Organizations steeped in text documents have an ally in their quest to streamline business processes. Natural language processing, a branch of AI focused on communication, is helping companies such as Accenture surface high-value information and cut costs by bringing text-based, unstructured communications into the machine learning age. \u00a0\n\nWith more than a million contracts in its records system and thousands more added monthly, Accenture\u2019s legal organization of about 2,800 professionals was struggling to find specific information across contracts, thanks to a tedious, costly process for which detailed cross-document search capability was limited.\n\n"If we have specific events that happen globally, for instance, we have floods in Chennai, or if we have certain events that are outside of our control, we, for years, had done that very much manually to understand what our client or our contract obligations were," says Mike Maresca, global managing director of digital business transformation, operations, and enterprise analytics at the global professional services provider.\n\nSo the company\u2019s Internal IT Enterprise Insight team set about leveraging NLP and AI to help improve the searchability of its contracts records system \u201cso that when we need to respond in a certain way, for instance, how much revenue is going to be impacted in that case of a flood in Chennai, we're able to look across our contracts to understand that," Maresca says.\n\nThe project, Accenture's Legal Intelligent Contract Exploration (ALICE), was launched with two needs in mind. First, the team needed to help the legal organization perform general text searches across the million-plus contracts in the company's Manage myRecords (MMR) system. Second, it needed to enable search for contract clauses.\n\n"We have a staff of data scientists that started to model analytics solutions that could index that large repository of contracts and build specific analytic algorithms to extract terms and conditions," Maresca says.\n\nThis was no easy task, as Accenture's historic contracts have few if any labeled clauses, meaning the team could not build and evaluate a classification model for the project. It was working with an unseen collection of documents and had to determine how to develop the clause identification component without labeled data.\n\nTo add to the challenge, many of Accenture's contracts originate on client paper, with different clients laying out their terms and conditions in different formats. That meant the team couldn't apply a template approach to identifying clauses \u2014 force majeure, liability, data privacy, and so on. Contracts don't necessarily reference clauses directly, so the solution required finding related keywords that could identify the presence of a clause.\n\nThe power of natural language processing\n\nTo solve this problem, the team turned to \u201cword embedding,\u201d an NLP method that facilitates comparisons between words based on semantic similarity. The model pulls out a list of keywords and their relevance scores from contracts, allowing the clause extraction solution to compute a similarity score that indicates how relevant each paragraph in a document is for a particular clause type, Maresca says.\n\nAccenture\u2019s word-embedding model goes through contract documents paragraph by paragraph, looking for keywords to determine whether the paragraph relates to a particular clause. For example, words like "flood," "earthquake," or "disaster" commonly occur with the clause "force majeure."\n\n"The analytics allow us to search on key terms, conditions, certain clauses, legal clauses, and started with English," Maresca says. "We've been layering in translation capabilities."\n\nExecutive leadership brought the project to IT with specific use cases around business continuity. The legal organization were the business owners of the solution.\n\nMaresca's team went from conceptual idea to pilot in a series of sprints that took six months. The team tested the pilot with various Accenture legal user groups and used the results to build a minimum viable project, which was then handed to a solution delivery team for full production at enterprise scale. It went live eight months later.\n\nLaunching legal analytics\n\nMaresca cites strong collaboration between the Internal IT Enterprise Insight Studio, the legal organization, and an IT development team as essential to taking the project from prototype to deployment into production. Data experts in the MMR team were key to understanding the metadata structure and how to link that metadata to the contract text. Legal experts helped to build the initial clause extraction component. The development team included data scientists, UI\/UX engineers, software engineers and functional architects.\n\nToday, ALICE, which has earned Accenture a CIO 100 Award in IT Excellence, is fully deployed and has improved Accenture's ability to identify and understand risks. Maresca says it has significantly reduced the amount of time attorneys have to spend manually reading through contracts for specific information.\n\n"Then it was extended to other value opportunities, like 'what if?' scenarios," Maresca says. "Have we ever structured our limits of liability in this way or that way? So, the use cases have grown as we continued to use this capability and to stretch it and enhance it as we see additional value opportunities. We're finding new ways to harvest value from the data that we have."\n\nThe company envisions ALICE as one piece of a broader digital transformation of its legal analytics capability. In that vision, ALICE will be the hub for various legal analytics services.