Estimating a job with AI: what the tool does, and what it leaves to the estimator

An estimator spends a good part of the week measuring drawings: areas, lengths and rooms, one by one, in PDF files. That is the work AI promises to speed up. The promise only holds if the estimator can trust the result.

The problem with a guessed number

A tool that returns a total without showing where each quantity comes from forces the estimator to redo everything. Worse, a quantity made up by a language model can hide inside a total that looks right. In a bid, a takeoff error becomes a pricing error.

Mensor’s rule

Mensor is built on one simple rule: every quantity is measured on the drawing, or else flagged to the estimator.

  • A green quantity is measured and backed by a dimension written on the drawing.
  • A yellow quantity has no written proof. It never turns green on its own.
  • A red quantity must be corrected by the estimator.

The AI model reads the drawing and finds the rooms, dimensions and scales. It never writes a number in the bill of quantities: quantities are calculated by code, from what was measured. A scanned drawing that the tool cannot read with certainty is rejected rather than guessed.

What stays with the estimator

The estimator fixes the red items directly on the annotated drawing, by moving an outline or tracing a foundation. The estimator handles the rooms that have no written dimension. Above all, the estimator keeps the judgment: reading the specifications, the addenda and the site conditions, which no drawing tells in full.

Where Mensor stands

Mensor is in development. We are aiming for vector PDF drawings as input and a bill of quantities as output, exportable to CSV, Excel and annotated PDF, organized by MasterFormat. We are also working on access from assistants such as Claude or ChatGPT.

Are you an estimator who wants the tool to fit the way you work? Write to me. Your drawings and your takeoff habits will help us build it right.