Authored by Katherine J. Neikirk
This article was originally published in the Delaware Business Court Insider | November 14, 2012 

Courts outside of Delaware have addressed the use of predictive coding in discovery and now the Court of Chancery has addressed this technology in a transcript ruling in EORHB v. HOA Holdings, C.A. No. 7409-VCL (Del. Ch.). Predictive coding is a technology that allows attorneys to train a computer program to review documents for responsiveness. Predictive coding is supposed to reduce the costs of document production by cutting down on the amount of document review by attorneys. Instead of associates, contract attorneys or overseas lawyers reviewing hundreds of thousands of documents for responsiveness, a computer program codes documents as responsive or nonresponsive.

Attorneys may still review documents coded responsive by the computer program and do a quality control sampling review of documents coded as nonresponsive, but the goal is to cut down on the amount of attorney-review time. To train the computer program, attorneys will code a "seed set" of documents. The computer program then codes the remainder of the documents responsive or nonresponsive based upon the seed set. Attorneys will review some or all of the results in order to further train the computer program. According to some studies, predictive coding is at least as accurate, if not more accurate, than human review.

In EORHB, which involved indemnification claims arising from a sale transaction, the court raised predictive coding sua sponte after granting partial summary judgment in favor of the defendants and refusing to dismiss the defendants’ counterclaims for indemnification. Vice Chancellor J. Travis Laster informed the parties that he believed EORHB was the type of non-expedited case where the parties would benefit from using predictive coding. Not only did Laster indicate he was in favor of predictive coding, he also ruled that if the parties did not want to use predictive coding, they would have to show cause for why predictive coding was inappropriate. He further ordered the parties to discuss using a single discovery provider and if they could not agree upon one, to submit a list of discovery providers from which he would select a discovery provider for the parties to use.

So does EORHB mean that all parties appearing in the Court of Chancery must use predictive coding? Well, no. Laster indicated that EORHB was a non-expedited case where predictive coding made sense. Predictive coding will not always make sense (at least not yet). Predictive coding is a relatively new technology and, thus far, not inexpensive. Costs will likely drop as more parties use predictive coding and more vendors offer it, but for now, it can be expensive. Predictive coding makes the most sense in a case with lots of custodians, lots of documents and high stakes. For example, in Da Silva Moore v. Publicis Groupe & MSL Group, a Southern District of New York case where the parties had agreed the defendants would use predictive coding, there were more than three million documents collected. Predictive coding currently makes less sense when there are few custodians, few documents and low stakes. If, for example, the ultimate recovery is $100,000, then it is unlikely the Court of Chancery would order, or the parties would wish to utilize, predictive coding, the cost of which could exceed the ultimate recovery.

The EORHB ruling suggests that the Court of Chancery could be open to a party wishing to use predictive coding, under the right circumstances, against the wishes of the opposing party. In teeing up a predictive coding dispute for the Court of Chancery, parties should keep in mind the general practices for raising discovery disputes in the Court of Chancery. First, it would be risky to use predictive coding without informing the other side. The better practice would be to first discuss with counsel for the opposing party the possibility of using predictive coding and how it would be used. Using predictive coding without informing the other side creates a risk of having to redo a document review and production at even more cost when the other side finds out and points out flaws in how predictive coding was used. Second, a party must be prepared to explain to the other side and the court why predictive coding makes sense or does not make sense. Claiming predictive coding is "cool" or "too untested" or "too expensive" is not enough; a party must be prepared to explain why predictive coding makes sense or does not make sense in its case. A party should be ready to provide specifics. For example, x number of custodians over x number of years could mean x number of documents and x number of dollars in review and production costs versus predictive coding costs. Third, a party should not run to the court without first discussing the issues with the other side and trying to find common ground. The parties may be able to narrow their disputes or find a way to use predictive coding without having to involve the court at all.

Predictive coding could be a way to reduce the high costs of e-discovery and the Court of Chancery has indicated that it is open to the use of predictive coding. Accordingly, attorneys litigating in the Court of Chancery should be prepared to use this technology in the right case.