Industries / Finance and business services / Finance and banking / Debt collection
Account records
Account records are the data and documents that travel with an unpaid debt when a lender hands it to a collection agency or sells it: who owes it, how much, how the balance was built up, and who owns it now.
6
Milestones achieved, from 1989 to 2023.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1989
Bad loans sold at auction
Congress created the Resolution Trust Corporation in August 1989 to close failed savings and loan associations, called thrifts; it closed 747 of them and moved from selling assets one by one to selling them in packages by auction from mid-1990. The FTC says large-scale selling of consumer debt began in this crisis, when the corporation auctioned off nearly $500 billion in unpaid loans, and that the money those sales raised persuaded other lenders to start selling their own bad debts.
Credit
Resolution Trust Corporation
2013
A census of debt buyers
The US Federal Trade Commission studied nine large debt buyers, which bought 76.1% of the consumer debt sold in 2008 by its count: more than 5,000 portfolios holding nearly 90 million accounts with a face value of $143 billion, bought for about 4 cents on the dollar. The data files carried names, balances and key dates, but in a sample of 3.9 million accounts bought in 2009 buyers received documents such as account statements for only 12% of accounts, were not told whether a consumer had already disputed a debt, and verified only about half of the debts consumers disputed.
LessonA balance in a data file is not proof of a debt, and when the documents stay with the seller, disputes go unverified.
Credit
Federal Trade Commission
2013
Industry rules for debt sellers
The Receivables Management Association International (RMAI), the US trade group of debt buyers, began certifying businesses in 2013 with 19 standards; its version 14.0, in force from March 2026, has 67 standards and 532 active certifications, by RMAI's count. A certified buyer must hold a fixed list of data and document elements for each type of debt, for example 22 for credit cards, 39 for medical debt and 43 for business debt, and may not sell an account that is paid, settled, disputed, discharged in bankruptcy or created by identity theft, or whose seller cannot get the original documents; independent auditors check compliance midway through each three-year cycle.
Credit
Receivables Management Association International
2015
Selling debts already paid
The Consumer Financial Protection Bureau (CFPB), 47 states and the District of Columbia found that JPMorgan Chase had sold credit card debts that were already paid, settled, discharged in bankruptcy, fraudulent or owed by people who had died, some with wrong balances, and had backed more than 528,000 collection lawsuits with robo-signed sworn statements, meaning statements signed in bulk without checking the records. The July 2015 order made Chase stop collecting on those accounts, pay at least $50 million in refunds and $136 million in penalties and payments, give buyers account-level documents, keep records available to buyers for three years after a sale, and tell consumers when their account is sold.
LessonA seller answers for the records it passes on, not only the buyer who collects on them.
Credit
Consumer Financial Protection Bureau and state attorneys general
2021
China sells personal bad loans
A trial launched in March 2021 let Chinese banks sell personal non-performing loans, loans whose borrowers have stopped paying, in batches by online public bidding through the China Credit Assets Registry & Exchange; by the end of July 2021, 446 institutions had opened accounts, including 305 banks and branches and the national and local asset management companies that buy the loans. Bloomberg, as reported by investingLive, said such transfers reached 37 billion yuan in the first quarter of 2025, more than eight times a year earlier, that the regulator extended the rules past the end of 2025, and that the registry has since stopped publishing regular detailed figures.
Credit
China Credit Assets Registry & Exchange; National Financial Regulatory Administration
2023
EU loan sales get templates
An EU directive of 2021 on credit servicers and credit purchasers, applied from 30 December 2023, requires banks selling non-performing loans to give prospective buyers the information they need to value them, for consumer and business borrowers alike. Rules the European Commission adopted in September 2023, drafted by the European Banking Authority, set loan-by-loan templates of 135 data fields, 78 of them mandatory, covering the borrower, the loan and any collateral; a bank that gives inaccurate or incomplete information may face fines and damages claims, in the account of the law firm Clifford Chance.
Credit
European Banking Authority; European Commission
Watch for
Interest frozen after charge-off
RMAI's version 14.0 extends its ban on adding interest to balances after charge-off, the point at which a lender writes a debt off as a loss, from credit card debts to every type of debt held by certified debt buyers, from 1 August 2027; interest after a court judgment is still allowed.
