Benchmark

How fast is automated symbol takeoff?

SimplyCount counted 60 unique symbols across 168 pages in about one hour — roughly 168× faster than counting by hand.

Result

168
pages in the set
60
unique symbols
~1 hr
including review
168×
vs hand-counting

Methodology

A 168-page commercial plan set with 60 distinct symbol types was processed end to end: import the PDF, point at each symbol from the legend, let the engine detect every instance across all sheets, then review the marked sheet symbol by symbol. Elapsed time — including that review — was about one hour, entirely on a local machine — the estimator’s own desktop.

How we state it honestly — the assumption behind the multiplier

Review here means per-symbol batch acceptance: for each symbol, the estimator scans the marked sheet and accepts or rejects that symbol’s detections — not a one-by-one confirmation of every individual mark.

The measured side of the claim is the hour: 168 pages, 60 symbol types, counted and reviewed. The baseline side is an assumption we state rather than hide: it assumes ~45–60 minutes per page to count and record by hand, first pass only — no second passes or rechecks. The multiplier uses the top of that range (a page-hour): about 168 working hours, i.e. weeks of working days, for this set. “168× faster” is that arithmetic, nothing more; your baseline will vary — if your hand counts run faster than a page an hour, scale the claim down accordingly — the hour on our side is the measured part. And because the baseline is per-page, the multiplier scales with page count: 168× is a property of the assumed baseline and the size of this 168-page set, not a property of the engine.

Accuracy, stated honestly

Accuracy was measured against a hand-verified ground-truth set. After review — the per-symbol accept/reject pass that is part of the workflow — the exported count matched that ground truth exactly: 100%. That is by design: review exists to catch what the engine misses, so the number you export is the number you verified. We do not publish a raw engine-recall percentage — the count that matters is the one you sign off, not one you have to take on faith. The stronger test is your own plan set: request free access and run one.

This benchmark is the one time anchor for every speed claim on this site. When the homepage says hours of hand-counting become minutes, or a multi-day takeoff becomes an afternoon, those are this benchmark restated — not separate measurements. Each symbol’s detections are reviewed and signed off by a person, and any count that isn’t fully verified is shown as a floor (≥), never dressed up as a total.

See it on your own plan set

Request early access with your work email — the team reviews every request, nearest bid deadlines first — then run your first plan and see a number you can defend. 168 pages · 60 symbols · ~1 hour.

Free during early access · Your work email · No credit card · Counting runs on your machine

See it countRequest Early Access