BRENNEN

Industries / Finance and business services / Finance and banking

Mortgage lending

Mortgage lending is the work of lending money to buy homes and other property, with the property pledged as security for the loan, collecting the payments over the life of the loan, and selling many of the loans on to investors.

1775

Richard Ketley, landlord of the Golden Cross inn on Snow Hill in Birmingham, England, founded a society whose members each paid the same amount into a common fund until every one of them had a house. A plaque unveiled in Birmingham in 1958 says he founded it "in or before 1775" and calls it "the first known building society in the world"; the Building Societies Association dates it to 1775.

bsa.org.ukopenplaques.orgbsa.org.uk

42

Milestones achieved across 6 domains.

16

To watch for: announced, not yet confirmed outside.

14

Open deliverables. Nobody has shipped them.

In one minute

Where it started

In December 1770 an association of Silesian landowners in Prussia issued the first Pfandbrief, a bond secured on its members' estates, and in or before 1775 Richard Ketley's society in Birmingham began pooling its members' savings until each of them had a house. Until the 1930s an American buyer might have to put down as much as 40% and repay within a dozen years. From 1934 the US Federal Housing Administration insured loans of 20 to 30 years, paid off in equal monthly payments with as little as 10% down, and in 1938 the government chartered Fannie Mae to buy those loans from lenders.

Where it stands

US households owed $13.4 trillion on mortgages at the end of September 2026, by the Federal Reserve Bank of New York's count, and the Urban Institute puts 64% of the debt on US homes in bonds guaranteed by Fannie Mae, Freddie Mac or Ginnie Mae. Home loans outstanding in the European Union came to about EUR 7.1 trillion at the end of 2025, by the European Mortgage Federation's count, and China's personal housing loans fell 4.1% in the year to September 2026, to 35.91 trillion yuan, by the People's Bank of China's figures. The work is still slow and costly: the average US purchase loan took 36.8 days to close in March 2026, by ICE Mortgage Technology's count, and independent mortgage banks spent $10,936 on each loan they made in the second quarter of 2026, by the Mortgage Bankers Association's count.

What is open

Most open jobs are public tests that nobody has built: made-up pay stubs, appraisal reports, eNotes, closing packages and escrow statements with planted errors, and scripted calls to score the AI phone agents that lenders and servicers now use to take applications and to talk to borrowers who have fallen behind. Others are public tables and models: a tracker of the short-term credit lines that lenders without deposits depend on, a table of how each country pays for its home loans, and an open benchmark for predicting when borrowers pay off early. AI models that read long documents and hold spoken conversations put this work within reach of a student or a small team using invented borrowers and public records; MortarBench, an academic benchmark posted in June 2026 in which the best closed model got 77.1% of loan questions exactly right, shows where such tests start.

Domains

AchievedWatch forOpen

Advances to build with

1 research note

Open in mortgage lending

3 of 14 open deliverables

Open · Applications

Read the income papers

A public test of AI reading a borrower's income papers as scans and photos. Make 500 invented borrower files, every page marked SPECIMEN, each with two pay stubs, two years of W-2s (the yearly US wage statement), a federal tax return with Schedule C business profit for the self-employed, two months of bank statements and a filled-in URLA, rendered both as clean PDFs and as photographed or scanned pages. The answer key gives every field, such as employer, gross pay, year-to-date pay, business profit and large deposits, and about ten checks across documents per file, such as whether the employer's name matches on the stub, the W-2 and the URLA.

Built with what exists

Python with Faker for invented people and employers and ReportLab to draw pay stubs and bank statements; the IRS's fillable W-2 and 1040 PDFs, filled in with pypdf; Augraphy to make pages look printed, scanned or photographed; the ULAD schema from Freddie Mac's site. Systems to test: Claude, Gemini, and a small extraction model such as Meta's Muse Ember (see the Note "The jobs agents repeat"), with Gemini Decision 1 answering each file's yes-or-no cross-document checks in one pass (see the Note "A thousand questions, one pass").

Credit · unclaimed

Open · Underwriting

A debt-to-income answer key

A public set of made-up loan files with a worked answer key. Each file holds pay stubs, W-2s, tax returns with Schedule C (self-employment) and Schedule E (rental income) and a credit report listing several kinds of debt; the key works out qualifying monthly income, monthly debts and the debt-to-income ratio the way Fannie Mae's Selling Guide says. AI models are scored against it. Nothing is used on a real applicant. The files can come from the same generator as "Read the income papers" under Applications; this test scores the calculation, not the reading.

Built with what exists

The public Fannie Mae Selling Guide; the IRS's public PDF forms, filled by a Python script with pypdf or ReportLab; MortarBench's open code as a model for the scoring harness; Claude, GPT and Gemini models and the open-weight Qwen3 as systems to test, with a small extraction model such as Meta's Muse Ember for reading the forms (see the Note "The jobs agents repeat").

Credit · unclaimed

Open · Valuing the home

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.

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.

Credit · unclaimed

Where it sits