Evaluating The Fitness Of Predictive Coding

March 5, 2014

ediscoveryAs discussed in our Feb. 11 post, predictive coding is emerging as something of a Holy Grail of sorts for the legal industry. Many attorneys are hoping predictive coding will solve the issue of how to manage the volumes of electronic communication that, for better and for worse, are now part of the discovery process.

The availability of predictive coding means parties to litigation need to balance the attractive cost efficiencies it offers against the risks inherent with any new technology.

Here are a few things to keep in mind when evaluating the use of predictive coding:

Predictive coding still relies on humans: Both process and subject matter expertise are required to employ predictive coding effectively, neither of which can be attributed to a computer. Process expertise does not refer to in-depth knowledge of the machine-level technology, but rather the ability to employ predictive coding within a comprehensive workflow. Subject matter expertise is paramount to creating a seed set (the “rules” given to a predictive coding mechanism that tell it what to look for) that does not result in “garbage in, garbage out.” While using a seed set that has been generated from machine learning is possible with some predictive coding tools, the majority of predictive coding implementations still rely on subject matter experts in the creation of the seed set. Both process and subject matter expertise should not be underestimated. The up-front time investment required in these areas should be weighed against the savings expected to be achieved with the use of predictive coding.

Type and size of datasets matter: There are datasets that lend themselves to greater success with predictive coding than others. Specifically, if the dataset is not large enough to create a statistically meaningful seed set, or if your dataset includes a high volume of video, audio or image files (think of a patent or construction case with lots of diagrams and photos), it may not be the best candidate for predictive coding.

Technology isn’t foolproof: Predictive coding is not a silver bullet, but it can be a highly effective tool in reducing the number of documents requiring eyes-on review. Human review is notoriously error-prone and any opportunity to leverage technology to do the heavy lifting should always be considered. Predictive coding is not intended to replace attorney-reviewers completely, but when effectively implemented, can greatly increase the speed and accuracy of document review, resulting in overall cost savings for your clients.