For more than 10 years, corporate-generated electronically stored information (ESI) has been growing exponentially. The expense of maintaining these growing data stores often stresses corporate bottom lines, but litigation can bring them to the breaking point. Specifically, the greatest concern comes from the document review phase, which in most cases, can take up to 75 percent of the discovery budget. Worse still, many corporate budgets are still unadjusted to account for this reality, and the time frame for production remains the same. Hence, once litigation begins, it often tests the limits of these budgets.
These realities have brought our system of litigating civil disputes to a tipping point, but a solution is in sight. Two respected judges have endorsed a new spectrum of approaches for reviewing documents, known as technology-assisted Review (TAR). In this article, we will be discussing these approaches and when corporations can best apply them.
Before we delve into the analysis of approaches, let’s first level-set on what TAR actually means and why it is of interest in dealing with these issues. A recent law review article generally described these approaches as follows:
…a process that involves the interplay of humans and computers (meaning various search technologies) to identify documents in a collection that are responsive to a production request, or to identify those documents that should be withheld on the basis of privilege.
TAR improves on linear review by more quickly sifting through large volumes of data and identifying the documents that are potentially responsive, sometimes accelerating this review process by 75 percent or more. This acceleration then manifests in proportional cost savings, particularly when compared to linear review.
TAR can involve a number of methods to yield a defensible production set, but a standard methodology has not yet been established. The encouraging news is that recent court decisions have addressed two effective approaches:
- AI-based methodology leverages computer-provided artificial intelligence (AI) to identify potentially relevant data in a document collection
- Language-based methodology relies on a human’s understanding of language to identify potentially relevant data
Both deliver significant savings in time and cost, but each approach has specific instances in which its use is ideal.
AI-based TAR is generally the best fit, when two elements are in play:
- The need to arrive at quick decisions early in the litigation (assess the case early to decide whether to settle or litigate)
- Enough time remains to read an average of 10,000 documents in order to train the system with a “seed set”
The workflow leverages AI to look for potentially relevant data, meaning that a computer, rather than a reviewer, is performing the lion’s share of the decision making. The reviewer identifies a handful of potentially relevant documents, and the computer takes this input and looks for “more like this” across the current corpus of data to find what it concludes are also responsive documents.
While this process can be quick and relatively painless to counsel at the outset, as we have seen in recent court opinions, this approach can also present challenges, particularly in the crafting and updating of the seed set. Downstream from the review process, we have also seen that, when challenged, explaining how an analytics-based TAR seed set performs can be difficult.
Conversely, language-based TAR relies on a human’s understanding of the actual language in a data set to identify and prioritize documents that are potentially responsive. This results in sets of documents that are prioritized according to relevance (i.e., can’t possibly be relevant, possibly relevant, definitely relevant), and reduces the set of documents that are advanced for review by 50 percent, on average. Those “possibly relevant” documents then move on to a first-pass review phase in which the specific relevant language in each document is highlighted, and then other documents with similar language are tagged accordingly, taking advantage of the redundancy in language from one document to the next. At the same time, reviewers can defensibly set documents that “can’t possibly be relevant” aside with clear transparency as to why they have been excluded, and move those that are “definitely relevant” straight to second-pass, privilege or QC review. This process turns the process of search upside down, looking first for what’s not relevant, which dramatically reduces the time and costs of review when compared to linear review.
Because of this emphasis on human decision-making, language-based TAR is most appealing when transparency and insight into coding decisions are of paramount concern, when the ability to audit reviewers in real time is important (highlighting language enables real-time oversight), and when an organization wants to make this a regular business practice (a corporate dictionary is an output that companies can leverage in future cases).
In conclusion, TAR can deliver significant time and cost savings regardless of which approach an organization takes. If a corporation needs a quick decision on the exposure a case presents, and if there is enough time to read a significant number of documents prior to review, then an AI-based TAR approach may be best. If transparency and insight into coding decisions is important, or the process of a regular business practice is appealing, then language-based TAR is the way to go. In either case, both approaches are court-tested, making them reasonable for use in offsetting the risk and cost of document review.