FAQ
The questions estimators ask before trusting a tool with their numbers.
On CAD-exported PDFs, every symbol is vector geometry — lines, arcs, circles with exact coordinates. StructCount reads the legend to learn each symbol, then matches that geometry across the sheet as a constellation of strokes: rotation-invariant, scale-aware, verified stroke by stroke. It's the same category of matching a CAD system would do, not image recognition.
The engine is deliberately precision-first: a count only confirms automatically when the match is geometrically distinctive; everything uncertain goes to a review queue for a human decision. That means the numbers in your export were either exact geometry matches or your own explicit confirmations — and each one can be clicked to show its locations on the drawing.
AI reads; it never counts. A vision model can extract legend rows (when the legend is an image, or on request) and can pre-sort the review queue with suggested verdicts. All counting is done by the deterministic matching engine, and all uncertain calls wait for you. If you never touch the AI features, the product works fully without them.
PDF drawing sets. Native vector PDFs (plotted from AutoCAD, Revit, MicroStation…) give the best results. Scanned or rasterized sheets are detected automatically and handled with image template matching plus AI legend reading. DWG import is on the roadmap.
Supported — triage detects raster sheets and switches strategy. You crop the legend once, the model reads the rows, and template matching finds the instances. Results land in the same review queue, so the workflow doesn't change.
Yes. Add a class manually, place a few instances by hand, and Find Similar uses your placements as templates to recover the rest. You can also crop any region of the sheet and count matches of it directly.
Drawings are stored encrypted at rest in the UK (AWS London region), access is scoped to your organization, and your drawings are never used to train AI models. Deleting a set removes the stored file.
Sheet analysis runs in seconds per sheet on typical layouts. The bottleneck is the part that should take time — your review of the flagged edge cases — which is a keyboard-driven pass over thumbnails, not a re-count.
Upload a tender you've already counted by hand and compare.
Get started free