Leading Law Firm

Each high-risk document was coded according to privilege level, training a document model to (1) find more similar content and (2) check the quality of privilege decisions across the production. Using a combination of analytics and Predictive Coding, the Sidley Austin team satisfied the government subpoena while staying within a budget that would have been impossible with linear review methods.

Team size and project constraints

In this unusual case, Sidley Austin was retained to respond to broad government subpoenas across 125 GBs of mailbox data. The charity organization did not have an extensive internal eDiscovery capacity and relied on the strategic eDiscovery team. And since every dollar spent on the subpoena response was a dollar that couldn’t be used for providing charitable services, the team handling the matter had to be lean—in total there were about six attorneys—and had to rapidly identify the riskiest documents for priority review. 

Efficient workflow with analytics and machine learning.

The Sidley Austin team prioritized documents using Axcelerate and a workflow that emphasized analytics and machine learning. The firm chose OpenText™ Axcelerate™ OnDemand, a managed service that makes Axcelerate available for individual matters and cases, supported by OpenText Discovery project managers. Bolstering the team with a dedicated PM to help devise and run search term queries, productions, and samples allowed the case team to focus on applying the nuanced legal judgments required for responsiveness and privilege calls. 

To start, the team narrowed the data using metadata filters. Knowing that certain date ranges were outside the scope of the subpoena, they rapidly filtered the dataset according to document creation date, sent date and other time-based filters. Realizing that certain custodians and communications would trigger privilege, they further culled by domain information. Throughout the process, Axcelerate’s visualized metadata fields displayed how many documents featured the metadata they were seeking. To keep costs down, sampling was a must.

As the team interrogated their client’s data, they experimented with keywords and ran samples out of the larger responsive sets to determine whether the keywords were targeted or burdensome. They were able to rapidly evaluate huge datasets to determine whether they were valuable for review. Samples that returned rich results were fasttracked into a review workflow, which was aided by Predictive Coding. The Predictive Coding algorithm created a data model that suggested similar documents for review. And, since Axcelerate Predictive Coding is issue-based, they were able to simultaneously train the system to quality check their privilege decisions. Decisions that the computer disagreed with could be quickly identified for second pass review.