Play Nice: Recent eDiscovery Decision Emphasizes Consistent Cooperation

June 24, 2014
6 minutes

In the weeks since Magistrate Judge Leen denied a late request to use predictive coding in a case before her, the e-discovery world is still abuzz over certain language included in her ruling. In the case in question, Progressive Casualty Insurance v. Delaney, 2014 WL 2112927 (D. Nev. May 20, 2014), Progressive had implicitly agreed to transparency when it agreed to partake in an extensive joint ESI protocol, including detailed agreement on search terms. The court found that Progressive violated this agreement when it unilaterally reversed course and applied predictive coding. And for this dissonant approach to cooperation, it paid dearly. Any practitioner considering predictive coding[1] should fully consider the judge’s reasoning and the potential pitfalls associated with failure to consistently cooperate. 

The Progressive parties had submitted a Joint Proposed ESI Protocol, which the court approved. Pursuant to the Protocol, Progressive applied search terms to the documents it collected, narrowing the number of documents requiring manual review from 1.8 million to 565,000. Progressive then used contract attorneys to review 125,000 of the 565,000 “hit” documents. At that point, Progressive determined that manual review of the remaining documents would be too time-intensive and expensive. Unilaterally, Progressive decided to apply predictive coding to the remaining 440,000 documents. When Progressive later sought permission to amend the ESI Protocol to allow for predictive coding, the court firmly rejected Plaintiff’s amended proposal, stating that it lacked the “unprecedented degree of transparency and cooperation among counsel” that cases approving the use of technology-assisted review have “required.” Id. at 10. Instead, the court granted FDIC’s motion to compel and ordered Progressive to produce all 565,000 “hit” documents, without review for relevance (a privilege screen was allowed).

The takeaway from Progressive should not be that predictive coding cases require complete transparency on the part of producing parties. Instead, e-discovery practitioners should be mindful throughout the discovery process of the parameters and implications of their cooperation efforts, and avoid creating unreasonable limitations for themselves.

When it comes to discovery, cooperation is the new black. Many judges have signed the Sedona Cooperation proclamation, and courts as well as commentators have urged parties to cooperate on various issues throughout the discovery process. Most authorities agree that “failure to engage in cooperative discovery is likely to increase the costs, and complexity, of litigation to the detriment of everyone involved.” Paul W. Grimm & Heather Leigh Williams, ‘The [Judicial] Beatings Will Continue until Morale Improves’: The Prisoner’s Dilemma of Cooperative Discovery and Proposals for Improved Morale, 43 U. Balt. L. F. 107, 115 (2013). Cooperative discovery involves not just refraining from abusive practices, but also developing, testing, and agreeing on the nature and scope of information sought (to the extent consistent with clients’ interests). The Sedona Conference, The Case for Cooperation, 10 Sedona Conf. J. 339, 339 (2009 Supp.).

Parties may find it challenging, however, to put these principles into action. Progressive illustrates that even when parties cooperate to develop extensive ESI protocols at the beginning of a case, they may find later that another technology would be more useful. As the producing party, Progressive might have had more room to determine the best approach had the ESI Protocol allowed a flexible approach, permitting certain judgment calls by the producing party.  See, e.g. The Sedona Conference, The Sedona Principles: Second Edition Best Practices Recommendations & Principles for Addressing Electronic Document Production (June 2007), Principle 6 (“Responding parties are best situated to evaluate the procedures, methodologies, and technologies appropriate for preserving and producing their own electronically stored information.”).

Nor have the courts been entirely clear about the degree of cooperation required. The Progressive court did not define what it meant by “unprecedented degree of transparency and cooperation,” but it is debatable that any court has actually mandated the transparency this court assumed to be consistent across federal courts. For example, in both Da Silva Moore v. Publicis Groupe, 2012 WL 607412 (S.D.N.Y. Feb. 24, 2012) and In re: Actos, 2012 WL 7861249, MDL No. 6:11-md-2299 (July 27, 2012), the courts praised the parties’ agreement to share protocols and seed sets (documents used to train the predictive coding tool), but did not order it. Even in In re Biomet M2a Magnum Hip Implant Prods. Liability Litig., No. 3:12-MD-2391 (N.D. Ind. Aug. 5, 2013), where the court urged the parties to cooperate, the court recognized that cooperation has limits and refused to order the sharing of seed sets. Biomet at *2. 

Undoubtedly, the courts’ reluctance to order full transparency stems, as it should, from the fact that full transparency would give the opposing party access to attorney work product. In recent cases, courts have found that a variety of documents comprise opinion work product and need not be produced: documents related to methods for review and retention, S.E.C. v. Schroeder, No. C07-03798 JW (HRL), 2009 WL 1125579, at *12 (N.D. Cal. 2009); attorney’s instructions on how to conduct a computer search, Lockheed Martin Corp. v. L-3 Comm. Corp., No. 6:05-CV-1580-Orl-31KRS, 2007 WL 2209250, at *10 (M.D. Fl. July 29, 2007); and compilations of documents organized by legal theory, Kodak Graphic Comm.’s Canada Co. v. E.I. du Pont de Nemours & Co., 2012 WL 413994, at *4 (W.D.N.Y. Feb. 8, 2012).

Even more basic, requiring parties to disclose seed sets containing irrelevant documents “reaches well beyond the scope of any permissible discovery by seeking irrelevant or privileged documents.” Biomet at *2. The reason to deny such requests “seems self-evident,” considering that Federal Rule of Civil Procedure 26(b) limits the scope of discovery to relevant, nonprivileged matters. Id. at *1. Judge Richard G. Andrews of the District Court of Delaware also recognized that sharing seed sets would be a significant and unjustified departure from current discovery practices: “[w]hy isn’t it something where they answer your discovery however they choose to answer it…? How do you get to be involved in the seed batch?” Transcript of Discovery Dispute at 16, Robocast, Inc. v. Apple, Inc., No. 11-235 (D. Del. Dec. 5, 2012).

Another aspect of the Progressive case is troubling. The court criticizes Progressive for going forward with predictive coding in a manner inconsistent with the vendor’s “best practices.” The court does not elaborate on what specific “best practices” those are, but one could read that the court is referring to the vendor’s preference to load “all” data collected before applying predictive coding. That technique is much debated among experts and certainly has not been established as a best practice across the board. If the Progressive case holds that when a party uses predictive coding, all data must be loaded, without first culling using search terms, then the ruling is indeed problematic. Courts should not adopt this method as a gold standard unless experts reach some consensus on general methodology. Moreover, such consensus would have to allow for significant variations among vendor tools, case-specific needs and facts.

In the context of the existing case law, the language in Progressive is best read as a frustrated court’s response to Progressive’s unilateral decision to use predictive coding, despite a pre-existing joint Protocol with no such provision. Rather than taking Progressive as a signal that complete transparency has become mandatory, litigants should continue to explore what practices are best for each case, and continue to evaluate the most appropriate and defensible way to incorporate each vendor’s tool into their discovery workflow.

[1] Because the court used the term “predictive coding” instead of “technology-assisted review,” this article will use the term chosen by the court.