On February 24, 2012, a New York court issued the first-ever reported decision in the United States on computer-assisted predictive coding, approving it as "an acceptable way to search for relevant ESI [electronically stored information] in appropriate cases." See Moore v. Publicis Groupe, 11-civ-1279 (ALC) (AJP), 2012 U.S. Dist. LEXIS 23350 (S.D.N.Y. Feb. 24, 2012). Predictive coding, as explained below, is a method of using computer software to identify the relevant documents. Under the right circumstances, use of predictive coding can be a less expensive alternative to the traditional document-by-document review model. Because this is the first court decision to discuss predictive coding, it will likely influence how predictive coding is treated by other courts.

THE TRADITIONAL DOCUMENT-BY-DOCUMENT REVIEW MODEL

As anyone involved in litigation knows, the cost of discovery is often the single most expensive part of a case. The discovery process often involves millions of emails and other ESI that are then filtered using search terms, date restrictions, or other methods to reduce the number of documents for manual review. Even after this filtering process, discovery often involves teams of attorneys performing a document-by-document review of a substantial number of emails and other ESI for privilege and responsiveness. Not only is this process expensive, it can also lead to inconsistent results.

COMPUTER-ASSISTED PREDICTIVE CODING

Computer-assisted predictive coding has the potential to reduce the cost of document review while also increasing its accuracy. In predictive coding, senior attorneys carefully review a small sample of documents (a "seed set") and code the documents in the seed set as being relevant to various issues in the case. The computer software analyzes these documents and recognizes patterns in the documents, and soon the software is able to predict the likelihood that any given document is relevant to one of the issues in the case and tag it accordingly. Thus, after a limited amount of human effort, the computer can review all of the documents at issue and identify those that are likely relevant. This method can be cheaper than having attorneys review every document. It can also promote greater consistency of results. Until recently, it remained to be seen whether courts would endorse this process as a substitute for the traditional document-by-document review.

MOORE V. PUBLICIS GROUPE

In Moore v. Publicis Groupe, Magistrate Andrew J. Peck, one of the leading e-discovery jurists, a proponent of the predictive coding model, and an author of an important recent article endorsing the use of predictive coding, issued the first decision specifically addressing the use of predictive coding as a replacement for the traditional document-by-document review Importantly, in the case, both parties agreed to use predictive coding to locate relevant emails and ESI. But the two sides later disputed the manner in which predictive coding would be implemented. Among other things, the plaintiffs expressed concerns as to both the accuracy of the original coding and the possibility that the software would overlook relevant documents.

Partially in response to these concerns, and partially of its own initiative, the court imposed a number of protocols for performing the automated document review process. Perhaps most significantly, the court required that the party producing documents would be required to give the other side its entire coded "seed set" (excepting only privileged documents). In other words, after the senior attorneys coded the initial documents as either relevant or not relevant to issues in the litigation, all of the non-privileged documents (whether relevant or not) would be turned over to the opposing party to review. This process would allow the opposing party to verify that the initial coding was correct, and that no systematic errors were created through the initial coding process. Additionally, the court decided that, in order to ensure the accuracy of the process, the producing party would be required to continue to turn over the non-privileged documents judged by the software to be irrelevant, at least during the early successive rounds of coding. Of course, this process might require the producing party to let the other side examine what could be a significant number of non-relevant documents that would ordinarily not be produced.

As the Moore court noted, both parties had agreed to use predictive coding and merely disputed the method of implementation. Nonetheless, the court discussed how it might rule if the parties had not agreed that predictive coding was appropriate. Where the producing party wants to use predictive coding and the requesting party objects, the court suggested that it might overrule that objection. "The question to ask in that scenario is what methodology would the requesting party suggest instead? Linear manual review is simply too expensive where, as here, there are over three million emails to review. Moreover, . . . statistics clearly show that computerized searches are at least as accurate, if not more so, than manual review." Id. at *28-29. Where the requesting party wants the producing party to use predictive coding and the producing party objects, the court declined to suggest how it might rule, simply noting that "[t]he tougher question . . . is whether the Court, at plaintiff's request, should order the defendant to use computer-assisted review to respond to plaintiffs' document requests." Id. at *29 n.10. These questions will undoubtedly be answered in upcoming decisions.

PRACTICAL IMPLICATIONS

Companies with large amounts of email and other electronic documents may want to consider using computer-assisted predictive coding to identify relevant documents. Predictive coding may end up being cheaper and more accurate than traditional methods of document review.

Likewise, a requesting party may find predictive coding attractive in that it allows software to perform the judgment as to which documents are relevant, thereby reducing the danger of opposing counsel withholding relevant documents. The requesting party may also find it advantageous to review all the documents in the seed set used to train the software.

Nevertheless, if courts are going to require all non-privileged documents in the seed set to be turned over to the requesting party, some companies will justifiably be wary of using predictive coding in the wrong type of matter, such as a case against a competitor or that concerns sensitive technology. Likewise, because judicial approval of predictive coding remains an area of uncertainty, some companies may be concerned about using predictive coding technology if ultimately a judge may decide that the traditional document-by-document review method should have been used. The more conservative approach, for now, is to obtain approval from opposing counsel or the court before using predictive coding technology.

Companies should consult with counsel regarding these issues before deciding whether or not predictive coding is appropriate for their case.

Because of the generality of this update, the information provided herein may not be applicable in all situations and should not be acted upon without specific legal advice based on particular situations.

© Morrison & Foerster LLP. All rights reserved