January 28, 2013
In complex litigation, document review is often the timeliest and most costly aspect of the discovery process. A recent ruling in a Virginia court and a 2011 article published in the Richmond Journal of Law & Technology may change that. In June 2012, a Virginia state court judge granted a party’s motion to use predictive coding over the objection of the opposing party. Predictive coding is defined as software applied to a sample set of documents to identify responsive records. In the Virginia case, lawyers coded a sample of 5,000 documents (out of 1.3 million) as relevant or irrelevant. Through the use of predictive coding, a total of 173,000 documents were identified as relevant. Lawyers reviewed about 400 documents the program identified as relevant and found that approximately 80% were indeed relevant. The lawyers also checked a sample of irrelevant documents, finding that about 2.9% were possibly relevant. Thus, the program correctly identified documents 81% of the time, that’s 20% higher than the accuracy rating associated with manual review.
It certainly appears that predictive coding has a lot of benefits including:
-Cost and efficiency
-A reduction in the need to hire contract attorneys to review numerous documents
However, there are risks involved. Predictive coding is still relatively new and not widely used. It is not fully understood and is inconsistently applied. Also, there is risk that confidential or privileged information could be revealed. And human judgment remains integral to the process as lawyers code the sample set of documents that ultimately create the algorithm upon which the program is based.
Is predictive coding right for every type of case? Can a sample set of documents be coded clearly and consistently? Will the use of predictive coding help prioritize documents for manual review or reduce the volume of documents for manual review? The answer is, “it depends.” One thing is clear, lawyers will want to be actively engaged in the process, whether predictive coding is used or not.
Have you used predictive coding in a case? If so, what was your experience? Would you use it again or would you rather rely on manual review?