Delaware Court of Chancery Supports Use of Technology Assisted Review to Reduce the Burden of Document Reviews

On October 21, 2024, the Delaware Court of Chancery ruled on a motion to compel in Berger v. Graf Acquisition, LLC, et al. regarding the use of technology assisted review to address the burdens of substantial document review in discovery.

Berger is a putative class action brought by a former stockholder of Graf Industrial. The former stockholder asserted that Graf impaired his redemption right in connection with a September 2020 merger. The case is still in discovery, and though some discovery has been provided, the parties were at an impasse on whether more is warranted.

The defendants agreed that the documents sought by the plaintiff in discovery are relevant, but insisted that the plaintiff’s proposed search criteria, which would require the review of 125,000 documents, was burdensome and disproportionate to the needs of the case. As such, the defendants countered by proposing much narrower search terms or offering to review the universe of documents captured by the plaintiff’s broad search terms so long as they could employ a technology assisted review protocol. The plaintiff rejected this offer and insisted that the defendants conduct a manual review instead.

Ultimately, the court agreed with the defendants and concluded that utilizing technology assisted review protocol is “a reasonable means to address … burden concerns” inherent to the eDiscovery process.

What is TAR

Technology Assisted Review (TAR) is a form of artificial intelligence that is employed in eDiscovery. TAR leverages machine learning and/or continuous active learning techniques with the guidance of attorneys to facilitate the document review process.

The goal when using TAR is to define and prioritize the universe of documents that are likely to be responsive and exclude or not prioritize documents that are not likely to be responsive.

As noted by the Delaware Court of Chancery:

“TAR promotes efficiency in the discovery process in several meaningful ways. First, with appropriate quality control and system training, TAR reduces the document set for human review and limits it to documents more likely to be responsive. Second, TAR processes apply a more uniform relevance standard than manual document review using keyword searches. Third, and relatedly, a more manageable document set for manual review means lower costs for litigants. TAR can also yield superior results.”

How does TAR work?

The first step in using TAR is to create a set of documents that can be used to train your TAR model. There are a few different methods to train a TAR model, but typically this process involves a senior attorney reviewing and tagging a random sample of documents (typically, 500 documents or more) from the review set. This random sample is called your control set. Once the control set is ready, the TAR model uses the attorney’s tagging decisions to build a classification/ranking algorithm that will rank the rest of the documents in the review set.

After the TAR model is run across the entire review set, the senior attorney should review another random sampling to see if the rankings are appropriate. If the rankings are not appropriate, the attorney may need to further train the model (i.e., enhance the control set by tagging more documents).

Once the attorney is comfortable with the TAR model’s rankings, the review set can be ranked from most to least likely to be responsive. The documents scored most likely to be responsive should be reviewed first, and in some instances, the documents below a certain responsiveness score can be discarded.

Best Practices When Using TAR

• It is for the producing party to decide whether to utilize TAR, as that party has the best knowledge of the specifics of its own electronically stored information; however, as noted by the Delaware Chancery Court, the producing party should be transparent about the TAR process. We recommend keeping a log of the who, what, where and when of your TAR process.
• The initial sample review should be as accurate as possible at that stage of the discovery process. We recommend a senior attorney with substantial knowledge of the case review and tag the sample used to train the TAR model.
• Second-level and quality control review should be utilized throughout the TAR process to ensure that the parameters set up by the control set remain accurate in light of additional information uncovered during the review process. Discovery and document review are fluid processes. If a new fact or issue comes to light, you should re-train your model accordingly.

Artificial Intelligence is being deployed in eDiscovery more than ever before. TAR (both TAR 1.0 and 2.0) and Gen AI are powerful tools in the eDiscovery tool kit, but that does not mean they are fit for every case or document review.

Before using TAR or other artificial intelligence technology, attorneys should consult their firm’s or company’s artificial intelligence policy and/or consult someone familiar with legal artificial intelligence tools.

Co-authors Chris Orrin, eDiscovery counsel, and Yordan Karov, associate, are eDiscovery attorneys in the Litigation Department at Klehr Harrison.

Artificial intelligence and e-Discovery will be discussed during the upcoming Klehr Harrison Lunch and Learn program, “Leveraging AI in eDiscovery: Innovations, Challenges and Best Practices”. The program will be presented by eDiscovery Counsel Chris Orrin. Registration is still open.