Thomson Reuters announced today (23 May) that its products will be integrated with Microsoft 365 Copilot, becoming one of the first in the legal industry to do so. TR announced that working with Microsoft, it has developed a contract drafting solution, powered by its legal software, content, and Copilot. The intelligent drafting solution will enable users to draft working documents and access TR’s content within Microsoft Word, building the final document with integrated access to Thomson Reuters knowledge, content, and AI technology.
Microsoft will announce its partnership with Thomson Reuters and demonstrate the integration with TR legal drafting in day two of its Microsoft Build conference in Seattle.
The other legal technology vendor to have announced an integration with Copilot is Peppermint Technology, which we revealed on 18 May.
In addition to integrating Microsoft 365 Copilot, Thomson Reuters today shared plans to incorporate new generative AI capabilities across its product suite including Westlaw Precision; Practical Law; Checkpoint Edge; legal document review and summary; and legal drafting.
In a briefing, chief product officer David Wong said: “This is live and in progress and as prototypes become available we will invite customers to test them and learn as part of our beta.” The first beta will be Westlaw Precision, Practical Law and Drafting in the second half of 2023, with TR’s research offering being the first priority for the roll out of the new generative AI functionality. Legal drafting, integrated with Copilot, will be available in line with the wider Copilot release schedule. TR says that tax and compliance, risk and fraud generative AI solutions will follow in 2024.
Wong said that TR has been running a battery of tests since GPT-3 launched in 2020 – including running it side by side with conventional research exercises – when the accuracy was “very bad.” Wong said that with the launch of GPT-4, the accuracy improved but observed, “GPT-4 by itself is still not a good student.” The accuracy of those tests stood at around 70% and Wong said: “We knew that it was still not sufficient to answer legal research questions.” It was when they began to teach the system to use TR library that accuracy rose to over 90%, with Wong observing: “In line with our general approach to making the technology available we are going to use a gold standard of benchmarks that we expect the system to meet, and when it reaches those marks we will release it.”
This story is being updated.