Every company has one. The person everyone Slacks when something odd happens. The one who knows which customer gets net-60 terms, which supplier ships short, and the exact sequence of clicks that makes the month-end report balance. When they're out for a week, throughput drops. When they leave, you find out how much of your operation was never written down.
Most owners know this is a risk. Fewer fix it, because the standard remedy — "let's document our processes" — usually produces a folder of PDFs that nobody opens after week three. The problem isn't effort. It's method.
Tribal knowledge isn't laziness — it's compression
Experienced operators don't skip documentation out of carelessness. They compress. After the two-hundredth time doing a task, the decision points stop feeling like decisions. Ask your best person to write down how they handle a customer credit and you'll get six steps. Watch them do it and you'll count twenty, including three judgment calls they didn't mention because those felt obvious.
That gap is where the failures live. New hires follow the six steps, hit the first unlisted exception, and either guess or interrupt someone. Both outcomes cost you.
So stop asking people to write down what they do. Capture them doing it.
Record the work, then let AI draft the document
This is where the economics have genuinely changed. Documentation used to be expensive because the writing was expensive. Now the cheapest reliable method looks like this:
Sit with the person while they do the real task, on real data, and record it — screen and voice. Ask them to narrate, and interrupt with one question whenever they make a choice: how did you know to do that? That question is the whole interview. It surfaces the compressed judgment.
Then run the recording through transcription and use AI to produce a first-draft procedure: steps, decision points, exceptions, systems touched. The draft will be imperfect. It will also be eighty percent of the work, produced in minutes instead of days. The expert reviews and corrects rather than authoring from a blank page — a task most people will actually complete.
The leverage here isn't the AI. It's that AI removes the excuse that documentation takes too long.
Push the process into the system, not into a PDF
A written SOP is a suggestion. A system is a constraint. If you stop at the document, you've created a reference that decays quietly while the real process drifts.
The better move is to encode the procedure where the work already happens. In a well-configured ERP, that looks like required fields that can't be skipped, approval steps that route automatically, stages that won't advance until the prerequisite is done, and templates that carry the right terms and defaults. The knowledge stops being something people have to remember and becomes something the software enforces.
This is also where the exception handling belongs. If your expert knows that orders over a certain value need a second signature, that shouldn't be a line in a document — it should be a rule in the workflow.
Sequence it by risk, not by org chart
You can't capture everything, and you shouldn't try. Start where the exposure is highest:
- Processes where exactly one person knows how it works
- Handoffs between teams that generate rework or rekeying
- Anything touching cash: quoting, credit, collections, payroll
- Steps that stall when the owner is unavailable
- The tasks new hires ask the most questions about
Five well-encoded processes beat fifty documented ones. Pick the ones that would hurt most if the person behind them took a job elsewhere on Monday.
Give maintenance an owner and a trigger
Documentation projects die from lack of ownership. Assign each process a named owner and tie review to an event, not a calendar reminder nobody honors. Good triggers: a system change, a pricing change, a new hire in that role, or a complaint that traces back to that step. When the process changes in reality, it changes in the system the same week — or the system stops being trusted, and everyone goes back to asking the person who knows.
The quiet payoff: a business that runs without you
This work rarely feels urgent, which is why it gets deferred. But it's the difference between a business that scales and one that just gets busier. It's what lets you hire competently instead of hoping the new person is intuitive. It's what makes delegation real rather than nominal. And if you ever sell, it's the first thing a buyer probes — because a company where the process lives in three people's heads is worth less than one where it lives in the system.
At Infraxio, this is the work we do daily: sitting with operators, extracting how things actually get done, and building it into Odoo, integrations, and workflow so the knowledge stays with the business. Decades of operator experience, sharpened by modern AI tooling, applied to your specific mess.
If there's a person in your company you can't afford to lose, that's the conversation to have with us. Reach out and let's map where your knowledge is trapped — and what it takes to get it into systems that hold.