How AI Automates
Concrete Takeoff

What a concrete takeoff involves and the manual pain

A complete concrete takeoff produces volume in cubic yards by element type — slabs, footings, walls, columns, and piers — plus formwork area in square feet of contact area (SFCA), finish area, and edge or screed linear footage. The geometry is straightforward: a slab is area times thickness, a footing is width times depth times length, a wall is length times height times thickness. The difficulty is that the information is scattered. The foundation plan has the footprint, the footing schedule has the depths, and the section details have the step-downs and thickened edges. An estimator cross-referencing all three on a mid-size commercial foundation package can spend 10 to 25 hours on the takeoff alone.

Step 1 — Plan ingest and sheet classification

When a PDF set is uploaded, AI classifies every sheet by type and isolates the S-series: foundation plans (S1), framing plans, sections and details, and the footing and column schedules. The footing schedule is read to extract each footing mark's plan dimensions and embedment depth. The slab schedule provides thickness per zone and reinforcement callouts consistent with ACI 318 design data. Sections are tagged separately so the system can query them later when plan-view information is ambiguous — for example, when a grade beam depth appears only in a cut section and not on the foundation plan.

Step 2 — Scale detection and calibration

AI reads the scale annotation on each structural sheet and validates it against the column grid before measuring anything. Column grids carry explicit dimension strings between gridlines, so the system can confirm its pixel-to-foot ratio against labeled spacings. If a slab edge callout reads 120 feet and the measured vector at the calibrated scale reads 119.8 feet, confidence is high. If the discrepancy is material, the sheet is flagged for review rather than proceeding with a bad calibration. Calibration runs per sheet — not from a title-sheet scale block — because mixed-scale sets are common and each sheet needs independent treatment.

Step 3 — Object recognition and reading schedules

Vision models trace slab outlines as closed polygons, locate footing marks (F1, CF2, and so on) in plan, detect grade beams and continuous footings as linear elements, and identify columns and piers at their grid intersections. Each detected element is linked back to its schedule entry. When the AI sees footing mark F3, it retrieves the F3 entry from the schedule — say, 2'-6" x 2'-6" plan dimensions at 18 inches deep — and that linkage turns a detection into a quantity. Slab thickenings, depressions, and step-downs are flagged as geometry anomalies and handled as separate volume adjustments; a 12-inch thickened edge versus a 5-inch field slab carries more than twice the concrete per linear foot of perimeter.

Step 4 — Measurement and quantity computation

Slab CY = area x thickness / 27. A 10,000 SF slab at 6 inches is 5,000 cubic feet, or about 185 CY. Footings and grade beams are computed as width x depth x length / 27 per mark; walls as length x height x thickness / 27. Thickened edges and pits are added as incremental volumes. Formwork (SFCA) is computed separately: footing-side formwork is perimeter times depth, wall formwork is both faces times height, column formwork is perimeter times pour height. All SFCA figures are itemized alongside the associated CY in the BOQ, because form assembly and strip is typically priced per SFCA independently of the concrete price.

Step 5 — Assembly mapping, waste, and BOQ output

Volumes are mapped to mix designs from the structural general notes, formwork quantities to form system assemblies, and finish area to the specified finish type — broom, trowel, or exposed aggregate — from the slab schedule. A 5 to 10 percent concrete waste factor is applied to the neat-line volume for over-excavation, spillage, and ordering imprecision; 5% is the default for formed footings and slabs on grade, and the estimator can adjust it. The output is a CSI Division 03 BOQ: slab CY, footing CY by mark, wall CY, column CY, SFCA formwork by element, and finish SF, exportable to Excel.

Step 6 — Estimator review and accuracy

AI is strongest on the elements it can fully read: slab areas from clean plan outlines, and scheduled footing volumes where the schedule provides all dimensions. Slab-area accuracy is typically 96 to 98 percent on clean PDFs. The system is more cautious on elements that require inference — slab thickenings with no section, footing depths absent from the schedule, slopes to drains — which it flags rather than assumes. An estimated dimension wrong by half an inch across 15,000 SF of slab is roughly 23 CY, a material error worth flagging. In practice, estimator review of an AI concrete takeoff takes 1.5 to 3 hours versus 1.5 to 3 days for the same scope done entirely manually.

Element	AI Strength	Watch for
Slab on grade	96–98% area accuracy on clean PDFs	Thickenings not on a section
Scheduled footings	Full schedule read, per-mark CY	Depths missing from schedule
Cast-in-place walls	Length and height traced; formwork computed	Step-outs in section only
Columns & piers	Schedule-driven; perimeter formwork	Capital / base details
Slopes & depressions	Flagged for review	Need section or callout

Questions estimators actually ask