Predictive Coding: A Better Way to Deal with Electronically Stored Information

Download a printable PDF of this article.

Litigators bemoan the fact that the volume of documents and data produced and requested during discovery continues to increase year on year—as does the volume of electronically stored information (ESI) that clients must preserve. There is a glut of ESI arising as a consequence of the exponential growth in the creation and exchange of electronic files and email traffic. The result is that even in cases with a relatively low monetary value, there are often still hundreds of thousands or even millions of files relevant to the case, and a significant portion of the cost of litigation is driven by the obligation to preserve, search, review, and produce this ESI. However, the same computing power that has provided this overabundance of data may also provide the solution for reviewing, coding, and producing it in litigation with minimal human review.

Computer-assisted review, technology-assisted review, and “predictive coding” all refer to the use of computer-generated algorithms—based on human modeling—to code documents without the need for document-by-document human review. Predictive coding “predicts the relevance of discovery documents [or relative match of any document to the model set] based on the prior coding of a small sample of discovery documents by an attorney.” Nicholas Barry, Man Versus Machine Review: The Showdown Between Hordes of Discovery Lawyers and a Computer-Utilizing Predictive-Coding Technology, 15 Vand. J. Ent. & Tech. L. 343, 344 (2013). In other words, predictive coding is an iterative process because it depends on the coding first applied by attorneys familiar with the case and responds as the attorneys continue to interact with the documents. The computer becomes an extension of the attorney reviewer—coding documents in the review universe based on the algorithms derived from a model created by the attorney.