A pile of data is not an asset.

The standard diagnosis for a business that cannot see itself clearly is that it needs more data, better data, or data in one place. So it buys a warehouse, a lake, a dashboard, or one more app for the one problem that app was built for. The pile grows. The decisions do not improve.

This is not a failure of effort. It is a category error. Data is a record of what happened. Understanding is knowing which parts of it matter in this business — and that is a different substance entirely.

Consider what an operator actually knows that no system holds. That this client always asks for extra discovery sessions and never signs a variation. That the country on the back block comes into winter light. That the Friday afternoon shift runs slightly worse and always has. That a volume discount on printed drums is cheaper per unit and more expensive per week of floor space. None of that is in a warehouse. All of it decides outcomes.

General-purpose models have made the gap starker, not smaller. A capable model can explain cattle husbandry, draft a variation clause, or describe scrap-rate management in a plastics plant. What it cannot do is tell you that your feed draw is eleven per cent faster than your own last three winters at this stocking rate — because it has never seen your records, and it has no idea how your business runs.

So there are two halves, and they are usually owned by different things. The data is in the systems. The understanding is in a few people’s heads. The value is only released when the same thing holds both.

That is a harder product to build than a dashboard, for an unglamorous reason: it has to accumulate. It has to observe an operation for long enough to learn what normal looks like here, be corrected when it draws the wrong conclusion, and remember the correction. Useful in the first week, materially sharper by month six.

It also has to show its working. A claim without a source is a guess with better grammar. If a system tells an operations director that a print run is tying up fourteen times its own discount in floor space, it needs to name the unit cost sheet and the storage rate it read — because the only reason to act on it is that you can check it.

The uncomfortable implication for most data projects is that the sequence is backwards. Clean the pile first, understand later, and the understanding never arrives — there is always more pile. Start from the operation instead, and the data you already have turns out to be enough to be useful almost immediately.

A pile is not an asset. A pile that something understands is.