BRENNEN

Industries / Finance and business services / Finance and banking

Lending and credit

Lending is the work of deciding who may borrow, on what terms, and collecting the money back, from small personal loans and mortgages to company loans and the bank promises that pay for goods shipped across borders.

7

Milestones achieved, from 1462 to 2026.

2

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. 1462

    Pawn loans against usury

    A public pawnshop called a Monte di Pietà opened in Perugia, Italy, lending small sums to poor people against a pledged object at low interest, as an answer to usury, meaning interest judged excessive. A University of Graz account of historian Tanja Skambraks's research says these pawnshops "only charged four to ten percent" where private lenders asked 20 to 30 percent, and that about 200 more opened over the next hundred years. The Graz account calls Perugia's the first pawnshop of its kind, and a 2021 study by Francesca Picciaia and colleagues in The Accounting Historians Journal calls it "the first ever Italian Monte di Pietà".

    uni-graz.atresearch.unipg.it

    Credit

    Franciscan friars, who played "a key role" in founding them, by the Graz account; the city of Perugia

  2. 1983

    Loans to groups of women

    Muhammad Yunus negotiated a Grameen Bank Ordinance in Bangladesh in 1983. The bank lent to poor rural women organized in groups of five that took collective responsibility for each other's loans, in place of collateral, the property a borrower pledges to a lender. By February 2008 it had 7.4 million clients and $545 million in loans outstanding, and its model had been copied in more than 40 countries, by the count of David Hulme of the University of Manchester. In the late 1990s many borrowers took on more debt than they could repay, and between 2001 and 2002 the bank moved all 1,200 branches to a new model that dropped joint liability. Yunus was awarded the Nobel Peace Prize in 2006.

    LessonSix randomized studies on four continents found that microcredit's average effects were "modestly positive, but not transformative" (Abhijit Banerjee, Dean Karlan and Jonathan Zinman, 2015).

    hummedia.manchester.ac.ukaeaweb.org

    Credit

    Muhammad Yunus and Grameen Bank

  3. 2007

    Subprime mortgages collapse

    In the early and mid-2000s US lenders made high-risk home loans, called subprime mortgages, to borrowers with below-average credit histories. Most were funded by securitization: loans were pooled into bonds and sold to investors. When house prices fell, losses spread, and the crisis ran from 2007 to 2010, by the Federal Reserve History account. The US Financial Crisis Inquiry Commission (FCIC) wrote in its 2011 report: "We conclude this financial crisis was avoidable."

    LessonThe FCIC concluded that "collapsing mortgage-lending standards and the mortgage securitization pipeline lit and spread the flame of contagion and crisis".

    federalreservehistory.orggovinfo.gov

    Credit

    US mortgage lenders and the firms that securitized their loans, as found by the FCIC

  4. 2012

    Loans by mobile phone

    In late 2012 Commercial Bank of Africa and Safaricom launched M-Shwari in Kenya, a bank account run over the M-PESA mobile money service. It lends small sums for 30 days with a 7.5% fee, and it scores applicants from their phone and M-PESA records, so people with no bank or credit history can borrow. Within two years it had more than 4.5 million active users. Economists Prashant Bharadwaj, William Jack and Tavneet Suri (NBER, 2019) found that 34% of eligible people borrowed, and that their households were 6.3 percentage points less likely to go without spending after a shock such as an illness.

    LessonThe authors conclude that digital loans improve access and resilience but "are not a panacea for greater credit market failures".

    nber.orgnber.org

    Credit

    Commercial Bank of Africa and Safaricom

  5. 2021

    Apple Card review

    New York's Department of Financial Services (NYDFS) investigated whether the Apple Card, issued by Goldman Sachs, treated women worse than men. Its report of 23 March 2021 found that applications "from women and men with similar credit characteristics generally had similar outcomes", and no fair lending violation. It also found that "a perceived lack of transparency undermined consumer trust", and the companies dropped a six-month wait before an applicant could appeal their credit terms.

