Rule 1 of the Federal Rules of Civil Procedure calls upon courts and litigants to “secure the just, speedy, and inexpensive determination of every action and proceeding.” And so, it comes as no surprise that technology assisted review (“TAR”) is being widely embraced by the legal profession.
What is TAR?
TAR (also called predictive coding, computer assisted review, or supervised machine learning) is a document review process where humans work with software to train the software/computer to identify relevant documents.* The goal of TAR is to effectively and efficiently categorize documents. The potential for significant savings in cost and time, without sacrificing quality, is what makes TAR so appealing.
The process is relatively straight forward. First, electronic documents that have been preserved and collected are loaded onto a platform where the software builds an index during which each document’s text is analyzed.** Next, a human reviewer who is knowledgeable about the issues and circumstances of a lawsuit reviews and codes/tags documents as “responsive” or “non-responsive.” This “seed set” of documents are used to “train” the computer. This information is then ingested by the computer, and used to draw inferences about documents that have not yet been reviewed. The computer “learns” from the human reviewers’ designations which combination of terms or other features occur in responsive documents and the computer then develops a model that it uses to predict the coding on the remaining documents.
However, TAR should not be applied blindly. Rather, quality control and testing are essential to confirm the accuracy of decisions made by the software. There are a myriad ways to perform QC, including testing random samples of the predicted responsive set before production. It should be noted that there is no standard measurement to validate the results of TAR. Rather, it is based upon reasonableness and proportionality considerations.
For those of you interested in learning more about TAR and best practices associated with TAR, consider reading the Technology Assisted Review Guidelines released by Duke Law School in February.
*The phrase “technology assisted review” can imply a meaning broader than used in this blog. For example, “TAR” could encompass non-predictive coding techniques such as “clustering.”
**Although software algorithms differ, most analyze the relationship between a document’s words, word order, characters, and repetitive text. This analysis, in turn, allows the software to compare one document to another.