Industries / Construction and property / Commercial and residential construction / ConTech
Estimating
Estimating is the work of measuring a project's drawings to find how much of each material and task it needs (the takeoff), putting a price on those quantities, and turning the total into a bid, along with the software and online networks builders use to do it and to collect prices from subcontractors.
7
Milestones achieved, from 1980 to 2018.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1980
Estimates move onto computers
Timberline Software of Oregon put construction cost estimating on minicomputers, the mid-sized business computers of the time, with a program called AccuBid. The International Directory of Company Histories says Curtis Peltz designed it in 1980 and that its PC successor, Precision Estimating, followed in 1987; Sage dates a Windows version to the early 1990s. In July 2003 the UK's Sage Group and its US unit Best Software agreed to buy Timberline for US$102.9 million in cash, according to ENR, and Sage still sells the descendant as Sage Estimating.
Credit
Curtis Peltz, Timberline Software
1984
Measuring plans with a pen
By hand, an estimator measured each area on a paper blueprint with a ruler. A digitizer is an electronic tablet that records where a pen touches it, so the estimator could touch the corners of an area on the plan while a computer worked out lengths and areas. Marsh V. Racine filed a US patent in April 1984, granted in March 1986, for a "Computer Aided Coordinate Digitizing System for reading plans and calculating desired quantities." Mechanical contractor H. Kent Craig wrote in 1999 that in 1982 and 1983 he helped build "the second digitizer-based computerized construction estimating system ever created." A patent filed in 2000 still described both, the ruler and the digitizer pen, as the existing ways to do a takeoff.
Credit
Marsh V. Racine (patent); the builders of the earliest systems are not named in the sources found
1990s
Takeoff moves onto the screen
On Center Software, which says it has operated since 1988, sold On-Screen Takeoff, a program that measures directly from drawing files on a computer screen instead of from paper on a digitizer. On Center calls it "the first digital takeoff software for the preconstruction industry." Vertigraph, a rival takeoff vendor, wrote in 2020 that On-Screen Takeoff was the first product to measure quantities from drawing files, and that once a tablet was no longer needed, "digitizer tablet takeoff became obsolete." On Center has been part of ConstructConnect since 2016.
Credit
On Center Software
2000
A formula skips a row
J. Kokolakis Contracting bid US$4,242,500 to build a dining facility at MacDill Air Force Base in Florida, the lowest of eight bids. Its spreadsheet formula picked up the "site work package" prices from the eighth line instead of the seventh, so those prices never reached the totals, and the firm asked to raise its bid by US$330,500. The US Army Corps of Engineers let the firm raise its bid, and in January 2000 the US General Accounting Office (GAO), which rules on federal bid protests, upheld that decision because the saved spreadsheets and formulas showed both the mistake and the intended price. GAO said records of computer spreadsheets can count as evidence if they are "in good order."
LessonEvery total needs a check that it includes every row, and the original workbook and its formulas should be kept, because they become the proof when a mistake has to be shown.
Credit
GAO decision B-283857, Construction Technology Group, Inc
2005
A late quote goes missing
Shortly before bids closed at 9 a.m. on 10 October 2000 for Milwaukee's Zoo Interchange, a subcontractor for the bridge concrete work phoned James Cape & Sons to raise its price from US$425 to US$555 per cubic yard. The change never reached the handwritten tally that fed Cape's computer-generated bid of US$16,332,873.75, which the court says was US$450,450 too low as a result. Wisconsin's transport department refused a correction, awarded Cape the job and, when Cape refused to sign, kept its US$100,000 bid deposit. In July 2005 the Wisconsin Supreme Court held that a public owner there may not let a bidder "correct" a mistaken bid, but that Cape had met the rules to withdraw it and should get its deposit back.
LessonBid-day changes need a check that every revised quote reached the final number, because the law may let a bidder walk away from a mistake but not fix it..pdf).
