Lifting the Barriers to Using Technology in eDiscovery

Editor’s Note: The authors of this post work at Ernst & Young.

By David Remnitz, Global Forensic Technology & Discovery Services Leader and Eric Johnson, Executive Director in the Fraud Investigation and Dispute Services practice at EY

In Insiders’ Guide to Technology-Assisted Review (TAR), we discuss how an explainable workflow, proper reporting, and cooperation between parties are the foundations of judicial acceptance of TAR. In fact, they are significantly more important than the application of any specific machine learning algorithm.

To paraphrase William Gibson’s famous remark about the future (The Economist, December 4, 2003): TAR is already here, it’s just not evenly distributed. TAR refers both to the application of review technologies across the spectrum of review activities, and to the popular understanding (or misunderstanding) of how statistics and available technologies apply to those activities. What we have come to appreciate is that technological advances have so far served largely to digitize the static environment of a paper document collection and review, essentially adopting for eDiscovery the linear workflow that has dominated legal process for decades. As such, these advances have not supported the need for transparency in furtherance of defensibility, nor have they cost-effectively addressed the core challenges imposed by the underlying nature of electronically stored information (ESI).

To be effective, the review process strategy must be supported by a tactical plan tailored to achieve a well-defined end goal. Timing is critical. The proposing party needs to define and document the process prior to any “meet-and-confer discussions” with opposing counsel or commitments to any of the agencies. There are some basic questions on which agreement must be reached in order to have an effective process:

• Is TAR to be used to increase the cost-efficiency of manual review through effective prioritization?

• Is TAR going to entirely replace manual review to meet an aggressive deadline or significantly reduce costs?

• Will TAR replace manual review of low-risk data (e.g., from non-executive custodians) while being used to prioritize manual review of high-risk data (e.g., from the C-suite)?

Each of these considerations requires a different work-stream approach from the other. Each carries with it a different burden of transparency.

At the simplest level, TAR is a response to the fact that document review focuses increasingly on information originally created with technology. It includes ESI generated by users of smartphones, email, word processing, spreadsheets, databases and the internet, to name a few. In some of its manifestations, technology is so common and familiar to all of us that it’s virtually invisible.

The same also holds true for technologies used to help manage the eDiscovery process. For example, the extremely common practice of de-duping using identical MD5 hashes rests on a remarkable technological feat whose robustness is simply taken for granted. Similarly, web hosting for native review depends on countless technological underpinnings needed just to provide secure and reliable network communication — and so on for many other aspects of just about any review of ESI. Yet without an explainable step-by-step process and audit trail, each of these technologies presents an inherent threat to defensibility.

As these examples suggest, in a broad sense TAR has been faithfully and almost invisibly used since the inception of eDiscovery, and yet many more advanced and transparent TAR techniques, such as predictive coding, have thus far been met with fear and a restrained level of adaptation. State-of-the-art machine learning can yield amazingly high overall accuracy, but the performance of these systems is bounded by a compromise to have either better precision or better recall. For example, tolerances for low recall levels that are welcome and acceptable in the context of recommendation systems, like the book recommendations of Amazon or the film recommendations of Netflix, may not be acceptable in high-stakes legal disputes. A well-designed architecture will allow data mining and other manual adjustments that can be used to break through the bounds where a completely automated system may plateau in order to improve both recall and precision.

No matter the strategy behind using TAR, it is important to have a suite of tested TAR technologies and workflows designed to ensure defensibility. It is also important to know the difference between price and cost. Just as we must move away from outmoded process flows, we need also to view cost metrics differently. The initial investment in TAR will typically be higher, but the downstream upside should be measured not just by unit price, but by overall cost reduction and the defensibility of results.

The views expressed in this column do not necessarily reflect the opinion of Big Law Business or its owners.