Case study

A 71-sheet drawing set,
turned into a project brain
in one AI pass.

A mid-market general contractor handed us a full multi-discipline construction document set. One automated pass turned it into trade takeoffs, coordination findings, and ready-to-send RFIs. An afternoon of compute, not a week and a half of skilled labor.

71
sheets analyzed
225
takeoff line items
13
conflicts caught
~90%
of the time saved
The challenge

Days of skilled work, before a single bid goes out

A full set lands: civil, structural, architectural, mechanical, plumbing, fire, electrical. Someone has to read all of it, take off the quantities, catch the conflicts between disciplines, and write the RFIs. On a set this size that is roughly a week and a half of a project manager and an estimator's time, and the conflicts that slip through become change orders in the field. We pointed one automated pass at the PDF.

1

Analyze

Reads every sheet: type, scale, schedules, materials, cross-references.

2

Structure

Builds a queryable database and concept wiki of the set.

3

Take off

All-trade quantities, every line confidence-scored.

4

Coordinate

Surfaces cross-sheet conflicts automatically.

5

Draft RFIs

Writes formal RFIs and routes them to the right designer.

See it work

It reads the drawing and marks it up

The contractor opens a dense foundation plan. Seconds later the AI has read every footing tag off the vector layer and placed all 80 footings and pedestals with coordinates. That placement is the start of the takeoff, and it is exact, not eyeballed.

Before. The raw sheet
S-201_foundation_plan.pdf
raw foundation plan
After. AI-marked, in seconds
S-201_marked_up.pdf
AI-marked foundation plan

Foundation plan, title block removed for confidentiality. Marker color and labels are the AI's, one per counted footing.

What it produced

Two layers: a brief for every sheet, and a database you can query

Layer 1: every sheet understood

  • 71 per-sheet briefs covering title block, scale, schedules, materials, and cross-references.
  • 703 legend symbols catalogued across 6 disciplines.
  • Each sheet tagged with what it answers precisely versus by inspection.
  • Raster shop-drawing sheets flagged, never faked.
Sheets by discipline
Architectural19Civil13Electrical12Structural11Plumbing7Mechanical5Fire3Cover1

Layer 2: a queryable project database

Eight tables: sheets, schedules, instances, relationships, design basis, notes, placeholders, coordination. Every fact carries a reliability score.

86
schedules
85
instances
11
relationships
16
design basis
Ask it, "how many F-10 footings, and where?"
SELECT tag, grid FROM instances WHERE tag='F10';
An exact, sourced answer in seconds. The kind of question a flat PDF cannot answer.
Before and after

Every workstream, and the hours it gives back

The same pass produced the real deliverables below. The left side is the manual reality. The right side is what the AI handed back, shown exactly as it came out (anonymized).

Quantity takeoff
Measure and tally every trade across the set and key it into a spreadsheet. The one line everyone misses costs margin.
~32 hrs~1 hr + review
saves ~31 hrs
After. 225 line items, 7 trades, confidence-scored
Takeoff_Register.xlsx
takeoff register
Coordination and QA
Cross-check 7 disciplines against each other across 71 sheets, by eye. The conflicts that slip through become change orders in the field.
~12 hrsminutes
saves ~12 hrs
After. 13 conflicts flagged, with severity
Coordination_Flags.xlsx
coordination flags
RFIs
Write up each issue, cite the exact drawings, propose a fix, and route it to the right design firm. Fourteen of them.
~10 hrsminutes
saves ~10 hrs
After. Formal RFI, drawing-cited, ready to send
RFI-008.docx
formal RFI
Honest by design

It scores its own confidence, and flags what it cannot prove

High 68Medium 74Low 83
225line items

HIGH from vector tags and schedules. MEDIUM from vision or a single source. LOW where scaled or derived.

What it deliberately will not fake

  • Earthwork cut and fill. Needs a topo and grading model.
  • The metal building superstructure. Delegated to the manufacturer.
  • Sprinkler heads. Design-build by the fire protection contractor.
  • Pipe and duct runs. Flagged as plan-view estimates.

It quantifies what it can prove and flags the rest. The estimator verifies before pricing.

The value

About a week and a half of work, produced in an afternoon

~64 hrs
first-pass work compressed
~90%
of that time recovered
$5-7.5k
skilled labor per set*
13
issues caught pre-field
ManualWith Fulcrum AI
10h0.3hUnderstandtheset32h1.2hTakeoff12h0.3hCoordination10h0.4hRFIdrafting

*Manual hours are a conservative first-pass estimate for a project manager and estimator on a 71-sheet set. The AI pass produces the same first-pass output in a few hours, including human review of every flagged item before anything is priced or sent. Labor value at a blended rate near $95 per hour loaded. Figures are illustrative. Your set, trades, and rates will differ.

Fulcrum Foundry

This was one drawing set.

The same workflow runs on every bid. Faster turnaround, fewer missed scope items, cleaner margins.

Get In Touch
Fulcrum Foundry builds AI workflows that create leverage in your business, on your real projects.