AI quantity takeoff from construction drawings — measured by the machine, approved by your sales team
Eight salespeople scale-measuring facade drawings in PDF tools or on printed A3 with pen and paper roughly 1,400 quotes a year, where only ~20% become orders. We apply a vision pipeline that reads the full drawing package and produces a quantity list with confidence scores and traceability, for the salesperson to review instead of measure.
Challenge
The problem pattern.
A request lands: hundreds of pages of PDF, often without a drawing register. A salesperson hunts down the relevant facade, plan and detail drawings, scale-measures them on screen — or prints them in A3 and measures with pen and paper — then sums metres, corners and types into a pricing spreadsheet.
At one typical building-products manufacturer this happens ~1,400 times a year across eight salespeople. They quote ~250,000 metres of product annually and win 50,000–70,000 — meaning most of the manual measuring is spent on quotes that never become orders. The measuring isn't hard. It's slow, fully manual, and it caps how many quotes the team can turn around.
Solution
What we build.
A computer-vision-assisted takeoff — explicitly assistance, not autonomy. A drawing package goes in. A quantity list with full traceability comes out. The decision stays with the salesperson.
- Upload the whole document folder, no pre-sorting — the AI classifies pages (plan / facade / section / other) and finds the relevant drawings even without a register.
- Extraction per view: footprints and dimension arrows from plans; counts per facade cross-validated against plan × storey count; existing PDF annotations read directly from the document layer where present.
- Rule-based derivation of corners, end pieces and posts from segment geometry.
- Every extracted value carries a confidence score and a link back to the exact region on the source drawing. References between main and detail drawings are linked to avoid double counting.
- The salesperson reviews in a verify interface (drawing and proposal side by side), adjusts, and exports the approved list straight into the existing pricing spreadsheet. Pricing logic and ERP stay untouched.
Architecture
How it’s put together.
A staged pipeline: page classification (title-block OCR + visual classifier) → per-view extraction → rule-based derivation → structured output (quantities, types, segment lengths, accessories) with confidence and source references → human verify UI → Excel/CSV export.
Questions about takeoff
Can quantities be taken off construction drawings automatically?
They can be measured by machine and approved by a person. The full drawing package is read at once, every measurement carries a confidence score, and anything uncertain is raised rather than assumed. The salesperson checks a list instead of scale-measuring by hand, which matters most where a large share of quotes never turns into an order.