Credit
Wisconsin Supreme Court, James Cape & Sons Co. v. Mulcahy, 2005 WI 128
2011
Models arrive, estimators wait
Building information modelling (BIM) means drawing a building as a 3D computer model whose parts carry data, so quantities can be counted straight from the model instead of measured from drawings. A 2011 survey of US construction practitioners by Anoop Sattineni and R. Harrison Bradford II found that "while most contractors were using BIM, only a small minority were using it to automate estimating processes." Most organizations said their BIM staff lacked the estimating experience to produce working estimates from models, and the reverse.
LessonModel-based takeoff depends on people who know both modelling and estimating, not only on software.
Credit
Anoop Sattineni and R. Harrison Bradford II, Proceedings of the 28th ISARC, Seoul
2018
Bid invitations join one network
A general contractor, the firm that signs with the owner, gets a price for each trade by asking subcontractors to bid. BuildingConnected, founded in San Francisco in 2012, built an online bid-management network where owners and general contractors find and hire subcontractors. In December 2018 Autodesk agreed to buy it for US$275 million net of cash acquired, closing in January 2019, and said its network had "more than 700,000 construction professionals."
Credit
BuildingConnected (co-founder and chief executive Dustin DeVan); Autodesk
Watch for
Scope sheets drafted by AI
A scope sheet lists the work a trade's price must include, and estimators usually write it by reading the drawings and specifications. Provision Software Corporation of Ontario says its Scope Agent drafts these sheets and reached "97.01% match accuracy on scope item extraction" on its own internal validation dataset, measured against human-prepared scope sheets from real projects; its human baseline scored 91.3% and took roughly four days per project. Provision also says the general contractor EllisDon's use of the output on the New Brunswick Museum project is "the first documented case" of AI-generated scope going straight into a bid package at a top-tier general contractor. The release does not say how many projects were tested or name an outside checker.
Achieved whena general contractor or researcher outside Provision tests it on projects Provision did not choose and publishes the match rate against estimator-written scope sheets.
Claimed by
Provision Software Corporation, August 2026
AI levels subcontractor bids
Bid leveling means lining up subcontractor quotes so they cover the same work and can be compared fairly. MeltPlan says its Melt Bid tool reached 95.3% overall accuracy against expert estimators' own leveled answers on 100 bid proposals across 40 packages, scored over 8,500 data points, against 75 to 80% for general AI models from Anthropic (Claude), OpenAI and Google (Gemini). The study is MeltPlan's own, and the article reporting it was produced in partnership with MeltPlan.
Achieved whenan outside party runs the tool on quotes it did not pick, including gaps that every bidder left out, and publishes the results.
Claimed by
MeltPlan, reported August 2026
Bid review in minutes
Zebel, a preconstruction software company in Walnut Creek, California, launched AI Bid Review, which reads incoming subcontractor bids, maps each number onto the bid form tied to the estimate, flags scope omissions, and applies nothing until the estimator approves. Zebel says reviewing and leveling one trade by hand takes "an estimator about an hour," more than a working week across a project, and that in Zebel "the same review takes minutes." It gives no accuracy figure, and a quoted customer said "only time will tell" whether it keeps up across hundreds of bids.
Achieved whena customer or outside party publishes measured review times and error counts over a large number of real bids.
Claimed by
Zebel, October 2026
Open
01
A planted-gap bid-leveling test
A public test set for bid leveling. Take one public project's drawings and specifications and write about 30 mock subcontractor quotes across six trades. Plant known problems in them: exclusions, conditions, missing addendum acknowledgements, arithmetic slips like the Kokolakis formula error, and gaps that every bidder for a trade leaves out. Publish the quotes, a scope baseline and the answer key, then score tools against it.
The briefCredit · unclaimed02
Public unit prices, one table
An open table of unit prices, meaning the price per unit of work, such as per ton of asphalt or per square foot of sidewalk, drawn from bid tabulations, the tables in which public agencies publish every bidder's prices. Collect files from at least ten agencies, map each agency's item names onto one shared list, and publish the median and range per item per year, with links back to the source files.
The briefCredit · unclaimed03
Addenda change lists, checked
An addendum is a change to the bid documents issued before bids are due. Build a public set of 20 bid packages from public agencies, each with its original drawings and specifications and every addendum, plus a hand-made list of every change that affects price, recorded by sheet or section and by trade. Then test whether software can produce the same list.
