There is a foreman. Call him Johnny. Twenty years on sites. He knows which suppliers will push back if you give them a reason to and which ones won’t. He knows the ground on a particular stretch behaves differently in wet weather. He knows his crew, which groups can be planted on a single task and left to grind through it, and which ones need to keep moving, keep progressing, a different job every few days or they lose the run of themselves. He just knows how things work.
When you need to understand what is actually happening on a site — not the report version, the real version — you call Johnny. And he will tell you.
Nothing here is an argument that replaces that.
But Johnny is one person, on one site. The business has ten sites. Ten foremen, each excellent at what they do, each with their own version of that knowledge. None of them talking to each other in any systematic way. Each one operating with full visibility of their own job and almost none of anyone else’s. That is not a criticism. It is just how it works. It is where the gap is.
Siloed knowledge — and where it costs you
Here is something that happens constantly in construction businesses and almost never gets identified as a data problem.
Three sites running simultaneously. Three experienced foremen, each doing their job flagging issues, escalating what needs to be escalated, keeping things moving. On site one, there is a recurring problem with a particular materials delivery. On site two, the same supplier is causing delays. On site three, the quality on a recent batch isn’t quite right.
Each foreman has reported upward within his own chain. Each issue, on its own, looks like a one-off. A delivery problem. A scheduling irritation. A minor quality concern. Nothing that demands escalation beyond the site itself.
But there is a pattern there. Nobody is raising it because nobody is in a position to see the full picture. The information exists, it just exists in three separate conversations, on three separate sites, in three separate chains of communication. No single person has sight of all three at once.
Each foreman did what he was supposed to do. This is not a communication failure. It is a structural problem. The knowledge is there. The connection is not.
What your spreadsheet can’t tell you
The data that gets analysed in construction has always been the data that fits in a spreadsheet; costs, resourcing, scheduling, procurement, material volumes, delivery dates. Discrete, structured, countable things - that data matters, it is how you run a business.
But it tells you what happened. Not why.
The why has always lived somewhere else. In conversations and site visits. In the phone call you make to Johnny when the numbers are not telling the full story. But by the time you’ve noticed the problem andyou’re making those calls, the money is already gone. You are looking backwards at a problem that has already cost you. Worse, you are only looking at all because something went wrong visibly enough to prompt the question. The patterns that never surface — the slow bleed, the supplier who underdelivers just enough across enough jobs — those never get the phone call, how much is that costing you every year?
What is different now
Natural language — the way people actually speak, the way they describe what is happening around them, the observations and opinions and judgements they express out loud — can now be processed and understood in ways that simply were not practical before.
Large language models are exceptionally good at reading human language. At extracting what is relevant. At identifying connections across time and context that would otherwise sit in separate silos. At spotting patterns of the kind that only become visible when you can read across everything at once.
The running commentary your foremen provide every day — in voice notes, in conversations, in the observations they make as they move around a site — no longer has to disappear. It can be captured, structured, timestamped, and put alongside your costs and your resourcing and your schedule. Not replacing the expertise on the ground. Listening to it. Tying all of that context, the institutional knowledge, the business-relevant judgement calls, together in one place.
The context becomes part of the dataset.
That cross-site supplier problem, the one that looked like three separate one-offs, becomes visible. Not because anyone went looking for it. Because the information was there, and for the first time there was something able to read across all of it at once.
When people leave
People leave. Every business knows this and does what it can to manage it — handovers, documentation, knowledge transfer. Most of it is inadequate. Not out of negligence, but because the knowledge that is hardest to transfer is the knowledge that was never written down in the first place.
Johnny’s twenty years of experience, his instinct for how a job is going to go, his ability to read a situation and make a call — that is his. It stays with him. No system touches it. But that is wisdom. What also lives in Johnny’s head is knowledge — specific, articulable, business-relevant things. That this supplier is unreliable on short-lead orders. That this ground holds water after heavy rain. That this crew needs momentum or it loses its edge. Those things can be expressed in words. If they are being expressed in words regularly, they can be kept.
When Johnny retires or moves to another firm, six months later someone is back on a similar site, facing similar conditions, and the thing that would have saved them two weeks of figuring it out the hard way — the thing Johnny knew — is gone.
Not because it could not have been kept. Because there was no mechanism for keeping it while it was being generated, day by day, in the ordinary course of doing the job.
That mechanism now exists. It will not recover everything. But a business that is consistently capturing the natural language of what is happening on its sites — the observations, the decisions, the context behind the numbers — retains something real when people move on. Not their wisdom. Their knowledge.
And that compounds. Every experienced person who leaves takes something with them. The question is how much of it had to go.
The next post covers how to actually capture this — what works, what doesn’t, and what fits into a working day without asking anyone to change how they work.