Achieved whenIndependent certification audits after August 2027 find no interest added between charge-off and judgment in certified buyers' account records for any type of debt.
Claimed by
Receivables Management Association International, February 2026
Open
01
Checking debt sale files
An open, machine-readable schema for the data that must travel with each placed or sold consumer account, mapped field by field to RMAI's element lists, the conditions in the 2015 Chase order and the EU's 135-field template; a synthetic portfolio of invented accounts with known defects planted in it (paid, settled, discharged, deceased, disputed or identity-theft accounts marked for sale; a balance that does not add up from principal, interest, fees and payments; a missing original creditor or charge-off date); and a checker that flags each defect before the file changes hands.
The briefCredit · unclaimed02
Chain of title by machine
A benchmark of invented debt-sale paperwork, meaning bills of sale, account schedules and assignment statements, for chains of one to five sales of the same account, some with planted breaks: a schedule that leaves the account out, a seller whose name does not match the previous buyer, dates out of order, or balances that change without a payment. The task for an AI model is to trace each account from the original lender to its current owner and name any break.
The briefCredit · unclaimed
Open 01 · the brief
Checking debt sale files
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, machine-readable schema for the data that must travel with each placed or sold consumer account, mapped field by field to RMAI's element lists, the conditions in the 2015 Chase order and the EU's 135-field template; a synthetic portfolio of invented accounts with known defects planted in it (paid, settled, discharged, deceased, disputed or identity-theft accounts marked for sale; a balance that does not add up from principal, interest, fees and payments; a missing original creditor or charge-off date); and a checker that flags each defect before the file changes hands.
Why it is still open
The FTC found documents for 12% of accounts in its 2009 sample, and Chase sold paid and discharged debts with wrong balances. RMAI sets element counts per type of debt, and the EU sets a template, but this research found no open, machine-readable schema joining them and no public synthetic portfolio or test to check a sale file against; European DataWarehouse said in 2023 it planned a private-area service for disclosing NPL portfolio data.
Built with what exists
Python with pandas; Faker for invented names and addresses; Pandera or Great Expectations for data rules; a general AI model such as Claude Fable 5.2 (see the Note "Three working devices") to draft rules from the text of the Chase order and the RMAI standards, with each rule checked by a person.
First step
Put the Chase order's "do not sell" categories and RMAI's six sale restrictions side by side, and write one test for each.
Done looks like
The schema published under an open license with a mapping table to all three sources; 50,000 synthetic accounts with at least 10 defect types planted at known rates; a checker that finds at least 99% of planted defects with fewer than 1% of clean accounts flagged; another team reruns it from the published code.
Open 02 · the brief
Chain of title by machine
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A benchmark of invented debt-sale paperwork, meaning bills of sale, account schedules and assignment statements, for chains of one to five sales of the same account, some with planted breaks: a schedule that leaves the account out, a seller whose name does not match the previous buyer, dates out of order, or balances that change without a payment. The task for an AI model is to trace each account from the original lender to its current owner and name any break.
Why it is still open
Debt changes hands several times and documents often stay behind: the FTC found documents for only 12% of accounts at purchase, and the Chase order required hand-signed statements based on real records after Chase filed robo-signed ones. Searches for this research found vendor tools for real-estate title records but no public dataset or benchmark of debt-sale chains of title.
Built with what exists
Python with ReportLab to generate documents as PDFs, with scan noise added; Tesseract for text recognition; vision-language models such as Claude Fable 5.2 (see the Note "Three working devices") and Mistral Large 4, whose weights can be run on one's own machines (see the Note "A trillion open parameters"). All names and accounts invented.
First step
Write the generator for a two-sale chain, original lender to first buyer to second buyer, with one bill of sale and one account schedule per sale.
Done looks like
1,000 synthetic account chains, 30% with a planted break, released with answers and a scoring script; published scores for at least three models on chains traced correctly and breaks found, with at least one system finding 95% of breaks with no more than 2% false alarms.