Predictive Coding, Seed Sets, and the Work Product Doctrine


Predictive Coding, Seed Sets, and the Work Product Doctrine

  • 1 Tags

By Philip Favro


The use of predictive coding has been both contagious and controversial since its use began in eDiscovery. On the one hand, predictive coding has found welcome recipients in clients, counsel, and the courts, all of whom are seeking to expedite the ESI search and review process. Nevertheless, there have been disagreements regarding various aspects of predictive coding. One of the key items in dispute concerns whether the attorney work product doctrine should protect the documents that counsel selects to help train the predictive coding process. This is a significant issue which, if left unresolved, could very well undermine the proliferation of this search and review methodology.

There is little disagreement over the significance that those documents – commonly referred to as training or seed sets – play in a properly functioning predictive coding workflow. Seed set documents are essential for training a predictive coding algorithm to identify the documents that “are the most relevant to the case and most representative of those for which each side is looking.” Where the predictive coding cognoscenti – and courts – disagree is whether a lawyer’s selection of seed set documents is protected from discovery by the attorney work product doctrine.

Some courts and commentators have taken the view that counsel should identify seed set documents for its litigation adversaries despite the potential for disclosing work product. Those who support the disclosure of seed set information justify their position on the need for transparency and greater certainty in the predictive coding process. Disclosure, they argue, will reduce satellite litigation over the process the party used to search for, review, and produce responsive information. All of which could arguably make discovery less costly, more efficient, and ultimately focused on disclosing information to enable the parties to resolve matters on the merits.

The allure of this position, however, fails to recognize that a seed set may reflect a lawyer’s perceptions of relevance, litigation tactics, or even its trial strategy. These conclusions regarding key strategic issues – reflected in counsel’s selection of documents – have frequently been protected in analogous circumstances as work product since they may reflect counsel’s “mental impressions, conclusions, opinions, or legal theories.” There are several cases that are instructive on this aspect of the work product doctrine.

For example, in Disability Rights Council of Greater Washington v. Washington Metropolitan Transit Authority, the court protected as fact work product the plaintiffs’ selection of certain documents made available during an inspection of a third party’s documents. Forcing the plaintiffs to divulge the precise documents that they selected from the third party could improperly reveal their counsel’s “selection process” and therefore, its “mental impressions.”

Similarly, the court in United States ex rel. Bagley v. TRW, Inc. protected the defendant’s selection of certain documents made available for inspection by the government as opinion work product. While the court acknowledged that full disclosure of that information would advance the laudable objectives of “conservation of resources” and “judicial efficiency,” such considerations would have to yield to a legitimate work product claim:

[T]he document production procedure insisted upon by the government would reveal the opinion work product of TRW’s counsel and give the government a free ride on the effort and investment of TRW’s counsel in reviewing and selecting documents and in preparing TRW’s defense.

The rationale from Disability Rights and Bagley is equally applicable to the issue of seed set protection. Just like in Bagley, opposing counsel should not be given a “free ride on the effort and investment” of counsel who meticulously identifies specific documents for its predictive coding seed set. And as cautioned by Disability Rights, allowing the opposition to understand counsel’s “selection process” would wrongfully enable it to access counsel’s “mental impressions” and prepare its case accordingly. Under these cases and other authorities, the identity of those selected documents should likely receive some form of work product protection.

While there are limitations to this rule and though there are potential benefits to disclosure, the convenience of cooperation should not be used to coerce a lawyer into disclosing its work product as reflected in the development of a predictive coding seed set.

About the Author

Philip Favro brings over fourteen years of expertise to his position as Senior Discovery Counsel for Recommind, Inc. Phil is an industry thought leader, a global enterprise consultant, and a legal scholar on issues such as eDiscovery, information governance, and data protection. Phil’s expertise has been enhanced by his practice experience as a business litigation attorney in which he advised a variety of clients regarding complex discovery issues. Phil is a member of the Utah and California bars. He is an active member of the American Bar Association and also contributes to Working Groups 1 and 6 of The Sedona Conference. Phil also serves as a Judge Pro Tempore for the Santa Clara County Superior Court based in Santa Clara, California.