How to Manage Privilege When Producing Antitrust Documents to Government | Epic

Identifying privileged content remains one of the most time-consuming undertakings of any discovery project. In recent years there have been significant developments in the ability to use analytic tools and artificial intelligence (AI) to identify the existence of privilege. This helps legal teams streamline the review process while preserving privileges.

Below is an overview of the steps to consider in managing the identification and review of potentially privileged content in anti-trust issues. Using advanced tools and following the proper steps can save a lot of time when producing documents for the government in an antitrust action.

Step One: Compiling a Privileges Screen

The legal team (corporate and outside attorneys) and the data provider should work closely together to compile a “privilege screen” of attorney names and terms that may signify the existence of privileged data. The privilege search terms report will display the total and unique number of “hits” for each name and term. Once the screen is finalized, it is executed on any reactive data. As with any search term analysis, it is important to test the results of these search terms, especially terms that might have wider use and may generate false positives.

Step Two: Highlight Terms That Indicate a Privilege

All hits are then highlighted in the review database. The team can also deploy custom analysis workflows to identify documents and document families that only have privilege terms appearing in email footers, as it is often common practice to include the “privileged and confidential” disclaimer on all emails.

Third step: Deploy an AI model

To reduce the number of false hits and broaden the scope to include privileged documents that privileged terms might miss, attorneys can also use analytics and AI tools to supplement the standard privilege screen. It starts with feeding sample coded documents into an AI system. These systems typically use the documents’ text and metadata to score each on a scale of 0 to 100 – the higher the score, the more likely the document is to be favored. Scores provide another way to identify privileged people documents and help reduce the number of false results returned by the standard privilege screen.

A privileged expert (usually from the review provider) can then train the system using examples from the collection. The training effort is typically less than 3,000 documents and continues while the system derives reasonable value from additional training. The expert also uses social network analysis, domain analysis and other analytical tools to identify potential privilege actors. Statistical sampling and targeted searches of the null set are then used to validate the results of the favored AI model.

Fourth step: sort the documents

After deploying the AI ​​model, the documents identified as potentially privileged are sorted into two groups:

  1. Higher Probability Documents are those containing certain attributes that generally signal the existence of privileged content. Common examples include external communications Jural advisor for the end customer and communications with internal counsel. These documents can be moved directly into a privilege log review and redaction workflow, eliminating the need for a first-level privilege review. The examiner performing the privilege log review will then confirm that the privilege exists.
  1. Low Probability Documents are those that exhibit some indicia of privileged content, but still require validation through the first-level review process to confirm the existence of the privilege. External counsel will decide whether to review all documents flagged as potentially privileged or to halt review for those with lower probability ratings.

Step Five: Identify Sources of Additional Privileges

Once the review begins, the team will identify additional names of individuals or organizations who may create or break the privilege. New preferred names are forwarded to outside counsel to determine if they should be added to a preferred search term report. Any new names added to the Preferred Search Terms report will then be normalized into previously reviewed non-preferred documents.


Combining traditional, rigorous filtering with AI privilege tools will provide the best results when examining privilege in anti-trust issues. Moreover, the process involved is defensible. The lawyer can easily explain the above process to a regulator who questions the methodology. Coupled with the protection of a protective order regarding the inadvertent production of privileged material, this provides a complete workflow to properly identify and record privileged data.

This blog post is derived from the chapter titled “Outsourced Document Review: Data Intelligence, Technologist Lawyers, Advocacy Support” by Edward Burke and Allison Dunham, which appears in the treatise by Thomson Reuters eDiscovery for Corporate Counsel (2022). Reproduced with permission, © 2022, Thomson Reuters. Jason Butler also contributed to this blog.

A link to the book appears below:

[View source.]

Comments are closed.