JPMorgan Chase & Co.

JPMorgan's lawyers and loan officers spent roughly 360,000 hours a year on mundane tasks like interpreting commercial-loan agreements, including reviewing about 12,000 new wholesale contracts annually. Human error in interpreting these contracts was a source of loan-servicing mistakes.

Starting in June, JPMorgan rolled out COIN (Contract Intelligence), a machine learning program running on the bank's private cloud network, to automatically read and interpret commercial-loan agreements. The system was part of a broader push to automate document-filing tasks and was funded out of the bank's technology budget.

COIN reduced the annual review work from 360,000 hours down to a matter of seconds and helped decrease the bank's number of loan-servicing mistakes tied to human misinterpretation of contracts. Bank executives framed the outcome as freeing staff for higher-value work rather than a pure displacement play.

• A narrowly scoped ML system can compress an enormous legacy manual-review workload (hundreds of thousands of hours/year) to near-zero.
• Built and run in-house on proprietary cloud infrastructure rather than a third-party legal-tech vendor.
• Internally framed as staff augmentation ("freeing people to work on higher-value things") even at 240,000+ employee scale.