By Regina A. Jytyla
It is estimated that first-level document review comprises between 58 percent and 90 percent of litigation costs. At this price-point, the cost of document review can be untenable, especially when the amount in controversy does not support the cost of discovery. Recent court decisions have laid the groundwork for a reliance on technology to bring efficiencies to this costly process.
In the 2010 case of Multiven v. Cisco Systems, Inc., the federal court in San Francisco intervened to expedite the discovery and review process. With a slow-moving document review, numerous discovery disputes and a looming deadline, the parties were ordered to split the costs of an e-discovery vendor and the court appointed a special master to oversee the remainder of the discovery process.
Multiven highlights the challenges of modern discovery - managing costs, meeting deadlines and conducting efficient reviews - that can be resolved through supplementing discovery projects with technology.
Intelligent review technology (IRT) applies human logic in an automated fashion, resulting in increased speed, consistency and accuracy. Workflow automation, supervised learning and statistical quality control are the cornerstones of an effective IRT system that allow review to be conducted faster, more efficiently and accurately than even the best human review teams equipped with current discovery technology.
IRT begins at the foundation of review by offering a solution to workflow management: a tedious process with relatively simple substance and highly complex logistical qualities that make it ripe for automation.
Rather than relying on a finite number of human review administrators with limited capacity, automated workflow provides greater processing bandwidth resulting in increased work capacity, a remarkably high attention to detail and the flexibility to handle varying workloads. Managing the review process on an automated workflow platform allows the system to simultaneously track all stages of the project and produce accurate, real-time metrics, progress reports and statistical analysis.
Another important feature is supervised learning, which replaces the need to rely on a search expert to build search queries or predictive models.
Instead, a supervised learning algorithm creates a classifier, which is a predictive model capable of sorting through documents to determine the responsiveness and then assigning each document to the correct class with a high degree of accuracy.
This technology iteratively analyzes samples of documents that have been classified and categorized by a human reviewer in order to "learn" the distinguishing characteristics of what constitutes responsive v. nonresponsive, privileged v. nonprivileged, etc.
The resulting expression of this difference is the classifier, which is then applied to nonreviewed documents. The result is a rating system that can be used to prioritize documents on a graduated scale of responsiveness rather than a more tenuous one-or-the-other categorization.
The substantial benefits of this technology include more effective search queries, greater consistency and a reduced need for human resources.
Statistical quality control is also a critical part of any review process, whether conducted traditionally or using supervised learning.
Through the use of IRT, statistical sampling guides the review process by accurately identifying the most responsive documents, allowing the human reviewers to be appropriately tasked to those documents in which their skills are most suited.
Perhaps more importantly than guiding attorneys during the review process, sampling can help signal when the review should end by estimating the proportion and probability of responsive documents remaining.
In Mt. Hawley Ins. Co. v. Felman Prod. Inc., the Southern District of West Virginia made it clear that sampling is a critical quality control process that should be conducted throughout the review.
The court found the plaintiff failed to perform this task, which weighed heavily on its decision to waive privilege on an inadvertently produced "smoking gun" e-mail aptly referred to by the court as the "proverbial tip of the iceberg," considering it is almost certain to sink the plaintiff's claim.
Although complicated and challenging, proper statistical sampling likely could have prevented this costly error from occurring. IRT makes the complicated sampling process simpler and more accurate by testing data samples to improve the accuracy and efficiency of calculating statistical data, ultimately resulting in a more defensible and reliable review.
As the scope and complexity of e-discovery becomes increasingly arduous, the disparity between cost and effectiveness will only increase. IRT pinches the gap and applies logic to deliver greater cost predictability, strengthened defensibility and increased speed.
Regina A. Jytyla serves as managing staff attorney at Kroll Ontrack. She tracks evolving law and emerging trends in corporate information management and legal discovery. Special thanks to Ben Kirk, law clerk at Kroll Ontrack, for his research and assistance in writing this article.