The takeoff is measured. The estimate is yours to sign.

The teams that come to us have usually already tried AI on a takeoff, and it failed. That is not surprising, and it is worth being precise about why. A general chatbot reads a drawing the way it reads a photo: it can describe it, but it cannot hold a spatial reference. Two footings sharing one F1 callout, a scale bar that applies to one detail and not the sheet, a symbol legend three pages away. Detailing has no approximation in it, and approximation is all a general model can do on a dense sheet.

Ruh runs takeoff as a purpose-built pipeline instead: sheet classification first, then scale calibration, then measurement and symbol counting with the callouts resolved against the whole set. Every quantity comes out labeled measured, derived, or assumed, with a confidence score and a page-coverage number, so you know exactly how much of the bid is measured versus allowance before it goes out. Partial runs finish honestly as partial. That is the difference between a tool an estimator can sign and a demo that falls apart on real drawings.

Pricing then runs on your book, not a national average. Your assemblies, your labor multipliers, your history, across nine stages from scope and budget through sub bid leveling to a proposal that fills your own workbook. Sub quotes are leveled against scope, and holes are priced as labeled allowances instead of silent zeros. The estimate that comes back is consistent from desk to desk, because the judgment that used to live in one senior head is encoded where every estimator can use it. Exports land in Excel and your own workbook, beside the Procore and Bluebeam workflows your team already runs.

The result is throughput. Quoting volume is revenue for most shops we talk to, and every quote is overhead until it wins. When the first pass arrives measured and priced, the same team prices more work and spends its day on scope gaps and strategy. One structural steel client, ADF Group, measured bid turnaround roughly 50 percent faster. The easiest way to judge any of this is the way estimators judge everything: on your own set. Send one and compare it to your own numbers.

Preliminary and full estimates are treated as the different animals they are. A preliminary number from limited documents runs fast with assumptions labeled loudly, which is exactly what a budget conversation needs. A full estimate runs the complete pipeline: detailed takeoff, pricing, sub bid leveling, and assembly. Estimating leads who reviewed early output told us it holds up against junior estimator work, with flagged rates keeping the accuracy honest, and that is the right bar: the machine does the junior pass at machine speed, and your senior judgment does what it always did.

On accuracy, the honest framing is thresholds and labels, not marketing percentages. Detailing teams benchmark takeoff tools against the high nineties, and the way to earn a place in that company is transparency: every quantity labeled, every low-confidence item flagged for a person, and side-by-side comparison against estimates your team already trusts. Nobody credible promises day-one perfection; teams told us themselves they do not expect everything replaced on day one. What compounds is the correction loop, because every fix your estimators make teaches the system your standards.