BRENNEN

Industries / Construction and property / Commercial and residential construction

Estimating

Estimators measure the drawings to count every item of material and hour of labor, price them, and turn the total into a bid or a budget.

7

Milestones achieved, from 1868 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1868

    Surveyors form a profession

    Quantity surveyors measure and price a building from its drawings. Forty-nine surveyors met at the Westminster Palace Hotel in London on 15 June 1868, elected John Clutton president and founded the Institution of Surveyors, which became the Royal Institution of Chartered Surveyors (RICS). RICS says it now sets standards for more than 130,000 members and candidates in more than 140 countries.

    rics.org

    Credit

    John Clutton and the founding members

  2. 1942

    A price book for builders

    Robert Snow Means, a civil engineer, kept construction costs in leather-bound books he made at his kitchen table, and by the early 1940s other builders were asking to buy them. His company dates its first cost data to 1942. Its database, now owned by Gordian, holds more than 92,000 priced line items, filed under the CSI MasterFormat numbers.

    rsmeans.comfims.doe.gov

    Credit

    Robert Snow Means

  3. 1978

    One numbering for all work

    The Construction Specifications Institute (CSI) and Construction Specifications Canada released MasterFormat, a numbered list of the kinds of work in a building. Specifications, cost books and estimates filed under the same numbers can be matched line by line. CSI last updated it in 2020.

    csiresources.orgfims.doe.gov

    Credit

    Construction Specifications Institute; Construction Specifications Canada

  4. 2002

    Estimates get an uplift

    HM Treasury had the engineering firm Mott MacDonald review 50 large UK public projects from the previous 20 years. On conventionally procured projects, capital costs came in on average 47% above the estimate made at approval, and works took 17% longer. Treasury guidance since then tells appraisers to add an "optimism bias" uplift to early estimates, of up to 24% for standard buildings and 51% for unusual ones.

    LessonAn early estimate should be corrected by how similar past projects actually turned out.

    parliament.vic.gov.augov.uk

    Credit

    HM Treasury; Mott MacDonald

  5. 2004

    Holyrood's cost doubles

    The forecast cost of the Scottish Parliament building at Holyrood rose from £195 million in September 2000 to £431 million in February 2004, by the Auditor General for Scotland's count. He found that the design was still being worked out while building went on, that the 2000 cost plan was "an indicator of the costs rather than a reliable estimate", and that under the contract form used, the client carried the risk.

    LessonWhen building starts before the design is finished, the estimate is only an indicator, and the client who carries the risk must manage the design.

    audit.scot

    Credit

    report by Robert Black, Auditor General for Scotland, June 2004

  6. 2017

    Costs reported one way

    A coalition of 45 professional bodies, among them RICS, the Indian Institute of Quantity Surveyors and the International Cost Engineering Council, launched the International Construction Measurement Standards (ICMS) in July 2017. ICMS sets one way to group and report a project's costs, so that projects in different countries can be compared.

    theconstructionindex.co.uk

    Credit

    ICMS Coalition

  7. 2026

    AI counts reliably

    Takeoff is the step of measuring quantities off the drawings. Three researchers compared the AI takeoff tool Togal AI with a contractor's takeoff of 29 line items on a live commercial project, in what they call the first quantitative validation of the tool in professional practice. Its counts of items such as doors and windows nearly matched; it measured areas consistently short, most of all on irregular exterior walls.

    LessonAI takeoff can be trusted first on counts; areas still need an estimator's check.

    easychair.org

    Credit

    Togal.AI; tested by Hooman Sadeh, Dominick Geloso and Dimitar Todorov (Associated Schools of Construction conference, June 2026)

Watch for

  1. A takeoff model beats estimators

    Handoff says its h1 model scored 81.6 on TakeoffBench-V1, its own test of 10 residential plan sets with 2,009 checked line items, against 77.6 for professional estimators using takeoff software and 61.4 for the best general AI model it tried. Handoff paid the estimators, an AI model matched the line items, and the plan sets are released only on request.

    Achieved whenAn outside group runs the test, or one of its own, on plan sets it chose, including commercial buildings, and gets a similar result.

    handoff.ai

    Claimed by

    Handoff, July 2026

Open

  1. 01

    A public commercial takeoff test

    A public set of commercial building drawings, each with a full takeoff made by two estimators and reconciled, and a script that scores any tool's takeoff against it item by item, with counts, lengths and areas scored separately.

    The brief
    Credit · unclaimed
  2. 02

    Why change orders happen

    An open, machine-readable dataset of change orders on public school building contracts: the original contract amount, each change, its cost, its description and its reason, and the revised total. A change order is a priced change to a signed contract. Use an AI model to read the published reports, and check its work by hand.

    The brief
    Credit · unclaimed

Open 01 · the brief

A public commercial takeoff test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of commercial building drawings, each with a full takeoff made by two estimators and reconciled, and a script that scores any tool's takeoff against it item by item, with counts, lengths and areas scored separately.

Why it is still open

The one outside test of an AI takeoff tool covered one project and 29 line items. Handoff's TakeoffBench-V1 covers houses only, and its drawings and answers are released only on request. CEQuest, an academic benchmark, tests question answering about drawings, not full takeoffs.

easychair.orghandoff.aiarxiv.org

Built with what exists

Drawing sets of finished buildings whose owners agree to publication, such as school districts or city agencies; a PDF measuring tool for the human takeoffs; Handoff's evaluation harness, which is public under the Apache-2.0 license, as a starting point; commercial AI takeoff tools and general AI models to be scored.

  1. First step

    Ask three public school districts or city building departments for permission to publish the drawings of a finished building, and count the doors, windows and floor finishes on the first set by hand.

  2. Done looks like

    At least 10 commercial drawing sets and 1,500 line items published under an open license, with a table of scores for at least three AI tools and two human estimators.

Open 02 · the brief

Why change orders happen

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open, machine-readable dataset of change orders on public school building contracts: the original contract amount, each change, its cost, its description and its reason, and the revised total. A change order is a priced change to a signed contract. Use an AI model to read the published reports, and check its work by hand.

Why it is still open

Chicago Public Schools publishes a change order log each month with each contract's original amount, number of changes, revised total, a description of each change and a reason code such as "Added Scope of Work" or "E&O AOR/EOR" (errors and omissions by the architect or engineer), but only inside PDF board reports. This research found no dataset that collects such logs across years.

cpsboe.org

Built with what exists

A PDF text extractor such as pdftotext; a general AI model to read each change and check its reason code; Python and SQLite.

  1. First step

    Download the last 12 monthly Chicago Public Schools board reports and turn one month's log into a table by hand, to serve as the answer key.

  2. Done looks like

    At least 1,000 contracts across at least three years, every field checked against the PDFs on a random sample of 100 rows with the error rate published, and a table of how much each reason code added to the original contract totals.