Law Technology Today

Optimizing Technology Assisted Review

The age of artificial intelligence is here, as diverse industries advance machine learning to develop self-driving cars, predict patient risk of disease, and even compose poetry. But even as technological innovations surround us, lawyers routinely default to the comforts of “what has been done before” when engaging in litigation discovery.

September 03, 2019

While there are benefits to leveraging past experience, the legal industry also must embrace the dramatic advancements in technology assisted review (“TAR”) applications that can significantly reduce litigation discovery costs, improve document review accuracy, and improve litigation outcomes. As these technologies have evolved, more and more courts are accepting and even encouraging parties to use TAR. Litigators who understand and effectively optimize available TAR applications and protocols can save clients time and money, while simultaneously improving litigation results.

Many litigators fail to optimize TAR’s capabilities because they have a deficient understanding of its capabilities. Litigators must first understand that TAR, itself, is a broad umbrella term that simply refers to any document review that is assisted by technology. TAR can therefore refer to anything from simple keyword searching to document de-duping, document clustering, email threading, and even more sophisticated algorithmic document profiling and predictive coding. While many discovery vendors advertise TAR services, litigators should talk with their vendors and educate themselves about what kinds of specific TAR applications are available, and which applications are best suited to their litigation needs before beginning a review project. Not understanding the available technology upfront can lead to unnecessary litigation expenses down the road, such as having to reprocess large amounts of data in order to apply more sophisticated TAR capabilities beyond those originally contemplated.

Gilbert LLP (“Gilbert”) has built relationships with numerous vendors that specialize in TAR capabilities. These relationships enable our litigators to command the technology and develop workflows and audit processes that help us best serve our clients’ needs. Gilbert litigators have used TAR applications in litigations spanning multiple industries and applied TAR capabilities across numerous discrete projects, resulting in more efficient completion of such projects at significantly reduced cost to our clients without sacrificing quality. For instance, by using computer active learning (“CAL”), we have been able to optimize computing and human resources (from lead litigators to contract reviewers) to create workflows that reduce review time on both offensive and defensive discovery by as much as an estimated 80% as compared to a straight linear manual review.

Utilizing one such workflow, lead attorneys use keyword searching to identify documents that are most likely to be of relevance, code a representative sample of those documents, and feed those documents into a database that builds profiles for “relevant” or “responsive” versus “non-relevant” or “non-responsive” documents. The CAL application then stratifies (potentially millions of) un-reviewed documents based on those profiles, organizing the non-human reviewed documents from most to least likely to be of interest. Junior reviewers can then review and issue-tag those documents that are most likely to be of interest. Designated confidence intervals are set based on the proportional needs of the particular review project in order to assure appropriate accuracy. Under this exemplar, litigation savings are realized because documents that are least likely to be responsive or relevant do not require human review. Additional benefits are also achieved, such as identifying significant documents earlier in the review process.

The CAL application can also continue updating the document profiles based on human reviewer tagging and highlight documents that are tagged inconsistently with the dominant profiles. This feature allows lead attorneys to more easily audit and correct deficiencies in the human review tagging. The benefits of CAL’s stratification of documents based on their relative importance compound at the end of the review: now, lead attorneys have a document set they can review not only by issue, but also by level of importance, increasing the likelihood that key documents are not lost in a digital heap when it comes time to prepare for depositions, motions practice, and trial.

And yet, many lawyers remain reluctant to apply TAR capabilities. This hesitation seems largely based on their own deficiencies and potentially misplaced notions about the superiority of human review. See Youngevity Int’l, Corp. v. Smith, 2019 (“Predictive coding or TAR has emerged as a far more accurate means of producing responsive [electronically-stored information] in discovery than manual human review of keyword searches.”). Where the size of the review justifies the upfront costs associated with more advanced TAR, there is growing consensus that clients may be better served by applying TAR and reducing the costs associated with manual linear reviews. See Zimmer, Inc. v. Beamalloy Reconstructive Med. Prods., LLC, 2019 (disapproving of responding party’s employment of document custodians “to cull documents from their email accounts instead of using a TAR protocol,” and rejecting argument that production of emails would be unduly burdensome because counsel could use TAR to avoid reading “every email for relevance and privilege”). Indeed, it “is now black letter law that where the producing party wants to utilize TAR for document review, courts will permit it.” Entrata, Inc. v. Yardi Sys., Inc., 2018 (quoting Rio Tinto PLC v. Vale S.A., 2015).

Before implementing TAR, counsel needs to be fully apprised of any applicable court rules, including local and chambers rules. For example, some courts have imposed restrictions on the use of TAR, requiring disclosure to adverse parties. In Winfield v. City of New York, 2017 the court noted that courts are split as to the degree of transparency required by the producing party as to its predictive coding process, with some courts ordering the production of the seed set documents and other courts refusing to do so. The court, therefore, urged the parties to cooperate and encouraged transparency.

In evaluating whether and how to integrate TAR applications into a review project, litigators are also well-advised to pay attention to guidelines and best practices that are both currently available and forthcoming. One widely-accepted best practice is to cooperate with the other side regarding the use of TAR, although cooperation is generally not required. See Hyles v. New York City, 2016 (“Cooperation principles, however, do not give the requesting party, or the Court, the power to force cooperation or to force the responding party to use TAR.”) (citing the Sedona Conference Cooperation Proclamation). Parties should also plan to validate documents coded as responsive or non-responsive using TAR, as they are expected to do with a manual review. See In re Broiler Chicken Antitrust Litig., (2018) (setting forth a TAR validation protocol).

Judges, practitioners, and industry experts have also convened through the Bolch Judicial Institute at Duke Law School to offer guidance as to the use of TAR. That convention released “TAR Guidelines” in January 2019, and a draft “TAR Protocol” is currently available for public comment (the final version of the “TAR Protocol” is expected to be released later this summer)

As technology assisted review evolves, so will opportunities and potential pitfalls. While TAR can deliver tremendous benefits, those benefits could be wiped out if not applied appropriately or if parties become embroiled in motions practice. At Gilbert LLP, we will be following legal and technological developments closely in order to continue to deliver the most value and best results possible to our clients.

Gilbert LLP is a Washington-based law firm specializing in litigation and strategic risk management, insurance recovery and complex dispute resolution.