The Tested Effectiveness of Equivio>Relevance in Technology Assisted Review

Winston & Strawn has been working for several years with the Equivio>Relevance predictive coding technology (“Equivio”). As a continuation of that work, and at the request of a large institutional client (the “Client”) seeking to evaluate the potential cost savings and effectiveness of several different “Technology Assisted Review” (“TAR”) methodologies/ technologies, Winston & Strawn participated in a study (the “Project”) whereby we: (i) applied Equivio’s predictive-coding technology to a set of Client documents that previously had been reviewed and coded for the Client by outside contract attorneys (the “Data Set”); (ii) estimated the approximate time and costs required to conduct a defensible TAR of those documents using Equivio predictive coding for initial review and culling; and (iii) compared the results of our predictive-coding process to those of a typical, purely human review. The overarching goal of the Project was to complete a substantially accurate predictive-coding analysis within the shortest time possible, requiring the least amount of human (expert) review possible. Based on this stated goal and the size of the Data Set, Winston & Strawn at the outset predicted an estimated completion of its predictive-coding process within 14 days—which proved accurate.

The chosen Data Set consisted of approximately 300,000 documents from a previously resolved pre-litigation matter in which contract attorneys reviewed the documents in Relativity, coding them for relevance, privilege, importance, and a few additional categories (the “Human Review”). These same documents – along with the Human Review coding results – were provided to Winston & Strawn for application of predictive coding using its licensed Equivio>Relevance software. Equivio>Relevance uses a computer-based algorithm to identify potentially responsive and privileged documents and to cull out non-responsive documents.

The attached is a summary of the workflow, the results of its application to the chosen Data Set, and an estimation of the likely time and cost for conducting a full TAR of the Data Set, compared to a purely human review.