In ESI Discovery: Are keywords, concepts and other searches simply antiques?
The discovery of electronically stored information (ESI) is loaded with potential pitfalls and failure unless the parties add two components to the mix: cooperation and collaboration. Lacking those components, ESI discovery, at least sometimes, can be one of the more painful experiences for the average trial lawyer.
The problem to overcome is largely that trial lawyers, by their nature, are competitive souls and tend toward competition rather than cooperation. Add to this personality that of the client who expects her lawyer to win everything, every time and we are off to the races.
In a recent case, the Honorable Magistrate Judge Peggy Leen seems to deal with overly competitive parties and lawyers not inclined toward collaboration; in the recent decision in Progressive Casualty Insurance v. Delaney, 2014 WL 2112927 (D. Nev. May 20, 2014).
The Progressive case is reported to involve approximately 1.8 million electronically stored documents from several different sources. Although we can not tell from the decision the date upon which the complaint was filed , we know that the parties had agreed to a Joint Proposed ESI Protocol order entered in October of 2013. The defendants moved to compel their discovery requests on December 27, 2013.
Throughout January, February and into March the parties struggled with meeting to resolve their ESI disagreements and the court cooperated by extending status conference deadlines. The history certainly paints a picture of two parties who could not reach agreements on the methods to be employed to manage the culling of ESI.
Apparently, at some point, the plaintiff and the defendant had agreed to keywords in order to conduct a search of the 1.8 million documents. Applying the agreed keywords to the whole collection resulted in 565,000 documents. The plaintiff agreed to begin a manual review of the resulting subset of documents and after contract attorneys had reviewed 125,000 of the documents the plaintiff realized that manual review would be time consuming and expensive.
Without consultation with the defendant or the court, the plaintiff decided to take the 565,000 subset documents and review them through the use of predictive coding technology. After application of predictive coding, the next subset of documents was reduced down to 90,575 potentially relevant documents. Applying privilege filters to the subset of 90,575 documents further reduced it to approximately 63,000 as less likely privileged documents.
The plaintiff maintained that the only way to cull down the document subsets was through predictive coding. Predictive coding is well described by the co-founder of the software company that Progressive (plaintiff) wanted to use:
“Predictive coding is essentially a learning technology,” says Warwick Sharp, a co-founder of Equivio, a company that develops text analysis software and is at the forefront of integrating predictive coding into its e-discovery tools. “What predictive coding is able to do is get input from a human being, who reviews samples of documents and marks them as relevant or not, and then those decisions are input into the predictive coding engine, which is able to generalize those decisions across the entire collection.”
There are some things that must exist for predictive coding to succeed in the culling of large document collections:
- The parties must have a real desire to cooperate together in the using the most efficient methods.
- The parties must collaborate in “near transparency” while laying down the structure to be followed in the predictive coding process.
- Developing the methodologies to be used must be mutually developed.
- The costs should be shared to engender a mutual interest in cooperation.
- Full access to the collection of “trained documents” should be granted to both parties.
The court ultimately discussed the original agreements by the parries to keyword searches; although conceding predictive coding might be a better approach. The plaintiff departed the cooperation and collaboration track when they unilaterally used predictive coding on the subset of documents realized through keyword searching. The court concluded that the first subset, 565,000 documents, should be produced for review to the defendant, without first reviewing for privilege or relevance. The court had clearly run out of patience with this discovery dispute.
What can be learned from this case? Is predictive coding ultimately the best approach to culling and coding large document collections? Have keyword, concept and emotive searches become antiquated?
Predictive coding technologies are not magic. Keyword, concept and emotive searching, while a little gray around the temples, still have a real place in ESI culling. Searches, if conducted properly, can play a valuable precursor to ultimately employing predictive coding and may even give better insight into training the predictive software.
What will not change and what has historically always been a hindrance to litigation that increased the costs for parties, particularly plaintiffs, is failing to cooperate. In today’s digital world, a resistance to collaboration further adds to the difficulty and expense of ESI discovery.