Every company has a person who "just knows." They know which client needs a call before the invoice goes out. They know which report gets pulled from which system, and which column to ignore. They know why the onboarding checklist has a step nobody follows anymore.
Nobody wrote any of this down. It didn't seem necessary. It worked.
Until that person goes on holiday, gets promoted, or leaves. Then the company discovers that a significant part of how it operates was never actually owned by the company. It was owned by someone's memory.
Tribal knowledge feels free. It isn't.
Undocumented processes don't show up on any P&L. There's no line item for "time lost because only one person knows how this works." That's exactly why they're so expensive: the cost is real, but it's spread across dozens of small moments that nobody adds up.
It shows up in four places.
Onboarding time. A new hire who could be productive in six weeks takes four months, because every question has to be routed to the one person who knows the answer — and that person has a day job.
Repeated errors. The same mistake happens every quarter, because the fix was never recorded. It was explained once, in a meeting, to whoever was in the room.
Key person risk. When one person holds the process, that person becomes a single point of failure. Their absence doesn't slow things down. It stops them.
Decision latency. Decisions wait. Not because they're hard, but because the information needed to make them lives in someone's head, and that someone is busy.
Putting a number on it
Try this exercise with any one process in your business.
Take a new hire. Estimate how many hours in their first three months are spent asking questions that a written process would have answered. Add the hours the experienced colleague spends answering them. Multiply by the number of people you hire in a year.
Then take one recurring error. Estimate how much time it costs each time it happens — to spot, to fix, to explain to the client. Multiply by how often it happens.
Most founders who do this honestly are surprised. Not because any single number is dramatic, but because the total is far larger than the cost of simply writing things down.
Documentation isn't a manual nobody reads
The usual objection: "We tried documenting. Nobody read it."
That's usually because the documentation was built wrong. A 40-page process manual is not documentation. It's a monument.
Good process documentation is short, lives where the work happens, and has an owner who keeps it current. It answers three questions: what triggers this process, what are the steps, and who is responsible for each one. If a new person can follow it without asking for help, it works. If they can't, it needs rewriting — not more pages.
Where AI actually helps here
There's an obvious objection to all of this: documenting processes takes time nobody has. This is where AI genuinely earns its place — as long as you're clear about what it's doing.
AI is good at turning raw material into a first draft. Record a twenty-minute screen share of someone actually doing the work, narrating as they go. Feed the transcript to an AI tool and ask it to produce a step-by-step process: trigger, steps, decision points, owner. What took an afternoon of writing now takes twenty minutes of talking and ten minutes of editing.
It's also good at consistency. Once you've written one process well, AI can reformat the others to match it — same structure, same level of detail, same headings. That consistency is what makes documentation usable.
What AI cannot do is know why the process is the way it is. It won't know that step four exists because of a compliance requirement, or that the exception at step six comes from one client's contract. It will produce something fluent and plausible that quietly omits the reason any of it matters.
So the rule is simple: AI writes the draft, the process owner verifies it. Never the reverse. A documented process that nobody checked is worse than no documentation, because people will follow it.
This is the same principle as before — AI multiplies what's already clear. Here, the documenting is the act of getting clear. AI just makes the writing part fast.
Where to start
You don't need to document everything. Start with the processes that meet at least one of these criteria:
Only one person can do it. That's your highest risk.
It goes wrong regularly. The recurring error is telling you something.
New people struggle with it. If onboarding keeps stalling at the same point, that's where the missing documentation is.
Pick three. Document them properly. Assign an owner to each. Review them in three months.
The real cost
The cost of an undocumented process isn't the time it takes to write it down. It's everything that happens because you didn't.
Your business already runs on processes. The only question is whether the company owns them — or whether a few people are holding them for you, until the day they can't.