With all the recent focus, it is not surprising that parties to litigation are including technology assisted review (TAR) provisions in protocols governing the production of electronically stored information (ESI). Parties are beginning to allow for at least the possibility of using TAR, even if only to supplement a traditional linear review. These protocols vary in detail, and some require the parties to collaborate together to develop the TAR process. As this technology’s reliability improves, litigants and courts may become more comfortable allowing TAR to have a larger role in identifying responsive documents in litigation.
TAR is a relatively recent development in the legal software industry that is used to prioritize or code documents based on seed sets coded by humans by extrapolating the human judgments to the remaining document collection using sophisticated algorithms or systematic rules. TAR can be used for culling documents or can be used in combination with search terms for prioritizing review. TAR can be used to respond to requests for production in litigation, or in response to government or internal investigations. Despite the fact that TAR can be a useful and effective tool, it may not be a good fit in every litigation or investigation. Practical concerns may weigh in favor of sticking with more traditional methods for culling and reviewing data. This article discusses in general terms some of those concerns, and provides some thoughts to consider when deciding whether to use TAR in your next litigation.
Type of Litigation: Since TAR tools are dependent on document text to identify similar concepts within a document population, it tends to not fare as well in matters that are image-, chart- or diagram-heavy as compared to litigation that is based primarily on more textual documents. Similarly, financial documents, such as spreadsheets or ledgers, also do not typically enhance the TAR process. Therefore, litigation relying on these types of documents (e.g., patent litigation) may not be the best candidates for TAR.
Level of Transparency or Collaboration: As noted by The Sedona Conference, producing parties are best positioned to evaluate the procedures, methodologies, and technologies appropriate for preserving and producing their own ESI. However, some courts and commentators have suggested that increased transparency and collaboration among parties, or even consensus, are required if TAR is to be used in the discovery process. In Moore v. Publicis Groupe, the Southern District of New York advised defendants that if they used TAR methodology, they would be required to turn over their seed set to plaintiffs, including documents tagged as non-responsive in the training rounds. In re Actos (Pioglitazone) Prods. Liab. Litig from the Western District of Louisiana, meanwhile, outlined the “search methodology proof of concept” that required participation by all parties.
In a jurisdiction where there is precedent requiring collaboration and as the number of parties to a matter increase, it may become more difficult to reach a consensus on TAR disclosures, processes, or metrics. This is true not only among adverse parties, but among co-defendants and co-plaintiffs as well, who may have diverging interests, unique IT systems, and business processes. Similarly, if the discovery negotiations are particularly adversarial, TAR may not be the best fit. It may be that efforts expended to achieve alignment among the parties (e.g., motions, hearings, expert testimony) could outweigh any potential benefits of using TAR.
Existing Litigation Readiness: Many defendants are often targeted in litigation. Accordingly, these defendants often invest in litigation readiness platforms and procedures that are typically based on keyword searches. Defendants with these types of systems and processes typically have the reasonable expectation of obtaining the benefit of those investments by deploying these systems and processes and may be reluctant to bypass these internal tools in favor of additional expenditure for TAR. However, there are ways that search term culling and TAR can be used to create efficiencies. For example, TAR could be used to identify certain subjects or issues to help streamline review. Or, the parties may agree that search term tools could be used in concert with TAR to help enrich seed sets to help expedite TAR training.
Expected Volume of Documents in the ESI Collection: Although the technology is improving, TAR still requires significant set up and planning to “train” the system. The commitment and expense of training the system may outweigh the expected savings and efficiencies of using TAR if there is insufficient ESI volume remaining after training.
Expected Richness Rate: Richness is the percentage of responsive documents across the whole ESI collection for a particular matter. It may be worth pulling a random sample to determine the richness of the ESI collection because that could influence the expected burden to train the system. Depending on the system and training methodology, TAR training may require review of random samples of documents pulled from the ESI collection until a statistically significant number of responsive documents are identified. If the ESI collection has a very low responsiveness rate, then more randomly selected documents would need to be reviewed to reach that statistically significant number of responsive documents. On the other hand, if the richness rate is particularly high, then fewer documents would need to be reviewed. The difference in time and expense spent to train the system could be significant.
TAR can be a useful and cost-effective tool, but it is only one arrow in the quiver and should be used in the appropriate situations. When deciding whether to use TAR, a litigant may want to consider the type of litigation they are involved in, the existing litigation processes and systems that may be in place, as well as the characteristics and makeup of the ESI collection itself. Doing so may help shed some light on whether TAR is the right call in your litigation.