Industries / Finance and business services / Finance and banking / Mortgage lending
Valuing the home
This is the lender's check on what a home is worth before lending against it, by a licensed appraiser's report, a software estimate from an automated valuation model (AVM), a mix of the two, or for some loans no appraisal at all.
7
Milestones achieved, from 1900 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1900
Germany defines a lending value
Germany's Mortgage Bank Act of 1900 introduced the mortgage lending value (Beleihungswert), the price a property can be expected to fetch at any time during the loan, leaving out short-term swings and speculation. The Pfandbrief Act, in force since 19 July 2005, keeps the idea: a valuer independent of the credit decision sets the value, it may not exceed the market value, and only the part of a loan up to 60% of it can back mortgage Pfandbriefe, the bonds German banks sell to fund home loans.
Credit
the German legislature; the Association of German Pfandbrief Banks (vdp), whose explanation is used here
1996
Valuers not liable for slumps
After UK property prices slumped in the early 1990s, many lenders sued valuers for overvaluing homes they had lent against. On 20 June 1996 the House of Lords decided three such appeals together, known as SAAMCO, and Lord Hoffmann held that a valuer who gives wrong information is "responsible only for the consequences of the information being wrong", which caps a negligent valuer's damages at the amount of the overvaluation and leaves the lender with the loss from the market's fall. The High Court of Australia appeared to reject the principle in 1999, by John Murdoch's account.
LessonA valuation states what the property is worth on the day; it does not insure the loan against a falling market.
Credit
Lord Hoffmann and the House of Lords; analysis by John Murdoch, University of Reading (2001)
2009
Loan staff kept from appraisers
Inflated appraisals were widespread in the US subprime boom. Under an agreement between New York's attorney general, Andrew Cuomo, Fannie Mae, Freddie Mac and their regulator, the Federal Housing Finance Agency (FHFA), the two companies adopted the Home Valuation Code of Conduct, which took effect on 1 May 2009 and barred lenders' loan sales staff from "substantive communications" with appraisers; the Dodd-Frank Act of 21 July 2010 repealed the code but wrote several of its rules into law. Economists at the Federal Reserve Bank of Philadelphia found that appraisals below the agreed price rose from 9.3% to 15.0% of appraisals between the six months before the code and the six months after, and that purchase loans the code covered were more often denied.
LessonSeparating appraisers from the people who sell loans cut inflated values, but in a falling market it also brought more low appraisals and more denied loans.
Credit
New York State Attorney General; Fannie Mae; Freddie Mac; FHFA; US Congress; measured by Lei Ding and Leonard Nakamura (2014)
2019
Appraisers anchor on the price
Michael Eriksen, Hamilton Fout, Mark Palim and Eric Rosenblatt studied 6.5 million appraisals for purchase loans made from 2013 to 2017 and found that appraisers weighted the comparable sales they chose in ways that put an extra 23% of values at or above the contract price, the price buyer and seller had agreed. A 2020 paper by the same authors found that appraisers who knew the contract price were more than twice as likely to reach it as those who did not. In September 2025 the US Government Accountability Office (GAO) cited the work as a reason automated models, which do not see the contract price, avoid this bias.
LessonAn appraiser who knows the agreed price tends to confirm it, so the check is weaker than it looks.
Credit
Eriksen, Fout, Palim and Rosenblatt, whom GAO describes as "academic and Fannie Mae researchers" (Journal of Urban Economics, 2019); GAO
2021
Europe sets valuation rules
The European Banking Authority's guidelines on loan origination and monitoring, published in May 2020 and applied from 30 June 2021, require a valuer to assess a property with a full visit, inside and out, when a loan is made. In "well-developed and mature property markets" a valuer may instead do a desktop valuation supported by "advanced statistical models", the EBA's term for AVMs, but stays responsible for the value and must be independent of the credit decision. The EBA found that only three member states then regulated such models directly.
Credit
European Banking Authority
2022
Appraisals without a visit
In a desktop appraisal the appraiser values the home without visiting it, from tax records, listings and other data. FHFA let Fannie Mae and Freddie Mac accept them in March 2020, early in the Covid-19 emergency, ended them in May 2021, and from March 2022 both companies' guides allow them for most one-unit homes with a loan of 90% of the home's value or less. In October 2023 FHFA's inspector general reported that desktop appraisals were less than one percent of the two companies' loans.
LessonAllowing a remote appraisal did not by itself get lenders to use it.
Credit
FHFA; Fannie Mae; Freddie Mac; FHFA Office of Inspector General
2026
One form for every home
On 2 November 2026 Fannie Mae and Freddie Mac made one redesigned appraisal report mandatory for new appraisals sent to them, replacing "all existing GSE appraisal forms with a single flexible and dynamic report" for every property type. It is built on version 3.6 of the Uniform Appraisal Dataset (UAD), the shared list of data fields and definitions, and lenders with a temporary exception may send old reports until early 2027, by Cotality's summary.
