Your Company Keeps Books on the Money—and Nothing Else
Somewhere in your company this week, a written procedure will fail, and no one will notice.
An order will arrive with a field left blank. Someone will recognize the customer, recall the special circumstances for this account, and fill it in correctly without pausing. No ticket will be filed. No exception will be logged. The work will move on.
Management will assume the process works.
The process does not work. The person works. What if that person leaves? What if she’s out sick? What if she wants a promotion, but is too much of a linchpin in her current position?
Every serious company keeps two kinds of knowledge, but treats them with wildly unequal seriousness. The first is written down, reconciled monthly, and legible to any competent outsider including lenders, auditors, and the new controller who arrived on Tuesday. The second is how the work actually happens, and it lives in heads, habits, inboxes, and the phrase Sarah normally handles that.
Financial books let you inspect claims about money. Very little in your company is designed to let anyone inspect claims about work.
The friar’s contribution
The above wasn’t always true about the money either.
In 1494, a Franciscan friar named Luca Pacioli published the first widely read description of double-entry bookkeeping. He did not invent it (Venetian merchants had used versions of it for generations) but he did codify it… which is the more interesting bit here. Codification took a practice that lived in the hands of men who had learned it by standing next to other men, and made it teachable, checkable, and portable beyond any particular person.
In double-entry accounting, every transaction is recorded as equal but opposite changes to different accounts—debits, credits, and assets—so the books must balance. If they do not, there’s something wrong.
That self-test is what made everything else possible. A lender could inspect the books. An investor could evaluate the enterprise. A manager 300 miles from the transaction could understand what had happened without reconstructing it from the founder’s recollection. Credit, audit, the joint-stock company… all of it downstream of a business becoming representable outside the heads of the people running it. Visible came first. Scalable followed.
But 500 years later, the operational side of your company still runs pre-Pacioli.
Why the oral tradition survives
Because it works. Or rather: because your people make it work.
Every written procedure is incomplete. That’s just life. Reality contains more cases than any document contains clauses. Karl Polanyi gave us the general form of the problem (we can know more than we can tell)—and it’s as true of institutions as of individuals.
So the distance between what the procedure says and what the work requires gets crossed, every day, by someone who reads the situation and supplies the missing clause. The procedure says manager approval required. The veteran knows this means Dave, that Dave approves anything under 15%, that above 15 you walk it through Finance first or he will be unpleasant about it, and that during close week you don’t ask at all unless the building is on fire.
None of that is written anywhere, but that doesn’t mean it’s not load-bearing.
That daily act of interpretation is the most underpriced input in your business. It appears on no balance sheet. It’s why an undocumented company can feel healthy for a decade (or more). The documentation debt is still real, it’s just being serviced, quietly, by experienced people out of their own attention.
And here’s the perversity at the center of the thing: the better your team, the more of that debt you can carry. Strong operators absorb ambiguity. They remember exceptions, repair broken handoffs, and know whom to ask. Because they keep the system running, the company feels no pressure to make the system explicit. Competence conceals the liability. Your best people are the reason your operational books do not exist.
If it works, why change now?
Well, you want AI to help optimize your business. You paid for the subscription—but it can’t do the job.
An LLM is not stupid about ambiguity per se. It is, however, unsituated.
It wasn’t in the room when the customer made the original request. It can’t see that no one has used the official form since March, that the documented approval chain is routinely bypassed, or that Dave’s title badly understates his authority. It can’t swing by someone’s desk to get a quick check or search the company SharePoint with a real hope of seeing what’s real versus what’s in forever-draft. It sees what you hand it, and nothing else.
Give it an incomplete procedure and it will not stop and ask. It will produce a confident, plausible completion, and you will not be able to distinguish that from a correct one until the invoice goes to the wrong entity.
That’s the difference. A confused employee asks a question. A confused model produces a bad and overconfident guess.
Which is why the demo dazzles and the pilot dies.
The demo proves the model can perform the visible task. The pilot discovers that the visible task rested on an invisible institution of exceptions, relationships, and tacit authority that nobody ever wrote down.
I’ve watched this happen from the other side of the same problem. An ERP implementation and an AI deployment look like different species of project. They aren’t. Both eventually ask the company a single question: how does this business actually work? And both are astonished by the answer, which is: it depends who you ask. That’s why every serious system project opens with an archaeological dig, interviewing employees to recover an operating model the company has used every day for 15 years and has never once represented coherently.
AI didn’t create this deficiency, but it has repriced it. The tacit layer was always a cost. But it was unbilled because your people were paying it in memory and improvisation. Now you want to delegate work to something that you do not want improvising, and the bill has arrived.
You’re falling behind because the economy is shifting and your business isn’t set up for AI—yet.
Put your operations on the books
None of this is an argument for eliminating judgment. No operating model survives contact with a sufficiently strange customer, and the goal was never to reduce the business to a flowchart. The goal is to know where judgment is required, and to give the person exercising it enough context to be right more often than not.
So take one consequential workflow, like onboarding a new client, and walk it through from start to finish. Make sure there are no gaps, everyone and everywhere work happens is captured, and “what to do” is so clear, a computer could follow it. Figure out where decisions get recorded, where the evidence is stored, and what happens when something falls outside the rule. Then put that answer somewhere a capable stranger could find it without asking Dave.
That’s the start of your operational books.
If you are staring down an ERP cutover, an AI pilot, or a key departure, and you suspect the operational books do not exist, reach out! That’s my specialty.