In e-discovery, teams make decisions based on their unique needs and preferences. There is no one-size-fits-all approach for analytics. Similarly, a heavily frosted white cake might be a classic wedding fixture, but plenty of brides and grooms opt for donut towers or cheesecake in lieu of tradition.
Inspired to tailor their use of Relativity Analytics in a way that satisfied their specific appetites, JND built an application on top of the platform that echoes the layered approach to a wedding cake.
Check out the origin story of JND's innovation, Layer Cake, for inspiration on making something deliciously efficient for your team.
The Case Prompting Innovation
When Lowey Dannenberg Cohen & Hart, P.C. needed to organize data for a global current manipulation matter, they sought the expertise of JND to make it happen. This was a hefty undertaking: the case encompassed 2.2 million documents and nearly 30 defendants, covering many data sources.
“When you get to a case this large, it is too time-consuming to review every single page of every document,” said Ben Sexton, director of e-discovery at JND. “Our goal was to get as many useful documents into the review as possible.”
First, JND familiarized themselves with their client’s data so they could adequately assess which features in Relativity would drive an efficient project. The team had specific goals in mind before jumping into the review stage.
“We wanted to combine all the key Relativity features in a cohesive approach to get measurable results,” said Ben. “What was very important to us was getting the work done behind the curtain. We didn’t want to train everyone at the firm on how to be an analytics expert; we wanted to let attorneys be attorneys.”
Catering the Recipe to the Data
After digging into their client’s data, JND realized they would need to separate the document set into two pools: communications, such as emails and chats, and transactional data, such as trade confirmations. The decision to filter into those buckets was driven by the substantial variances in the data types and the types of evidence within the documents.
For the communications bucket, JND first coded several “hot documents” as seed documents to train Relativity Assisted Review on how to categorize the data. Then, JND would sample the results from Assisted Review to ensure accuracy. Once JND was comfortable with the recall rates, they had to categorize the documents.
JND considered two core factors when categorizing the document pool: a document’s conceptual similarity to the seed documents, and whether the document hit on key people, timeframes, and terms.
JND’s categorization system is comprised of seven tiers, with the first tier containing the documents that rank highest in terms of relevance.
“The more overlap between the buckets, the more likely that a document will be key to the case,” Ben explained. “Tiering the documents allowed us to merge these conditions into a seamless review.”
When it came time to categorize the transactional data, JND had to adjust their recipe—it was not as cut and dry as communications between different parties. Analyzing the transactional data involved going through the various emails and other forms of communications and finding instances were transactions were described.
To identify the seed documents, JND ran a keyword search, clustering, and regulator expression testing. Once they fed the seed documents into Assisted Review, the recipe was similar to that of the communications set, except the team was working with different key people, terms, and time frames.
After categorizing the documents for both buckets, JND wanted to facilitate a seamless process for the firm's reviewers. So, they eliminated 1.2 million documents from the review set and batched 65,000 documents according to the tiers. JND found the documents promoted by Layer Cake were 100 times more responsive by the time they made it to those reviewers.
This richness yielded major savings for Lowey Dannenberg Cohen & Hart; they saved more than 50,000 review hours and an estimated $2.5 million dollars over a traditional, linear review.
Making a Winning Workflow Repeatable
These results were big—for their client and JND. They soon realized they could turn this workflow into an application they could apply to many matters, and decided to build the application in RelativityOne.
"When we moved to RelativityOne, this gave us more flexibility to customize and build unique workflows," said Ben. With the platform, JND automated the manual processes that went into creating the tiered system, layer by layer.
“Layer Cake exemplifies our commitment to innovation on the platform and building customized workflows for our clients that provide value,” Ben said.
