Case Study

Complex Contracts AI


Complex contracts must be read, tracked, and reconciled with electronic transactions, taxing middle office labor specialists with highly manual work. These analysts must track each contract and reference several other systems to ensure the electronic records match the original and confirming contracts.

Our work with major, global asset managers presented several major challenges:

  • Reading both economic and legal data from PDFs. We needed to develop natural language processing system that would read complex contracts and turn them into electronic tables.
  • Learning how to read specific product contracts from specific parties. It was important to train algorithms to understand the format and language differences applied by different types of complex contracts and the parties who wrote the contracts.
  • Accurately matching the PDF data with electronic trading records. We needed to develop an AI algorithm that would very accurately match both records without human intervention.
  • Tracking and resolving the issues. The process requires automatically sending electronic links to broker/dealers, counter parties, legal teams, and investment professionals.

As illustrated in the diagram below, our AI platform leverages both visual and natural language processing algorithms to intelligently read and parse the relevant economic terms and legal language in contracts. Once extracted, our system automatically retrieves the matching trades and electronic records to perform the paper to electronic reconciliation seamlessly.

How It Works

Complex contracts are automatically read with our AI using a variety of algorithms. Once this is performed, the system automatically retrieves the matching electronic records. It begins to learn - with human feedback - to improve its matching capabilities.


There a number of major outcomes we have achieved in this complex contract case.

  • OTC review time reduced 82%. We measured the baseline mean time to manually read a paper confirm and reconcile it against trading records. Our algorithm reduced the labor time to run these complex checks by 82 percent.
  • Broker/dealer delay transparency. Our system now tracks and provides algorithmic benchmarks for delays in sending and approving confirm contracts across 18 broker/dealers.
  • Tracking and prediction of disputes, breaks, and root causes. The algorithm keeps track of all disputes and automatically correlates disputes by break type.

See a demo