A relatively run-of-the-mill employment class-action lawsuitrecently set the legal technology world aflame. The case, Monique Da Silva Moore, et al. v. Publicis Groupe & MSLGroup, centers on five women who accused advertisingconglomerate Publicis Groupe of gender discrimination. But littleof this is relevant to the fire. Instead, a discovery disputebetween the parties ignited the blaze, providing the perfect stormfor what is considered the first judicial opinion to endorse predictive coding as adefensible way to review documents.

Magistrate Judge Andrew Peck of the U.S. District Court inthe Southern District of New York opined in his Feb. 24 decisionthat defendants could use predictive coding, a software tool thatuses algorithms to automatically tag documents, to review as manyas three million electronic documents as part of the parties'e-discovery protocol. Because the plaintiffs and defendants bothagreed to use predictive coding to cull from the massive number ofdocuments, the case serendipitously allowed Judge Peck, who is awell-known e-discovery technology advocate, a chance to plant hisflag and advance the field.

Although many technology vendors may point to Judge Peck'sopinion as a validation of predictive coding as a judiciallyendorsed product, this is not necessarily the case.

“This opinion provided him with a way to make his 'Search,Forward' article (that offered tentative, circumstantial judicialapproval of predictive coding) more official,” says James Hanft,counsel at Schiff Hardin. “This doesn't have that much legal weightto it because both parties agreed to it. It simply gives insightsinto what [Judge Peck] and a lot of other judges are thinking inrespect to the acceptability of computer-assisted coding.”

Getting Transparency

What Judge Peck does say in his opinion is that predictivecoding is an appropriate tool in certain cases, but only if it ispart of a defensible process that is subjected to the samequality-control testing appropriate to any type of documentreview.

“Judges don't want you to be perfect,” explains Cozen O'ConnorMember Dave Walton, “they want you to be reasonable. Usingpredictive coding—if it's been tested and sampled, there ismore-than-adequate research behind it to show that it's reasonable.But that still doesn't mean predictive coding is good for everytype of case.”

Another important point in Judge Peck's opinion that may havebeen lost in the furor is the issue of transparency. He wrote thathe granted the defendant's e-discovery protocol, in part, becauseit was willing to be fully transparent with the plaintiffs and showthem the results of the sampling as well as the results of themetrics to ensure nothing was missing.

“If you show you've got nothing to hide, a judge is going togive you the benefit of the doubt if you're trying to employcreative means and tools to reduce the ESI blob,” Walton says.

Despite the transparency, however, as of press time, there arestill unresolved issues. The plaintiffs have appealed Judge Peck'sruling to U.S. District Judge Andrew Carter for the finaldispensation of the predictive coding workflow, alleging Peckdidn't give them time to address his ruling, and that he expandedthe reasoning of his ruling after they filed their objections. Andsome of these questions are likely to be inherent to most casesinvolving predictive coding.

“The discussion around the disagreement about the approaches andthe best way to put the technology to use is an important one,”says Mikki Tomlinson, director of the strategic consulting servicesdivision for EDJ Group. “There are multiple disagreements that arebeing played out: How big should the sample sizes be? At what pointshould you stop? At what point are the parties going to call systemtrained? How is it being validated?”

Of course, questions like these will always be present and willvary on a case-by-case basis. But given time and experience usingpredictive coding, parties eventually will become more accustomedto it, and getting those answers will become easier.

Predicting Use

Judge Peck's opinion raises the question: When is the use ofpredictive coding appropriate? Experts agree that right now, giventhe high cost of the technology, it's best used in cases with asignificant amount of data to warrant that cost. Also necessary atthe moment is a large enough set of responsive documents within thedata collection to allow the software to be effective. Vendors,however, are working on adapting their tools to make predictivecoding more appropriate for smaller cases.

“What predictive coding is really going to do is replace thearmy of contract lawyers that are out there reviewing thedocuments,” Walton says.

Another point to consider is that, for predictive coding to beeffective, documents need to be in a form that the program caneasily parse. Large document sets that are PDFs will need to beconverted to a machine-readable form before being reviewed, whichcan create problems given the conversion accuracy of opticalcharacter recognition programs. Similar issues occur withhandwritten notes, drawings and other non-email documents.

There also can be problems with words, word association orpattern recognition. Attorneys see case after case with the samewords used over and over, inherently know context and can readbetween the lines. Predictive coding, as effective as it is, cannotalways find those same patterns.

“[With predictive coding,] you're going to miss the one-offthing where someone may use different word patterns or differentwords to say the same thing,” Hanft says.

Up to Code

At the moment, there's still more smoke than fire when it comesto predictive coding. It's exciting and buzzed about, but not a lotof companies or law firms are regularly using it.

“It's definitely being used,” says Symantec Corp. SenioreDiscovery Counsel Dean Gonsowski, “but it's kind of aPh.D.-level-use case right now. The workflows, the processes andthe understanding of how a traditional workflow comes into playwhen you get to predictive coding are magnified. It has a place,but you've got to get your graduate degree first, otherwise you'regoing to have litigators' heads exploding.”

Until those issues are solved, or at least better ironed out,mainstream adoption of predictive coding will remain limited.Gonsowski believes that 2012 is the year for awareness andexperimentation, and forward-thinking attorneys will start trying afew cases using the technology. By next year, he predicts, some ofthe transparency issues will have gone away, a workflow will havebeen established, and case law will have developed for attorneys touse predictive coding with greater regularity.