You Can't Automate a Report Nobody Agrees On

Ask five project managers for percent complete on the same job and you will get five numbers. All of them defensible. None of them the same.

I have sat in month-end meetings where the schedule said 60 percent, the cost report said 48, and the superintendent said "we're basically done with the hard part." Nobody was lying. They were answering different questions. The scheduler was counting activities. The accountant was counting dollars. The super was counting the things that keep him up at night.

This is the point where a lot of companies decide the problem is the tooling, and the fix is software. Buy a dashboard, wire it to the ERP, and the numbers will finally agree. I build software for a living now, and I can tell you the dashboard does not settle the argument. It just displays the argument in colour.

Reporting is a definitions problem before it is a tooling problem.

On heavy civil and marine work, the honest unit of progress is almost always a field quantity. Piles driven. Cubic metres placed. Metres of pipe in the ground and backfilled. These numbers have two properties that make them worth building on: someone can stand on the site and count them, and they mean the same thing to everyone. "Percent complete" has neither property until you define exactly what it is a percentage of.

So before any automation conversation, I push clients through a boring exercise:

1. Inventory what each PM actually reports today. Not what the template asks for. What they actually send. In most 15 to 30 person contractors this turns up four or five parallel systems: a spreadsheet, a weekly email, a whiteboard photo, and someone's memory.
2. Pick the few numbers that matter and define them in writing. One page. What counts as installed. When a quantity moves from forecast to actual. Who owns each number. This meeting is uncomfortable precisely because it surfaces the disagreements the old process let everyone avoid.
3. Make collection dumb-simple. If capturing the number takes more than a couple of minutes at the end of a shift, the field will stop doing it by the second rainy week. A shared sheet with locked columns beats a clever app nobody opens.
4. Only then automate the assembly. This is the step everyone wants to start with, and it is genuinely the easy part once the inputs are clean.

Here is what changed in the last couple of years: step four used to be a real software project, and now it barely is. Once you have a clean quantity table and a written set of definitions, a language model can draft a perfectly serviceable monthly narrative from it. Variance explanations, trend notes, the paragraph your client's board actually reads. Work that used to eat a PM's Friday now takes a review pass. I have built these pipelines. The model is not the hard part. The model was never the hard part.

But run the steps in the other order and you get something worse than the old spreadsheet chaos: you automate the argument. The AI happily writes fluent prose about numbers nobody trusts, and now the mistrust ships faster and looks more official. A confident summary of a bad number is a downgrade from an honest shrug.

There is a quieter benefit to the definitions work that has nothing to do with AI. When progress means the same thing across every job, you can finally compare jobs. Your bid-versus-actual data starts meaning something. Your forecasting improves because history is finally written in one language. The automation is the visible win; the comparable history is the compounding one.

The fix does not start with a purchase order. It starts with a meeting, a one-page definitions sheet, and the discipline to make five smart people agree on what a number means. That costs you a morning and some awkwardness. Every automation dollar you spend before that morning is spent decorating the disagreement.

If your PMs report progress five different ways and you want a second set of eyes on it, that is exactly the kind of thing I do at Manara. But whether you call me or not: write the definitions sheet. It is the cheapest fix in construction reporting, and almost nobody does it.