JPMorgan Chase

JPMorgan Chase’s lawyers and loan officers spent more than 360,000 hours each year reading and interpreting commercial loan agreements. Human error in interpreting about 12,000 new wholesale contracts per year also drove loan-servicing mistakes. The work was repetitive, time-intensive, and error-prone for the bank’s legal and credit teams.

The bank deployed COIN (Contract Intelligence), a machine-learning system that parses and interprets commercial loan agreements. The program went online in late 2016 and sits on JPMorgan’s private cloud infrastructure, learning by ingesting data to identify patterns and relationships. Machine-learning software supporting the effort was built with Cloudera.

Work that previously consumed over 360,000 hours a year is now completed in seconds, with fewer errors, by a system that runs continuously. Designers report COIN has helped cut loan-servicing mistakes tied to misinterpreting wholesale contracts. The bank is exploring further deployments for other complex legal documents.

• • High-volume, repetitive contract interpretation is a strong early automation target when error rates and hour counts are measurable.
• • In-house machine learning on private cloud infrastructure can automate core legal ops without a third-party legal AI product.
• • Measured time savings and error reduction create a clear case to expand the same model to related document types.