What is institutional knowledge, and why does it walk out the door?
Matt Cretzman · September 26, 2026 · 6 min read
Every company has two operating manuals. One is written down. The other lives in people's heads, and it's the one that actually runs the place.
That second manual is institutional knowledge.
What is institutional knowledge?
Institutional knowledge is the collective know-how of an organization: the facts, judgment calls, relationships and workarounds people pick up by doing the work. Most of it never makes it into a document.
It isn't the employee handbook. It's knowing things like:
- Which customer needs a phone call, not an email, before anyone shows up on site
- Why quotes for a certain kind of job always get padded, and by how much
- The workaround for the report that breaks every quarter-end
- Who at the supplier actually answers the phone
- What "done" means on one crew versus another
None of that is secret. It just never got written down, because the people who know it are too busy using it.
Institutional knowledge examples by team
- Sales: which objections are real, which discounts close and which ones just train customers to wait.
- Operations: the exceptions to every process, and the order things actually happen in versus the order the SOP says.
- Field and service: how a specific building, machine or customer behaves, learned over years of visits.
- Finance and admin: why an account is coded the way it is, and which vendor always invoices wrong.
- Leadership: why past decisions were made, so nobody relitigates them every two years.
If you've ever heard "ask Linda, she knows" in your company, you've found institutional knowledge. Some of it is also tribal knowledge, the riskiest kind, which I break down in tribal knowledge vs institutional knowledge.
Why institutional knowledge walks out the door
Institutional knowledge has no backup. It lives in a person, and people leave. They retire, they get recruited, they move. When they go, the knowledge doesn't transfer to the next hire. It evaporates, and the replacement spends months rebuilding it by trial and error.
You usually don't notice until something breaks: a bid gets priced wrong, a longtime customer gets annoyed, a compliance step gets skipped because nobody knew it existed.
There's a newer risk too. AI tools are only as useful as what they know about your business. If your company's real knowledge isn't captured anywhere, ChatGPT and Claude can't use it either. The companies that capture it get AI that answers like their best employee. Everyone else gets generic answers.
How to capture institutional knowledge
- Start with what would hurt most to lose. The person closest to retirement, the role with the most turnover, the process only one person understands.
- Interview, don't ask for documentation. People can't write down what they don't know they know. Ask about exceptions instead: What do new people get wrong? When do you ignore the SOP? What do you check that nobody asked you to?
- Pull in what already exists. SOPs, proposals, job files, call recordings, old email threads. It's messy, and that's fine.
- Structure it into small, specific pieces. One question type per piece, each with an owner who can say whether it's still right.
- Put it where people already work. Slack, Teams, Claude, ChatGPT, a text from the field. A knowledge base nobody opens is just a nicer graveyard.
- Track the questions that don't get good answers. That list is exactly what to capture next.
What this looks like in practice
A commercial fire protection contractor I work with had years of contracts, specs, SOPs and job files on a server nobody could search. Crews called the office for everything.
We turned 214,000 files into an assistant in Slack. It pulls project documents by job number and answers crews in English or Spanish. The knowledge was always there. Now it's reachable from a phone on a job site.
Where to start
Pick one role or one team, list the questions they ask most, and capture the answers from the people who know them. A first knowledge base for one team is usually live in weeks, not quarters.
That's the work I do. I build it on Skill Refinery, the knowledge platform I built for exactly this, and your team asks it in the tools they already use.
Want this built for your company instead?
See how I build AI knowledge bases