JPMorgan Chase

Interpreting commercial-loan agreements kept JPMorgan legal teams busy for hundreds of thousands of hours each year. Human error in reading roughly 12,000 new wholesale contracts annually also drove loan-servicing mistakes.

JPMorgan deployed COIN (Contract Intelligence), a machine-learning program that interprets commercial-loan agreements and reviews documents in seconds. The system runs on the bank’s Gaia private cloud and was built with Cloudera; it learns by ingesting data to identify patterns and relationships.

COIN took over work that had consumed 360,000 hours of lawyers’ time annually before it went online in June, reviewing documents in seconds. Designers say it is less error-prone than manual review and has helped cut loan-servicing mistakes tied to contract interpretation.

• • High-volume, repetitive contract interpretation is a strong fit for machine learning when paired with sufficient compute (here, a private cloud).
• • Measured outcomes can include both time displaced (hundreds of thousands of lawyer hours) and fewer servicing errors from misread terms.
• • Banks often treat legal contract AI as a beachhead for broader automation across other complex legal filings.