In Part One of this article, we summarized how electronic discovery’s All-Seeing Eye has grown bigger than its stomach, with parties and their attorneys churning out and reviewing hundreds of thousands of documents (and dollars!), only to see less than 1 in 10,000 wind up at exhibit lists at trial — with fewer still actually used. We discussed how, to combat this trend, two sets of courts — our Fellowship of the E-Discovery Kings — set on journeys to narrow the range of the All-Seeing Eye in patent litigation by issuing helpful Model Orders for e-discovery. But whether there is really One Ring to Rule Them All when it comes to search methodologies may not be as clear as the E-Discovery Kings propose.
“Far below the deepest delving of the dwarves, the world is gnawed by nameless things.”
— J.R.R. Tolkien, Lord of the Rings: The Two Towers
Our earlier post discussed the fallacies of the Old Ways of keyword searching, including its focus on specific terms (the little picture) without accounting for context and other ‘nameless’ features. We ended with the question: what then, is the One Ring to Rule Them All, which can capture both the named and the nameless? And is there even one?
In parallel with our Fellowship’s journey and the forging of their Model Orders in the hot fires of patent litigation, another power was on the rise: predictive coding. While such computer-assisted review tools (which automate a major proportion of document review) have been around for several years, it was June 2011 when one of the larger e-discovery vendors was issued a patent on the process, thrusting it into even greater prominence. Unlike keyword search terms, predictive coding teaches computers to “predict” the relevant documents based not only on key terminology, but features like dates, names, broader phrases, and other items of context. It is estimated that predictive coding can save up to 70% of review costs.
Courts and law firms have been slow to adopt this new technology. No one wants to be the ‘guinea pig’ trying out a new technology. Predictive coding is also somewhat of a ‘black box,’ and the law firm using it cannot always know what is going on inside the program, which can be problematic when it comes time to defend their methodology. However, doubts are slowly beginning to fade. In the same month the Fellowship of the Eastern District of Texas forged its Model Order, Magistrate Judge Andrew Peck of the U.S. District Court for the Southern District of New York approved the use of predictive coding in Monique Da Silva Moore, et al. v. Publicis Group SA, et al, 2012 U.S. Dist. LEXIS 23350 (S.D.N.Y. Feb. 24, 2012), aff’d, 2012 U.S. Dist. LEXIS 58742 (S.D.N.Y. Apr. 26, 2012). That same week a state court judge approved the use of predictive coding over one party’s objection. Global Aerospace v. Landow Aviation , No. CL 61040 (Vir. Cir. Ct. Apr. 23, 2012). And more recently, a Vice Chancellor in Delaware Chancery Court required both sides to use predictive coding . . . even though neither requested it. EORHB, Inc., et al v. HOA Holdings, LLC, C.A. No. 7409-VCL (Del. Ch. Oct. 15, 2012).
It is thus becoming clear that the judiciary will incorporate predictive coding into e-discovery where appropriate. In fact, Judge Peck noted that computer-assisted review “should be seriously considered for use in large-data-volume cases.” Judge Andrew Carter, who affirmed Da Silva Moore, acknowledged that manual review with its keyword searching “is prone to human error and marred with inconsistencies from the various attorneys determination of whether a document is responsive.”
All that is gold does not glitter, Not all those who wander are lost; The old that is strong does not wither, Deep roots are not reached by the frost. — J.R.R. Tolkien, Lord of the Rings: The Fellowship of the Ring
Despite the advantages of predictive coding, it may still not be the One Ring to Rule Them All. There are times in which the Old Ways of keyword searching may still be the best candidate. In patent troll cases, for example, where discovery tends to be disproportionately heavy on the accused infringer’s side and lighter for the troll (who usually has no product or business other than patent licensing and enforcement to speak of), focused keyword searching may be sufficient. The same can be true for smaller patent and IP litigation disputes. Even then, however, the limitations of keyword searching make it prudent to use it only in conjunction with other search tools. And for non-patent troll cases, and for more complex IP disputes (for example, trade secret misappropriation), predictive coding may indeed present a new One Ring to Rule Them All, ensuring the capture of the greatest amount of relevant and responsive materials while still conserving costs. At the very least, it presents a viable alternative to the Model Orders’ presumption of keyword searching.
Unfortunately, at the time the Fellowship’s Model Orders were written, predictive coding — much like Strider, the Ranger of the North — had still not quite revealed itself as an Aragorn, a potential heir to the throne. If the Orders had come out today or a couple of years in the future, perhaps they would have accounted for alternate methodologies. The Seventh Circuit E-Discovery Pilot Program, for example, is seriously considering the merits and pitfalls of various search methodologies, including predictive coding. Other courts, such as the Northern District of California’s “Guidelines for the Discovery of Electronically Stored Information” released in November 2012, leave many of these numbers – search terms, custodians, etc. – to the parties to negotiate. While it may be that the court believes the parties best suited to determine search methodology, perhaps the Northern District is tacitly acknowledging that guidelines too mired in traditional searching may restrict rather than facilitate e-discovery. In fact, the checklist accompanying the Guidelines order the parties to discuss the “search method(s)” used, “including specific words or phrases or other methodology.” This language appears to leave the door open to predictive coding and other non-traditional methodologies.
For now there remains a disconnect between the two more specific Model Orders and the growing popularity of predictive coding. Moreover, the best practices remain to be written, both for utilizing predictive coding and for choosing the best keyword searches in standard search methodologies. Thus, even if we could choose and carry One Ring to Rule Them All through the rocky terrain of patent litigation, the path on which to carry it remains unclear — much as it was for Frodo Baggins:
At last with an effort he spoke, and wondered to hear his own words, as if some other will was using his small voice. “I will take the Ring,” he said, “though I do not know the way.”