, the count is labeled as a minimum. Items that sat on the line between a business record and an excluded category were flagged but not characterized, for a person to review.

Pull up a chair.

Let's talk through where things stand, and where they could go.

Back to case studies