Note extraction refers to the process of pulling scope notes, keynotes, and callouts off construction drawings and organising them into a structured, reviewable list rather than leaving that information scattered across dozens or hundreds of individual sheets where it’s only visible one note at a time as someone flips through the set.
A drawing uses abbreviated messages to convey detailed information about the scope and includes many notes. Some are so detailed and so small that they can be easy to overlook. One sheet can have 20 or 30 notes. Each note outlines a responsibility, material, component, or requirement. When estimating a project’s cost, the estimator knows every single note and can’t rely on a quick review alone.
What a well-organized note extraction captures for each item:
- The note text itself, exactly as written on the drawing
- The sheet and location it appears on
- The CSI division or trade the note most likely belongs to
- Whether that scope trade assignment was confirmed or is still tentative
Auto-assigning a likely responsible trade to each extracted note, based on the note’s content and CSI division, speeds up the process considerably compared to reading and categorizing every note manually, though the assignment still benefits from a human check, since drawing notes occasionally use terminology that could reasonably belong to more than one trade depending on project-specific context that a purely text-based read might miss.
A quality check worth running periodically on any note extraction process, manual or automated: spot-checking a handful of extracted notes against their original drawing location to confirm the extracted text and assigned trade actually match what the sheet shows. This is especially worth doing after a drawing revision, since notes can change wording or location between revisions in ways that an extraction performed against an earlier version might not automatically reflect.
Handwritten field markups and older scanned drawings present real challenges for automated note extraction that clean, native digital drawings don’t, since text recognition accuracy drops considerably with poor scan quality or inconsistent handwriting. Projects working from a mix of document quality levels should expect correspondingly mixed extraction reliability and budget extra manual verification time specifically for the lower-quality portion of the set. Having extraction confidence scores empowers reviewers to mask lower-confidence scores, allowing them to devote valuable time to high-certainty results. This approach also discourages false negatives.
