Two contractors buy the same AI tool in the same month. Same price, same setup, same training. Six months later, one of them is getting real answers out of it and the other one quietly stopped using it. This the video I told you to watch twice.
Nothing in that story is about the tool. It's about what each company could put in front of it. And most contractors have no way to look at their own information and know which of those two companies they're actually going to be. So this video gives you a test.
Four words. Granular, structured, connected, and governed. If your information has all four, AI can do the real work for you. If it's missing one, you'll get answers that sound fine, but don't hold up.
Let's take them one at a time First, Granular means your information exists at the level of detail where the decisions actually get made. Here's the clearest way I can put it. Suppose all your system can tell an AI is this: “The project costs $4.2 million.”
That's true. It's also almost useless. There is nothing in that number to reason about. You can't ask why.
You can't ask what changed. And you can't compare it to anything except another single number. Now suppose your information goes further down. The project breaks down into phases.
Each phase into cost codes, Each cost code into activites, and each activity ties to a crew, to equipment, to production quantities, to labor hours, and to material. That second version isn't only more data, it's a structure full of relationships and relationships are what AI actually reasons over. Once the detail is there, a question like “why did this job cost more than we thought?” has somewhere to go.
The AI can walk down to the cost code, see that production was below plan, look at which crew and which equipment were on it, and come back with an answer instead of a number. This matters more in civil and infrastructure work than almost anywhere else. Your margin lives in production rates and equipment hours. If your information stops at the project level, all that is invisible.
Second. Structured. One quick clarification because I used that word differently in the last video. There I meant information that lives in fields and rows rather than a photo or an email.
Here I mean something more demanding. Structured means the information is specific enough to act on. Compare two sentences.
The first: “We spent more than expected.”
The second: “Labor Cost Code 3200 is 14% above budget because production is 18% below our historical average.” Both are true. Both might come off in the same job in the same week. Only one of them can actually start a conversation.
The first sentence is a feeling. No cost code, no magnitude, no cause, nothing to check, and nothing to do. If that's the level your information lives at, AI will hand it back to you in a nicer font. The second gives it something to work with.
A specific cost code. A number. A comparison against a baseline. and a stated reason that can be verified or challenged.
That's what structured means here. Not tidy. Specific. Here's a useful test. If somebody read this sentence out loud in a Monday morning meeting, would anyone know what to do next?
If the answer is no, AI won't do any better with it. Third, and this is the one that matters most. Connected. Granular gets you the detail, structured gets you the specifics.
Connected is what lets AI reason across the whole story instead of one chapter of it. Think about what has to line up for you to actually understand a job. It starts as an estimate. The estimate becomes a budget.
The budget becomes a project. The project consumes labor and equipment, produces quantities, and generates a cost. That's one continuous story, and it's how the work really happens. Now think about where that information lives at most companies.
The estimate is in one system the budget might be in accounting. Field production is somewhere else. Equipment hours are somewhere else again. Each of those systems is fine on its own, and each can answer questions about its own piece.
An AI looking at only one of them is like an estimator who's allowed to read the bid documents, but not the addenda. Everything they tell you will sound reasonable and some of it will be wrong. When those systems are connected, you can ask a different class of question. Not “what do we spend?”
but this: “We're over on this cost code, and is that because we underbid the production rate because the crew mix changed, or because the equipment was down?” Answering that walking from the estimate, to the budget, to the field, to the shop, and back again. If nothing connects, nobody can walk that path. Not AI and honestly, not your people either, not without a week, a spreadsheet and pattern recognition.
That's the real reason connected information matters. It's not a technical nicety. It's the difference between AI that reports and AI that explains, But connected on its own raises a question we still haven't answered yet. If everything talks to everything, who's allowed to ask it anything?
Fourth. Governed. Here's what that means. Granular, structured, and connected are all about whether AI can reason over your information.
Governed is about whether it should. For this question, from this person, right now, Think about what connected actually built in the last section. One continuous line from your estimate to budget to field data to your cost. That's powerful.
It's also a lot of exposure. If you haven't decided who gets to walk that line and how far. A foreman asking about crew hours on their own job is a different question than a competitor's former employee asking about your margins on every job you've bid this year. The data doesn't know the difference, so somebody has to tell it.
That's what “governed” means in practice. Access that matches the person, not just the password. A superintendent sees their jobs. A controller sees their numbers.
An estimator sees historical bids. Not because the system can't connect everything, we just spent two sections proving it can, but because it can and should are different questions, and only one of them is technical. Right now, for most of you, this mostly means who can pull up which report, but it changes shape fast. The next few videos in this series get into AI that doesn't just answer questions, but goes and does things.
Drafts a document. Flags an anomaly. Takes action on your behalf. When that happens, governed stops meaning who can view this, and starts meaning what is this allowed to do on its own without asking first.
Get that wrong in one way and useful information stays locked away from the people who need it. Get it wrong the other way, and you've built something they can see and eventually touch more of your business than anyone actually decided it should. That's not a future problem. It's the one to get right before any of the rest of this matters.
Now the honest part and this is a real limitation. Granular, structured, connected and governed is a test for whether AI can use your information responsibly. It is not a test for whether that information is right. You can have beautifully detailed cost codes that people have been miscoding for two years.
You can have access rules perfectly locked down and still be protecting numbers nobody bothered to check. Every one of these four qualities can be true, and the underlying data can still be wrong, and the AI will reason confidently over it either way. The answer will look exactly as trustworthy as a correct one. There's a second trap.
More data isn't automatically better data. Detail nobody maintains is worse than a simpler structure everybody keeps accurate, because the detail creates confidence that hasn't been earned. So there's a fifth thing that doesn't fit neatly into a checklist, and I'd rather say it than leave it out. Somebody still has to own this whether information is actually true.
That part isn't a technology problem. It's a discipline problem, and it's the least exciting sentence in this entire program. Skip it, and everything else in this video stops mattering. So here's the test, and it's worth writing down.
Granular. Does the detail go down to where the decisions get made? Structured. Is it specific enough to act on?
Connected. Can it be followed across systems from estimate all the way to cost? And Goverened. Does the right person or system have access and only the right one?
Run your own company through those four. Most contractors find they’re strong on one or two and thin on the rest, and knowing which is which tells you more about your readiness [for]
AI than any product demo will. Here’s why this is the hinge of the whole program. Everything from here gets more useful the better your answers are. The tools in the next two videos work harder when they have context.
The use cases after that assume you can get at the information, and the agents we cover near the end, the ones that actually go and do the work only become safe when all four of these are true, especially that last one. In the next video, we get practical. Generative AI. What you can actually do with it today and how to get more out of it than a summary.
