A quantity takeoff you can defend line by line
The situation
Estimating is where small contractors drown. Plans arrive, the bid window is short, and jobs go un-bid for lack of hours rather than lack of capability. Students heading into construction management need both halves of this: how to get quantities out of a drawing set quickly, and how to keep the assumption trail that makes a number defensible when the job is running and someone asks why the figure was what it was.
Steps
-
Write the assumptions before the first number
A one-page assumption sheet
Thickness where the drawings are ambiguous, which materials get counted separately, waste percentage, what is included and excluded. This list is the deliverable that outlives the bid.
What you only learn by doing it: State assumptions in the prompt itself rather than asking the model to choose. Left to decide, it picks something plausible and does not tell you which, and then your number has no provenance.
-
Read the set yourself and mark the scope first
The drawing set, and a marker
Go through the drawings and mark what is in and out before any automated help. You are looking for the things a model reliably misses: revisions, phasing, anything shown on one sheet and contradicted on another.
What you only learn by doing it: Note the sheet numbers and revision dates you worked from. Bidding against a superseded set is a failure mode no tool protects you from.
-
Ask for quantities by material, separately, totals only
OpenAI for Education Julius AI
Request each material as its own figure rather than one combined volume, and ask for totals rather than the full derivation on the first pass. Mixed materials in one number cannot be priced and cannot be checked.
What you only learn by doing it: Costello’s phrasing is the model to copy: calculate concrete, pavers and stone separately, assume the larger thickness, add 30% for fluff, report totals only. Specific instructions, specific assumptions, bounded output.
-
Re-derive two numbers by hand
Wolfram|Alpha Pro and Wolfram AI Assistant
Pick the largest quantity and one at random. Compute them yourself off the drawing. If both land within a few percent, the method held for this set. If either does not, nothing else on the sheet is trustworthy.
What you only learn by doing it: Always include the biggest line. An error there outweighs everything else combined, and it is the one a reviewer checks first.
-
Price it, then compare against a job you actually ran
Apply your unit costs, then set the result beside a completed job of similar type and look at the deltas. Historical comparison catches whole missing scopes in a way that line-item review does not.
What you only learn by doing it: Costello runs finished job costs back against the original estimate to find deviations. Doing that routinely is what turns one bid into an estimating practice that gets better.
-
Keep the assumption sheet with the bid, permanently
Your project file
File the assumptions with the submitted number. When the scope is disputed mid-job, the assumption sheet is the document that decides whether you are owed a change order or absorbing the difference.
What you only learn by doing it: Teach this as the professional artifact, not the total. Anyone can produce a number; the assumption trail is what makes it a bid.
Where this breaks down
A model reading drawings misreads scale, misses revision clouds, and takes a detail callout for a condition that exists everywhere. It will also answer confidently about a sheet it only partly parsed. Every number it returns is a draft until you have located it on the drawing yourself.
A bid is a contract offer. A takeoff error does not produce a bad grade, it produces a loss carried for the length of the job. That asymmetry is the lesson, and it is worth making students feel it before they feel it at work.
Waste and conversion factors are judgement, not arithmetic, and they vary by crew, site and material. Write yours down as assumptions rather than burying them, so a reviewer can argue with the assumption instead of the total.
Hazardous materials are out of scope for this entirely. Asbestos, lead and contaminated soil are survey-and-abatement questions governed by law and performed by licensed people. Nothing generated here touches them.
Provenance: adapted from James Costello, structural engineer and demolition contractor, interviewed in OpenAI’s ChatGPT for Pros newsletter, 5 August 2026. Costello estimates bids largely alone, stating his assumptions explicitly in the prompt — slab thickness, materials counted separately, plus 30% for fluff — and says the practical effect is volume: “Many of these projects wouldn’t be bid due to time constraints.” The sequence, the checkpoints and the cautions below are our construction, written for a classroom. Treat the source as one practitioner’s documented experience in a vendor publication, not as evidence that this works generally.