    LessonA lawful credit decision can still lose a customer's trust when nobody can explain it.

    dfs.ny.gov

    Credit

    NYDFS; Goldman Sachs and Apple

  6. 2021

    Peer-to-peer lender exits

    Peer-to-peer (P2P) lending websites match people who want to lend with people who want to borrow, without a bank in between. Zopa, which the trade publication Alternative Credit Investor calls the world's oldest P2P lender, gained a full UK banking license, and in 2020 its own bank funded £456 million of the £633 million lent through its P2P platform. In December 2021 Zopa closed its P2P lending and its bank bought the loans from retail investors.

    LessonA model built to cut out banks ended with a bank funding most of its loans.

    alternativecreditinvestor.com

    Credit

    Zopa, whose chief executive Jaidev Janardana called it "a tough decision"

  7. 2026

    A second mortgage score

    A credit score is a number, computed from a person's borrowing record, that predicts how likely they are to repay. On 22 April 2026 the Federal Housing Finance Agency (FHFA) and the Department of Housing and Urban Development (HUD) said that Fannie Mae and Freddie Mac, the two government-sponsored companies that buy mortgages from US lenders, were "immediately accepting Vantage-scored loans from approved lenders", their first new credit score models for mortgages "in decades". In September 2026 the two opened VantageScore 4.0 to all approved lenders. An analysis by the investment bank Keefe, Bruyette & Woods (KBW) put VantageScore 4.0 at 5.6% of their monthly volume in August 2026, about 99% of it from Rocket Mortgage and United Wholesale Mortgage, and found that "most lenders are still pulling FICO scores for almost all loans".

    fhfa.govhousingwire.com

    Credit

    FHFA and HUD, under the Credit Score Competition Act of 2018; VantageScore

Watch for

  1. Newer FICO score for mortgages

    FHFA and HUD say Fannie Mae and Freddie Mac are moving forward with FICO Score 10T as well as VantageScore 4.0, and that the Federal Housing Administration will accept both for the loans it insures. In September 2026 HousingWire reported that "FICO Score 10T is not yet eligible for delivery to either GSE" and that manually underwritten loans must still use Classic FICO only.

    Achieved whenFannie Mae or Freddie Mac accepts loans scored with FICO Score 10T and outside data, such as an analyst's count of delivered loans, show lenders using it.

    fhfa.govhousingwire.com

    Claimed by

    FHFA and HUD, April 2026

  2. India's shared lending interface

    The Reserve Bank of India (RBI) built the Unified Lending Interface (ULI), which lets lenders draw on shared data services, such as land records, when deciding loans. It is "often compared to UPI but for lending", UPI being India's instant payment system. RBI Governor Sanjay Malhotra said in October 2025 that 58 lenders, including banks, non-bank lenders and co-operative banks, had sanctioned 3.2 million loans worth ₹1.75 trillion through it between its launch in August 2023 and 3 October 2025, and that it lets lenders build credit models for people without credit histories. In July 2025 the RBI was reviewing the rollout with lenders after adoption came in slower than expected, Angel One reported.

    Achieved whenThe RBI or an independent study publishes repayment outcomes for ULI loans and measures what they did for borrowers who had no credit history.

    rbi.org.ininformistmedia.comangelone.in

    Claimed by

    Reserve Bank of India, October 2025

Open

  1. 01

    An open table of covenants

    A free public table of the financial covenants in company loan agreements that US public companies file with the Securities and Exchange Commission (SEC). A financial covenant is a promise in a loan agreement, such as keeping debt below a set multiple of yearly earnings, tested every quarter; breaking it lets the lender demand changes. For each covenant, record the test, the threshold, how often it is tested, the definitions it relies on, and a link to the clause in the filed document.

