BRENNEN

Industries / Construction and property / Real estate

Valuation

This is estimating what a property is worth, whether by trained appraisers (called valuers in many countries), by price indexes that track a whole market, or by computer programs called automated valuation models (AVMs).

7

Milestones achieved, from 1868 to 2024.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

In depth

The same ground is charted in more depth, as an industry of its own.

The road

AchievedWatch forOpen

Achieved

  1. 1868

    Surveyors form a profession

    Surveyors in London set up the Institution of Surveyors and elected its first council on 15 June 1868. It is now the Royal Institution of Chartered Surveyors (RICS). Its "Red Book" of valuation standards binds RICS members who value property, and the current edition took effect on 31 January 2025.

    rics.orgrics.org

    Credit

    John Clutton, the first president, and the founding members

  2. 1987

    Repeat-sales price indexes

    Karl Case and Robert Shiller built house price indexes for four US cities from homes that sold more than once. Comparing the same house at two dates keeps changes in price apart from differences in quality between houses.

    nber.org

    Credit

    Karl E. Case and Robert J. Shiller (NBER Working Paper 2393, September 1987)

  3. 1989

    Crisis brings appraisal standards

    During the US savings and loan crisis, nine appraisal organizations formed a committee in 1986 to write the Uniform Standards of Professional Appraisal Practice (USPAP), and in 1987 founded The Appraisal Foundation to maintain them. In 1989 Congress passed FIRREA, a law that made the Foundation the source of US appraisal standards and appraiser qualifications.

    LessonUniform US appraisal standards came in response to a crisis, not before it.

    appraisalfoundation.org

    Credit

    The nine appraisal organizations; The Appraisal Foundation; US Congress

  4. 2006

    Free online home estimates

    Zillow launched on 8 February 2006 with computer-made value estimates, called Zestimates, for about 40 million US homes, free to anyone. By 2011 it covered more than 70 million homes.

    sec.gov

    Credit

    Zillow

  5. 2021

    Zillow quits buying homes

    Zillow had been buying homes from sellers and reselling them, a model called iBuying. On 2 November 2021 it said it would wind the business down, citing errors in forecasting home prices far larger than it expected, and cut about 25% of its staff. It wrote down the value of its housing stock by about $304 million that quarter. News outlets reported the shutdown independently.

    LessonA price estimate good enough to show the public was not accurate enough to buy and resell homes on at scale.

    sec.govksl.com

    Credit

    Zillow Group

  6. 2021

    Appraisal gaps measured by race

    Freddie Mac, which buys mortgages from lenders, studied more than 12 million purchase appraisals from 2015 to 2020. Appraisals came in below the agreed price for 7.4% of homes in majority-white census tracts, against 12.5% in majority-Black tracts and 15.4% in majority-Latino tracts.

    wbez.org

    Credit

    Freddie Mac; reported by NPR

  7. 2024

    Rules for automated valuations

    On 17 July 2024 six US financial regulators issued quality-control standards for AVMs that lenders use on home loans. Lenders must ensure a high level of confidence in the estimates, protect against data manipulation, avoid conflicts of interest, test samples at random and comply with nondiscrimination law. The rule took effect on 1 October 2025.

    fdic.govgovinfo.gov

    Credit

    Consumer Financial Protection Bureau, Federal Deposit Insurance Corporation, Federal Housing Finance Agency, Federal Reserve Board, National Credit Union Administration and Office of the Comptroller of the Currency

Watch for

  1. Rules for AI in valuation

    RICS's first standard on using AI took effect on 9 March 2026; it requires AI risk registers, policies on responsible use and checks before buying AI tools. RICS is consulting from 3 November to 15 December 2026 on separate guidance for AI in property valuation, which it says will help valuers keep the professional judgment that "remains solely theirs".

    Achieved whenThe valuation guidance is published, and RICS's regulator or an independent review reports how firms that use AI for valuations comply with it.

    rics.orgrics.org

    Claimed by

    RICS, March and November 2026

Open

  1. 01

    A valuation test run ahead

    A public leaderboard that scores automated valuation models only on sales that had not been recorded when each model was frozen. Each month, every entrant's model is saved with a timestamp. When HM Land Registry publishes the next month's prices, the leaderboard runs each saved model on those sales and scores it. No model can have seen the answers.

    The brief
    Credit · unclaimed
  2. 02

    Rate condition from photos

    A public set of home photos rated for condition on the C1 to C6 scale that Fannie Mae and Freddie Mac set for US appraisals, where C1 is newly built and never lived in and C6 needs immediate repair to be usable. Two people rate each home separately. Then score AI models that rate condition from the same photos. Condition moves value, and today a person usually judges it on site.

    The brief
    Credit · unclaimed

Open 01 · the brief

A valuation test run ahead

Credit · unclaimed

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

What to build

A public leaderboard that scores automated valuation models only on sales that had not been recorded when each model was frozen. Each month, every entrant's model is saved with a timestamp. When HM Land Registry publishes the next month's prices, the leaderboard runs each saved model on those sales and scores it. No model can have seen the answers.

Why it is still open

Valuation firms test their own models and publish their own numbers. Gadsden, a UK AVM provider, reports that 58.8% of its estimates for standard sales fell within 10% of the price recorded by HM Land Registry, in its own 2026 test of 147,239 properties; it names no outside party that checked the test. This research found no independent, open test that scores several models on sales made after the models were fixed.

valuations.andrewgadsden.comgov.uk

Built with what exists

HM Land Registry's Price Paid Data, which covers sales for value in England and Wales, is updated monthly and is released under the Open Government Licence; Energy Performance Certificate data for floor areas; the Cook County Assessor's published valuation code as a model to adapt; Python; a scheduled job on GitHub Actions, whose commit history gives each saved model a public timestamp.

  1. First step

    Download a month of Price Paid Data, join it to Energy Performance Certificate floor areas for one city, and score a simple model on the following month's sales.

  2. Done looks like

    Twelve months of saved models from at least three entrants, scored on at least 5,000 sales, with each model's share of estimates within 10% of the price and its median error, by region.

Open 02 · the brief

Rate condition from photos

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 home photos rated for condition on the C1 to C6 scale that Fannie Mae and Freddie Mac set for US appraisals, where C1 is newly built and never lived in and C6 needs immediate repair to be usable. Two people rate each home separately. Then score AI models that rate condition from the same photos. Condition moves value, and today a person usually judges it on site.

Why it is still open

Restb.ai sells a C1 to C6 model and says it cuts valuation-model error by up to 18%, but its page cites no independent test, dataset or method. This research found no public photo set rated on the C1 to C6 scale.

restb.aisf.freddiemac.com

Built with what exists

Photos you have the right to use, such as your own and volunteers' homes or a brokerage's archive with its permission; the published rating definitions; Label Studio; AI models that read images (see the Note "Three working devices"); a small model taught the task on one graphics chip (see the Note "One GPU to fine-tune").

  1. First step

    Write a one-page rating guide from the published definitions, have two people rate 50 homes from photos, and measure how often they agree.

  2. Done looks like

    1,000 homes, each rated by two people with their agreement rate published, and a table of each model's exact-match and within-one-step rates against the agreed ratings.