Industries / Finance and business services / Finance and banking
Debt collection
Debt collection is the work of recovering money that people and businesses owe but have not paid, by letters, calls, payment plans and lawsuits, within laws that protect the people who owe it.
1869
Britain's Debtors Act ended arrest and imprisonment for most unpaid debts, with exceptions, so creditors had to recover money by persuasion and the civil courts rather than by locking debtors up.
47
Milestones achieved across 7 domains.
13
To watch for: announced, not yet confirmed outside.
14
Open deliverables. Nobody has shipped them.
In one minute
Where it started
For centuries a creditor could have an unpaid debtor jailed; Britain's Debtors Act of 1869 ended that for most debts. In Germany collection agencies grew out of credit reporting agencies in the 1870s, and early collection firms date from 1870 in Denmark and 1876 in Norway, by the Swedish collector Intrum's account. In 1977 the United States passed a federal law aimed at collectors, after Congress found "abundant evidence" of abusive practices.
Where it stands
The US industry has consolidated from about 4,500 collection firms in 2007 to about 2,500 to 3,000 in 2022, by the CFPB's count, while debt suits came to fill state courts; where courts keep data, more than 70% end in default judgments, given because the person sued did not answer, by Pew's count. Collection is still a leading source of complaints: the CFPB's public database holds 283,193 debt collection complaints received in 2025, about 41% of them about debts the person says they do not owe, though the Bureau says a growing share is filed in bulk by third parties and machines. Rules keep moving: the EU set common data templates for selling bad loans in 2023, Britain required lenders and collectors to consider forbearance case by case from 2024, and collectors are trying AI voice agents whose conduct no public test yet checks.
What is open
Most open jobs are public test sets built from invented accounts: debt sale files with planted defects, chains of ownership to trace, disputes paired with the records that answer them, default-judgment files, and simulated consumers that test whether AI collectors obey the rules on contact. Others turn rules now scattered across statutes into open data that software can check: contact limits, collector licenses, limitation periods, late payment rules, and credit report checks for the CFPB's own open tool. All of them can be built with AI models that read long legal and financial documents, with a person checking each rule against its source, and none contacts or profiles a real person.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
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 · 1 · 2
Last achievedEU loan sales get templates2023
Watch forInterest frozen after charge-off
2
openContacting people
Contacting people is the work of reaching someone who owes money, by letter, phone, text, email or visit, and the rules on when, how often and what a collector may say.
7 · 3 · 2
Last achievedAI voices count as artificial2024
Watch forIndia's stricter recovery rules
2
openDisputes and proof of debt
This is the work of answering a person who says a debt is wrong, not theirs or already paid, and of showing the records that prove what is owed.
6 · 1 · 2
Last achievedA standard dispute form2021
Watch forMachine-made complaints
2
openPayment plans
Payment plans is the work of agreeing how and when a debt is repaid: a one-off settlement, a monthly plan, a pause while someone gets advice, or a formal arrangement under insolvency law.
7 · 2 · 2
Last achievedWider help for arrears2024
Watch forUK limit on payment terms
2
openCollection lawsuits
This is the work of suing to collect a debt, and the court steps that turn an unpaid bill into a judgment that can be enforced against wages, bank accounts or property.
7 · 1 · 2
Last achievedNew York's pleading rules2021
Watch forAI checks of default files
2
openCredit reporting
For collectors, credit reporting is the work of sending each collection account to the credit bureaus in the agreed format, keeping it accurate, and correcting or removing it when a consumer disputes it or the law says it must go.
7 · 3 · 2
Last achievedCourt limits debt data storage2023
Watch forLenders share with every bureau
2
openCompliance
Compliance is the work of keeping collection inside the law: licenses to collect, rules on what collectors may say and charge, inspection by regulators, and care of the personal data a collector holds.
7 · 2 · 2
Last achievedGermany caps collection fees2021
Watch forFewer collectors under exam
2
open
Advances to build with
2 research notes

Claude Fable 5.2
Three working devices
Anthropic says its AI model Claude Fable 5.2 designed three prototype devices that worked without electrical modifications, with no outside test reported.

Mistral Large 4
A trillion open parameters
Mistral published its Large 4 AI model, a trillion numbers by its count, free to download and sell, though firms that rent out models need a separate license.
Open in debt collection
3 of 14 open deliverables
Open · Account records
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.
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.
Open · Contacting people
Open checker for contact rules
A free, open rulebook of contact limits in machine-readable form, and a checker that reads a log of calls, texts, emails and visits for invented accounts and flags every contact that breaks a rule. Start with four sets of rules: US Regulation F (seven calls in seven days per debt, the seven-day pause after a conversation, 8 am to 9 pm in the person's local time, opt-outs), Australia's guideline (three contacts a week, 10 a month, contact hours), India's 08:00 to 19:00 window, and the UK's Breathing Space, which bars contact to enforce a debt during a legal pause. Ship invented logs with known breaches planted in them.
Built with what exists
Python; the IANA time zone database for local times; rule text from the eCFR, the ACCC and ASIC guideline and the RBI directions; an AI model such as Claude Fable 5.2 to draft rule encodings from the legal text, each checked by a person (see the Note "Three working devices").
Open · Disputes and proof of debt
An open template count
A count, with code anyone can rerun, of how many US debt collection complaints were copied from templates. Take the debt collection narratives from the CFPB's archive (December 2011 to 14 August 2026), find exact and near-identical texts, and group them into template families by year and complaint issue. Then hand-label a random sample to separate a person who used a free template letter, which is lawful and common, from bulk filing by a third party. Work only with the public, anonymized texts and report totals, never single complaints.
Built with what exists
The CFPB archive's ZIP exports; Python with the open-source datasketch library for MinHash, a fast way to find near-identical texts; a general AI model for a first pass of the sample labels, such as Claude Fable 5.2 (see the Note "Three working devices") or Mistral Large 4 run on one's own machines (see the Note "A trillion open parameters").
Where it sits
Broader