News, Developments and Practical Advice on eDiscovery in the trenches of Litigation

Discovery Advocate

Home | Discovery Advocate

Magistrate Judge Joe B. Brown’s recent order permitting predictive coding in <i><a href="https://discoveryadvocate.bakerhostetlerblogs.com/wp-content/uploads/sites/12/2014/08/BridgestoneAmericasvIBM.pdf">Bridgestone Americas v. International Business Machines Corporation</a></i> has received <a href="http://www.recommind.com/blog/2014/08/12/bridgestone-v-ibm-approves-predictive-coding-use-rejects-progressive">a lot of attention</a> because it allowed the use of predictive coding on a population of documents that had already been screened using keywords pursuant to a case management order that did not provide for the use of predictive coding.

The question whether predictive coding should be deployed on a document population that has already been winnowed using keywords is a hot topic. Opponents say keywords can be a blunt instrument and are often developed and applied by those who don’t have a grasp of the range of claims and defenses at issue or the language used by the litigants and their key players. For example, is this a company that refers to the “business plan” as a “playbook”?

But where parties have already agreed to keywords and the use of a keyword screen, and the resulting document population is still sizable (in this case about two million documents), others say machine learning is a reasonable tool to separate the likely relevant from the likely irrelevant documents. In other words, advocates of this approach argue that the harm done – if any – through the use of a keyword screen will not be remedied by manual review of the resulting document population and a good predictive coding work flow should get at what matters more quickly.

While this debate is interesting, it is not really the focus of Judge Brown’s order which, instead, reflects a practical approach to the problem presented by the parties. It is this practical approach that provides the teachable moment for litigators, regardless of where they fall on the predictive coding learning curve. In short:
<ul>
	<li>The “uses of predictive coding is a judgment call.” Presumably Judge Brown would also agree that the choice <i>and implementation</i> of any search and review process involves many judgment calls throughout the process (albeit judgment calls that can be audited, evaluated, and fine-tuned based on the results of the process).</li>
	<li>Discovery is imperfect and – given a population of millions of documents – “[t]here is no single, simple, correct solution possible under these circumstances.”</li>
	<li>Imperfection increases the need for cooperation and transparency. Judge Brown concluded “openness and transparency in what Plaintiff is doing will be of critical importance,” and stated his expectation that the parties “communicate, through their attorneys and experts and companies doing the work, on a frequent and open basis.”</li>
	<li>Given the possibility that differences and disputes will arise as the discovery process and the application of advanced technology continues, an informal dispute resolution process is critical to keep the case moving forward. Judge Brown directed the parties to try to resolve their differences informally, but if they come to a “sticking point” that might create “unnecessary wrangling,” “not to allow it to go too far before scheduling a telephone conference or request an in-court hearing with the Magistrate Judge.”</li>
</ul>
This case, and the others in which bench and bar wrestle with the best way to find what matters in massive document populations, point to <a href="http://www.bakerlaw.com/files/uploads/News/Articles/LITIGATION/RPRTKeteltasInsightJAN-2014.pdf">the continuing need for search process quality standards</a> that are understandable to, and become part of the day-to-day workflow of, the participants in the legal process.

Predictive coding after keyword screening!? Don’t miss the point of Bridgestone Americas