JPMorgan's COIN: the famous 360,000-hours figure, and why it was never independently measured

The problem

Interpreting commercial-loan agreements is slow, repetitive legal work: reading each contract, finding and categorising the clauses that govern the loan. At JPMorgan’s scale this consumed large amounts of lawyer and loan-officer time and was prone to human error.

What was built

JPMorgan built COIN, for Contract Intelligence, an in-house system that learns to identify and categorise repeated clauses in commercial-loan agreements. It went online in June 2016 and was announced publicly in February 2017. Per an HBS write-up citing the bank, “the algorithm uses unsupervised learning” and “classifies clauses into one of about 150 different attributes of credit contracts” (source).

The outcome

The 360,000-hours figure. Bloomberg reported: “The program, called COIN, for Contract Intelligence, does the mind-numbing job of interpreting commercial-loan agreements that, until the project went online in June, consumed 360,000 hours of work each year by lawyers and loan officers. The software reviews documents in seconds, is less error-prone and never asks for vacation” (source). The ABA Journal relayed the same figure, that the software does “work that once required 360,000 hours of work each year by lawyers and loan officers” (source).

Contract volume and errors. Per Bloomberg, COIN “has helped JPMorgan cut down on loan-servicing mistakes, most of which stemmed from human error in interpreting 12,000 new wholesale contracts per year, according to its designers” (source). The magnitude of the error reduction was never quantified publicly.

Accuracy claim. The HBS write-up states the bank has said “the algorithm is more accurate than human lawyers” (source), while noting JPMorgan “has been tight-lipped about the details of the proprietary technology”. This is a company-attributed claim with no published measurement basis.

How this was verified

This case carries a green badge under TIN’s current standard: independently validated by TIN against the public record, with every figure re-researched, quoted exactly, and checked against a primary or independent capture. Read the badge as TIN validated this account is accurate, including the honest part the account foregrounds: the famous 360,000-hours and 12,000-contract figures are a single-origin 2017 corporate claim, relayed by reputable press but never independently measured, and JPMorgan has not updated them. Green certifies the claim and its single source are reported exactly and in context; it does not certify the number was independently measured. Direct client confirmation is the higher bar TIN grows into as the platform matures.

Sources

Cited in this case file. Tier 2 = independent press; Tier 3 = secondary analysis. Each quote above was checked against an archived capture of the source.

1. Bloomberg / Hugh Son, “JPMorgan marshals an army of developers to automate high finance,” 2017-02-28 (Tier 2, single origin of the figures). https://www.bloomberg.com/news/articles/2017-02-28/jpmorgan-marshals-an-army-of-developers-to-automate-high-finance
2. ABA Journal, “JPMorgan Chase uses tech to save 360,000 hours of annual work by lawyers and loan officers,” 2017-03-02 (Tier 2, relays Bloomberg). https://www.abajournal.com/news/article/jpmorgan_chase_uses_tech_to_save_360000_hours_of_annual_work_by_lawyers_and
3. Harvard Business School (RCTOM), “JP Morgan COIN: a bank’s side project spells disruption for the legal industry,” 2018-11-14 (Tier 3). https://aiinstitute.hbs.edu/platform-rctom/submission/jp-morgan-coin-a-banks-side-project-spells-disruption-for-the-legal-industry/