Operational data is the record of the real things your operation handles and the real actions it takes on them: the order, the lot, the machine, the work centre; scheduling, inspecting, holding, shipping. Most of it is already stored somewhere. Almost none of it is written down in words the whole building agrees on, which is why software can hold ten years of your history and still give you answers that would fit any plant anywhere. Before software can use your history, it needs your words, and writing them down is a job for people. I sat in on a workshop last spring where four people spent twenty minutes on the word shipped. Scheduling meant the truck left the yard. Finance meant the invoice went out. The warehouse meant the pallet was staged at the dock with the paperwork printed. Quality meant the hold was released. They had all worked in the same building for years, and not one of them was wrong. They had simply never needed the same definition, because a person had always been standing between the systems doing the translation by hand, for free, invisibly, a few hundred times a week. I am the marketing person in a room like that. I have never run a line and I have never closed a month. What I do is words, and that argument was entirely about words. Why operational data gets stored but stays unusable Two things are true in most operations at the same time. We have all the data. We cannot use the data. Both are true because storage and meaning are separate problems, and only one of them gets bought. A database has a table with a column called STATUS holding the value 4. Somewhere in the building a person knows that 4 means held for quality, that a hold on this product line needs a disposition before release, and that the disposition lives in a different system under a different record number. That knowledge is real, and the operation depends on it. It is also nowhere in the data. So the meaning gets rebuilt by hand, every time. An analyst writes a report and encodes the translation in a query nobody else can read. A supervisor knows which of the two dates on the screen is the one that matters. A quality lead can look at a record and tell you in four seconds what an auditor would ask about it. Each of those is a small act of translation performed by a human being, and none of it gets written down where the next person can find it. What counts as a noun A noun is something your operation can point at, count, and argue about. It gets created, it changes state, and it ends. In a plant that is the order, the lot, the work centre, the machine, the deviation. In a hospital it is the patient, the bed, the clinician, the discharge. On a campus it is the student, the section, the transcript, the cohort. In a benefits program it is the case, the applicant, the approval. A useful test: if two systems both hold it and the two disagree, it is a noun and it needs one definition with a name attached to it. Reports and dashboards are not nouns. They are screens that show nouns. The screen that shows late orders is not a thing your operation runs on. The order is. The other test is history. A noun is something you can ask a question about across time. How many times has this work centre caused a hold. How often does this transfer credit fail to reconcile. If you cannot ask that question because there is no stable thing to ask it about, you have fields, not nouns. What counts as a verb Verbs are the state changes. Schedule, release, hold, inspect, ship, admit, discharge, enroll, approve, escalate. Every one of them has a before and an after and somebody accountable for it happening. For each verb, write down who is allowed to do it, what has to be true before it can happen, what it changes about the noun, and what record it leaves behind. That sounds tedious until you try it, and then the first verb takes twenty minutes because you have run into the same disagreement those four people had about shipped. Verbs are where the disagreements show up, because a verb is usually the point where one department hands the work to another. Nouns tend to be shared. Verbs tend to be owned, and the owner has a definition that suits their job. None of this is a technology exercise. It is a vocabulary exercise, and it is the reason the same person keeps getting pulled into every project. How to write the list this week An hour. A whiteboard. Three people who do not report to each other and are known to disagree. Pick one thing that moves through the whole operation from the first touch to the last. An order, a batch, an admission, an application. Not the easiest one. Pick the one people argue about. Write down every noun that thing touches on its way through. Every real thing, in the words people actually say out loud. Write down every verb that happens to it, in sequence. For each verb, write who does it and what has to be true first. One line each. Circle every word where the room produced more than one definition. For each circled word, write the definition you are going to use and the name of the person who owns it. List what turned out to have no name at all. There is always something. Then run the test that matters. Hand the page to somebody who joined three months ago and ask them to read it back to you in their own words. If they can, you have a vocabulary. If they cannot, you have a page of internal shorthand and you have just learned what onboarding actually costs you. Keep it to one page. The version that takes six weeks and produces forty diagrams is a different project, and it usually stalls. What the list is worth once it exists The obvious returns come first and they are not about AI at all. New people get productive faster because the translation is written down instead of being learned one question at a time. Audit responses get shorter, because somebody can point at the record that answers the question rather than rebuilding the answer from scratch. Reporting requests stop turning into three-way meetings about which date is the real date. Then there is the AI part, which is what most people are asking about anyway. Point a model at a pile of tables and you get answers about tables, delivered with total confidence and no relationship to your operation. Point it at operational data organised around your real things and real actions, and it can reason about the order that is late because the supplier shorted you, or the discharge stuck waiting on a signature. Same model. The difference is whether your business was ever described to it. Your operation already has a vocabulary. It is just not written down. It lives in the four people who can translate between systems without noticing they are doing it, and who answer their phones because they always have. They will not always be in the building. The list is cheap to write while they still are. This week: Take one thing that moves through your whole operation, write down every noun and verb it touches, and mark every word that turns out to have two definitions.