Construction labour productivity grew about 1% a year for two decades while manufacturing grew 3.6%.

McKinsey Global Institute compared construction against the world economy and against manufacturing over twenty years for Reinventing Construction. Construction labour productivity grew about 1 percent a year. The world economy grew 2.8 percent. Manufacturing grew 3.6 percent. Twenty years of nearly everyone else getting better, and construction close to standing still.

I am 18, I work in electrical preconstruction, and a real share of that gap lives in tasks like mine. A new set arrives and my day starts the way it has always started. Find the fixture schedule on E-001. Go to the lighting plans six sheets later. Then work symbol by symbol across the set, marking each circle as I count it so it does not get counted twice, holding Type F, F1 and FA apart in my head while the phone rings.

Then revision 3 lands on Thursday. One sheet changed, maybe two, but the count is a single number and there is no way to tell which part of it is still good, so the whole set goes back through in order to trust one total. That is the part nobody budgets for. The first count is the work you planned. Every recount after it is the work you did not.

A 1 percent productivity curve is what that looks like once you add it up across an industry. Concrete got modelling. Steel got detailing. Mechanical got fabrication software. Lighting takeoff got a highlighter and a careful eye.

So I built LumenTakeoff for that one job, and this week I could finally show the whole thing running on the demo set in about a minute.

What LumenTakeoff does about it

You drop the PDF. A full multi-discipline set is fine, and a 64-page set processes in about a minute, page by page with progress on screen. Everything runs on your machine and no drawing content leaves it. Scanned or raster pages are reported as unsupported rather than guessed at, and the sheet strip shows every page with an in-scope checkbox, so the non-lighting sheets are unticked before anything gets counted.

Then every repeated symbol on every countable sheet becomes one family, with a ringed crop of one real instance and its exact instance count. It is deterministic on supported vector sheets, because the geometry is already inside the file: the same lines and arcs the engineer placed, repeated at different positions. Grouping is the hard part, so a symbol that differs by two segments stays its own family instead of being folded into the one beside it.

Each fixture family keeps its figures apart and never merges them: detected instances, production count, in bid scope, exact signatures, and how many times the schedule tag appears on the sheet, shown as a document-stated figure with its discrepancy. Naming is one keystroke at a time, most frequent first, with the set's own fixture schedule completing the type code as you type. Enter confirms, Esc marks it not a fixture.

Lock in runs a reconciliation check before anything leaves the tool, and if the counts do not reconcile it refuses to export rather than exporting a wrong number. What comes out is one XLSX with Coverage, Quantities, Findings and Evidence sheets, plus proof images with every counted instance ringed.

The repetition gets automated. The judgment stays where it belongs.

The machine counts. You decide.