Credit
Fannie Mae; Freddie Mac
Watch for
Order to widen valuation options
Executive Order 14393, "Promoting Access to Mortgage Credit", signed on 13 March 2026, asks six agencies, among them the Federal Reserve, the CFPB and FHFA, to consider "modernizing appraisal regulations and guidance to expand the use of alternative valuation models, desktop and hybrid appraisals, and artificial intelligence valuation tools", and to reduce appraisal requirements for low-risk loans. It sets no deadline for these steps.
Achieved whenOne of the named agencies issues a final rule or formal guidance that changes when an appraisal is required or permits AI valuation tools, and lenders begin using it.
Claimed by
President Donald J. Trump, March 2026
AI fills in appraisal forms
apprAIz and Restb.ai say their joint workflow "helps populate up to 90% of the UAD 3.6 appraisal form", using computer vision to pull property details from photos, while the appraiser keeps "independent analysis, judgment and USPAP-compliant conclusions" (USPAP is the US standard for appraisal practice). The release gives no accuracy figures and cites no outside test.
Achieved whenA regulator, Fannie Mae, Freddie Mac or an academic team publishes how often the fields the software fills match what a licensed appraiser records for the same homes.
Claimed by
apprAIz and Restb.ai, August 2026
Open
01
Test reports with planted errors
500 made-up appraisal reports in the UAD 3.6 format, each with photos, a floor plan and the written data fields, about half carrying planted errors: a bedroom count that does not match the floor plan, a condition rating the photos contradict, a required photo missing, a kitchen described as renovated but shown as dated. Publish the answer key and score AI review tools and general vision models on the errors they find and the false alarms they raise on clean reports. No real borrower or loan is involved.
The briefCredit · unclaimed02
Floor area by the rulebook
A public test of how consistently software, AI models and people apply one measuring standard, ANSI Z765-2021, which Fannie Mae requires for appraisals, to the same floor plans. Draw 300 made-up plans with known dimensions, including the cases the standard treats specially: stairs, sloped ceilings, levels partly below ground and openings to the floor below. Work out the correct above-ground and below-ground areas as the answer key, and score each tool's figures against it.
The briefCredit · unclaimed
Open 01 · the brief
Test reports with planted errors
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
500 made-up appraisal reports in the UAD 3.6 format, each with photos, a floor plan and the written data fields, about half carrying planted errors: a bedroom count that does not match the floor plan, a condition rating the photos contradict, a required photo missing, a kitchen described as renovated but shown as dated. Publish the answer key and score AI review tools and general vision models on the errors they find and the false alarms they raise on clean reports. No real borrower or loan is involved.
Why it is still open
Vendors sell software that reads appraisal photos and checks reports, but they publish claims, not tests; the apprAIz and Restb.ai release gives no accuracy figure. In September 2025 GAO found that public information on the inputs and accuracy of automated valuation models is "generally limited" and that vendors "may selectively report statistics on model accuracy". Searches for a public benchmark of appraisal-review tools found only vendor material.
Built with what exists
The UAD 3.6 sample scenarios and XML files, compliance rules and photo requirements that Fannie Mae and Freddie Mac publish; home photos that can be reused under their licence; the Zillow Indoor Dataset, 67,448 panoramas from 1,575 homes, for academic use; vision models such as Claude, Gemini, GPT or the open-weight Qwen2.5-VL; Gemini Decision 1 to put hundreds of yes-or-no consistency questions to each report in one pass (see the Note "A thousand questions, one pass"); Label Studio for a second person to check the key.
First step
Download the published sample scenarios, take one, write 20 variants with one planted error each, and run two vision models on them.
Done looks like
500 reports with at least 400 planted errors across at least 10 error types, an answer key confirmed by two people with their agreement rate, and a published scored run of at least three tools giving the detection rate and false-alarm rate for each error type.
Open 02 · the brief
Floor area by the rulebook
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of how consistently software, AI models and people apply one measuring standard, ANSI Z765-2021, which Fannie Mae requires for appraisals, to the same floor plans. Draw 300 made-up plans with known dimensions, including the cases the standard treats specially: stairs, sloped ceilings, levels partly below ground and openings to the floor below. Work out the correct above-ground and below-ground areas as the answer key, and score each tool's figures against it.
Why it is still open
Fannie Mae's Selling Guide says appraisers "must follow" the standard when measuring and counts a level as below ground "if any portion of it is below-grade". The trade magazine Working RE says "there remains some significant room for interpretation when it comes to the stairs" and describes one house an appraiser measured at 3,000 square feet against the architect's 2,800. A search for studies measuring how far figures differ under the standard found only practitioners' commentary.
Built with what exists
A purchased copy of ANSI Z765-2021; Python with the Shapely geometry library to draw plans and compute areas; CubiCasa5K, 5,000 annotated floor plans, for realistic layouts; floor-plan apps and vision models as the tools under test.
First step
List the standard's rules for stairs, ceiling height and below-ground levels, draw 20 plans that each test one rule, and compute their areas by hand and by script.
Done looks like
300 plans with an answer key checked by two people, and for each tool the median error in square feet and in percent, the share of plans within 1% of the key, and results for each hard case.