The briefCredit · unclaimed
Open 01 · the brief
A planted-gap bid-leveling test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test set for bid leveling. Take one public project's drawings and specifications and write about 30 mock subcontractor quotes across six trades. Plant known problems in them: exclusions, conditions, missing addendum acknowledgements, arithmetic slips like the Kokolakis formula error, and gaps that every bidder for a trade leaves out. Publish the quotes, a scope baseline and the answer key, then score tools against it.
Why it is still open
The only measured results found are vendors' own. MeltPlan's figures come from "an internal benchmarking study," and the same article notes that comparing quotes with each other cannot catch work all bidders left out: "an item all three excluded produces no disagreement to flag." No independent test of bid-leveling tools was found.
Built with what exists
pdfplumber, an open-source library that pulls text and tables out of PDF quotes; general AI models from Anthropic (Claude), OpenAI and Google (Gemini); a small extraction model such as Meta Muse Ember (see the Note "The jobs agents repeat"); repeated runs to check that a tool gives the same answer each time (see the Note "All twenty times"); a spreadsheet for the answer key.
First step
Pick one public project's bid documents, write five mock drywall quotes with 15 planted problems, and score one general-purpose model on them.
Done looks like
At least 100 planted problems, at least 10 of them gaps shared by every bidder, and a published catch rate and false-alarm rate for at least three tools, each run 20 times on the full set.
Open 02 · the brief
Public unit prices, one table
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of unit prices, meaning the price per unit of work, such as per ton of asphalt or per square foot of sidewalk, drawn from bid tabulations, the tables in which public agencies publish every bidder's prices. Collect files from at least ten agencies, map each agency's item names onto one shared list, and publish the median and range per item per year, with links back to the source files.
Why it is still open
Each agency publishes in its own format. Oregon's transport department, for example, posts yearly spreadsheets with its own fields: bid date, contract number, region, project name, item description, quantity, bid price and bidder rank. The only multi-agency compilation found, the SlabCalc Concrete Cost Index on IEEE DataPort, covers a single item, concrete sidewalk, from 1,233 winning bids at 16 public agencies.
Built with what exists
Browser Use 1.0, an open-source library for AI browser agents, to collect files from agency websites (see the Note "Deleting 80% to start over"); pdfplumber for tables in PDF reports; a classifier such as Gemini Decision 1 to sort item descriptions into the shared list (see the Note "A thousand questions, one pass"); plain CSV files in a public repository.
First step
Download Oregon's latest bid-price spreadsheet and one other agency's file, and map 20 common items between them by hand.
Done looks like
At least 10 agencies, 3 years and 50 shared items, published under an open license, with a hand check of a random 10% of item mappings and the error rate stated.
Open 03 · the brief
Addenda change lists, checked
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An addendum is a change to the bid documents issued before bids are due. Build a public set of 20 bid packages from public agencies, each with its original drawings and specifications and every addendum, plus a hand-made list of every change that affects price, recorded by sheet or section and by trade. Then test whether software can produce the same list.
Why it is still open
Pelles, a company that sells AI to estimators, wrote in October 2025 that revision comparison is "one of the most time-consuming parts of estimating" and that "Manual comparison works but doesn't scale." Its own examples of AI comparison use pasted text, not drawings, and it cites no measured accuracy. No independent test of addendum reading was found.
Built with what exists
pdfplumber for specification text; a long-document reviewer built on Claude (see the Note "Seven more flagged"); a vision-language model such as Liquid LFM3, fine-tuned on one GPU, to spot changed areas on drawing sheets (see the Note "One GPU to fine-tune"); a spreadsheet for the answer key.
First step
Download one public bid package with its addenda from an agency's bid website, list every change by hand, and time how long it takes.
Done looks like
20 packages, at least 200 hand-listed changes checked by a second reader, and a published recall (the share of real changes found) and false-flag count per package for at least one software method and for a person working by hand, with time taken for each.