In a merger between two healthcare industry companies, the Federal Trade Commission (FTC) issued a quick look as part of the second request process. Working under a compressed timeline, outside counsel engaged FTI Consulting to quickly structure a review and stabilize the Brainspace Technology Assisted Review (TAR) model so the parties could determine the risk level of overproducing and prepare the dataset for production to the FTC. A first set of documents had been submitted to regulators, with a second production due just as the COVID-19 pandemic reached the U.S. and self-isolation orders were being implemented.
In quick look matters, wherein the merging companies agree to produce a limited set of core documents to the FTC, data volumes are typically more manageable than extensive second request investigations. Still, the collection exceeded 3.9 million documents, and counsel was dealing with several complicating factors. In addition to a tight timeline of three weeks to complete the review, there were problems in getting access to the client’s data. The client had self-collected some documents, used a third party for collecting mobile data and identified dozens of boxes of paper documents at the last minute—all leading to delays in transferring the dataset to FTI Consulting’s environment.
These uncertainties, combined with the squeeze on time, meant that FTI Consulting’s team would need to develop a unique workflow that could meet the regulator’s parameters, quickly stabilize the Supervised Learning model and get through the documents at a fast pace.
As all of these factors came to a head, the COVID-19 pandemic was hitting the U.S., and states were beginning to issue stay-athome orders. FTI Consulting was committed to completing the review before review centers were closed, to avoid unexpected and costly delays for the client’s production to the FTC.
While coordinating with the client and its third-party collection vendor to gain access to the full set of 3.9 million electronic and paper documents, FTI Consulting set to work on preparing the review. Using Brainspace’s Continuous Multimodal Learning (CMML), a flexible interactive TAR tool, FTI Consulting customized an innovative workflow around this capability. This allowed the team to begin training the analytics model before all the data was received.
Reviewers were divided into three teams. One team of five reviewers (subject matter experts) was given sample sets of documents for which they determined, based on predetermined criteria, the relevancy of each document. Decisions were continuously used to train the CMML model until it reached a sufficient level of accuracy and recall based on an independent statistically significant validation set.
While the model was being trained, a second team of reviewers was provided with documents that could not go through the prediction process, including images, multimedia files and scanned paper documents from which text could not be easily pulled. The team reviewed these documents individually for responsiveness and privilege.
A third team of specialized reviewers worked on a highly focused review of the predicted responsive documents. FTI Consulting implemented a smart analytics approach to produce non-privileged documents from the predicted responsive population. The steps included ongoing keyword refinement, domain analysis, name count analysis, concept clustering, and leveraging all available metadata to determine what could be safely produced and what needed more intensive review.