Using machine learning to streamline litigation and cut costs

The challenge

One of the things that makes litigation time consuming and expensive is the process of making relevant information available to the other side and reading through what they make available – a process called disclosure.

If we could use automation to assess material and focus human attention only on those parts that matter we could reduce the time and cost of litigation and, most importantly, focus our efforts and skills on the information most likely to lead to a successful outcome.

The solution

We used a machine learning model to assess all the documents we were given in a case to rank them according to their relevance to the dispute, having used manual document review to train the system. Each document is ranked from 0 to 100. We then began our review not at a random document, but at the one ranked 100, the most relevant and important document. We worked our way through in order of relevance. The system learns from its users, who mark the relevance of each document as they go.

We were then able to make a well-informed decision about when to stop reviewing documents, confident that we were not leaving important ones un-reviewed.

The result

We were able to review only 93,000 of 184,000 documents involved in the dispute, with full confidence that we had seen the most important, relevant material. By not reviewing 91,000 documents we saved the client between 40% and 80% of the review cost they would have faced without this system.

The system also made other elements of the case more efficient and more effective, since our lawyers could ask the system in plain language for particular sets of documents for review.