Top Trends in eDiscovery #6: Predictive Coding
Predictive coding, which uses sampling to build and perform data set analysis, is quite a good tool for accelerating some manual analysis and review tasks. It is also not new: several vendors use predictive coding to accomplish eDiscovery tasks like analysis clustering, prioritization, threading or categorization. These activities shorten the review cycle by cutting down on unresponsive documents and organizing results into logical screens for reviewers.
In contrast, predictive coding for automated review seeks to replace manual review efforts for time and cost savings and for improved accuracy. This implementation of predictive coding learns progressively from expert attorney review examples, applies that learning to large review sets, and returns statistical samples for quality control. Ideally it will apply defensible review to large data sets in a fraction of the time that a team of reviewers could do it. In practice it is not nearly this easy, but it certainly holds promise for several review tasks.
Our take: Predictive coding for review is very promising but there is no case law for machine-automated document review. This is a serious problem for companies pushing predictive coding as review automation. We see three major drivers that will push for broader adoption: 1) established precedence, which will take some highly motivated trailblazers to accomplish; 2) corporate clients demanding a radical reduction in review costs and time, and 3) law firms building a competitive review practice who are willing to experiment with predictive coding.