    The brief
    Credit · unclaimed
  2. 02

    Test letter-of-credit checkers

    A public benchmark for checking trade documents against a letter of credit, a bank's promise to pay an exporter once the exporter presents documents, such as an invoice and a bill of lading, that match the credit's terms. Bank staff check each document against those terms and the International Chamber of Commerce (ICC) rules known as UCP 600, and any mismatch, called a discrepancy, can hold up payment. Build at least 300 invented letters of credit with matching document sets, every page marked SPECIMEN, plant discrepancies in some, have two experienced document checkers label each set, and publish scoring code. Labels cite UCP 600 article numbers rather than reproducing ICC text.

    The brief
    Credit · unclaimed

Open 01 · the brief

An open table of covenants

Credit · unclaimed

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

What to build

A free public table of the financial covenants in company loan agreements that US public companies file with the Securities and Exchange Commission (SEC). A financial covenant is a promise in a loan agreement, such as keeping debt below a set multiple of yearly earnings, tested every quarter; breaking it lets the lender demand changes. For each covenant, record the test, the threshold, how often it is tested, the definitions it relies on, and a link to the clause in the filed document.

Why it is still open

The Material Contracts Corpus (Peter Adelson and Julian Nyarko, Stanford, 2025) holds more than one million contracts filed with the SEC from 2000 to 2023, classified by type and party, and its description extracts no terms. A 2026 study by Maria Elena Bontempi of the University of Bologna searched 27 million pages of annual, quarterly and current reports to record 12 types of covenant for 4,538 firms, and its slides record the types, not the thresholds, and do not say the data are released.

arxiv.orgstata.com

Built with what exists

The Material Contracts Corpus bulk download; the SEC's EDGAR filing search; a model that reads long documents, such as Claude Fable 5.2 (see the Note "Three working devices") or the open-weight Mistral Large 4 (see the Note "A trillion open parameters"), to draft each entry; a second model that flags doubtful entries for a person to check (see the Note "Seven more flagged"); CSV files on GitHub.

  1. First step

    Download 20 credit agreements filed in 2023 from the corpus, extract their covenants by hand as a reference set, and compare one model's extraction against it.

  2. Done looks like

    At least 500 credit agreements from at least 300 companies, every financial covenant entered with its threshold and a clause link, under an open license, and a second person checking 50 randomly chosen covenants against the filed text and agreeing with at least 95% of them.

Open 02 · the brief

Test letter-of-credit checkers

Credit · unclaimed

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

What to build

A public benchmark for checking trade documents against a letter of credit, a bank's promise to pay an exporter once the exporter presents documents, such as an invoice and a bill of lading, that match the credit's terms. Bank staff check each document against those terms and the International Chamber of Commerce (ICC) rules known as UCP 600, and any mismatch, called a discrepancy, can hold up payment. Build at least 300 invented letters of credit with matching document sets, every page marked SPECIMEN, plant discrepancies in some, have two experienced document checkers label each set, and publish scoring code. Labels cite UCP 600 article numbers rather than reproducing ICC text.

Why it is still open

Software vendors claim high accuracy without a public test. Beam AI states "98% accuracy across flows" for its trade-finance agent and gives no test set or method, and a sponsored August 2026 article by the chief executive of Cleareye.ai says the ICC Banking Commission "has long flagged how many documentary credit presentations are rejected on first submission". This research found no public benchmark or independent accuracy study.

beam.aigtreview.com

Built with what exists

A spreadsheet or script that generates consistent document sets; a vision-language model that reads document images, such as Claude Fable 5.2 (see the Note "Three working devices"); open-weight models run locally for comparison (see the Note "A trillion open parameters"); a public scoring script.

  1. First step

    Write ten letters of credit for common goods, build their document sets in a spreadsheet with three planted discrepancies each, and ask one experienced checker to mark them blind.

  2. Done looks like

    At least 300 presentations with at least 1,000 labeled discrepancies and at least 100 clean sets, the two checkers' agreement reported, and a published table giving, for at least two AI checkers, the share of discrepancies found and the share of clean sets wrongly rejected.