What Is Predictive Coding, and How Does It Apply to Ediscovery?

Before electronic discovery (ediscovery), humans were required to manually review every document requested and provided during the discovery process. However, when modern data formats and technology made the volume of discoverable information explode, manual review became impractical. Predictive coding is now transforming ediscovery, using artificial intelligence (AI) to expedite the review process, saving firms both time and money.

What Is Predictive Coding?

Predictive coding leverages machine learning to surface likely relevant documents based off prior review decisions. Although machine learning may sound mysterious, it is actually a part of everyday life for most people. Consider the following examples:

  • Netflix and Amazon give movie and television recommendations based upon prior viewing choices.
  • Google search results are based on an algorithm that learns from past search and internet browsing behavior.
  • Spam filters allow us to divert dozens of unwanted emails from our inboxes.

The Role of Predictive Coding (and Human Reviewers) in Ediscovery

Within the realm of ediscovery, predictive coding—also known as technology-assisted review (TAR)—automates document review. A type of machine-learning technology, predictive coding gives computers the ability to learn from human input and make educated guesses as to how documents should be classified. Predictive coding is widely used to identify responsive or relevant documents and data in other review categories, including privilege, issue codes, and other vital classifications to dramatically reduce document sets—sometimes from several million down to a few thousand.

Predictive coding technology is used to find responsive electronically stored information (ESI) during the review phase of a case, shaping and altering the discovery process for law firms, lawyers, clients, and the courts. Here’s how it works:

  • The software is “trained” with a seed set of data—a sample of documents retrieved from the larger group of documents needing review.
  • Reviewers code each document as relevant or not relevant to the case and input this information into the predictive coding software.
  • As document review continues, artificial intelligence enables the software to continually learn from reviewer decisions and make faster and more accurate decisions.

The predictions generated by predictive coding can be extremely powerful. They can uncover a “smoking gun” document that will serve as proof in a criminal case, avoid an extensive manual review of documents to save thousands or even millions of dollars in review time, prioritize documents for relevance review to make the process increasingly efficient, and be used for quality control, making the review process significantly more accurate. Given these benefits, it is not surprising that according to Norton Rose Fulbright’s 2017 Litigation Trends Annual Survey, over 60 percent of corporate counsel are using predictive coding in their cases, and courts are routinely approving use of the technology in the cases before them.

Predictive Coding in the Everlaw System

Predictive coding with Everlaw learns continuously from reviewer decisions, enabling reviewers to use their standard workflow—assigning ratings, codes, and attributes to reviewed documents—in order to teach the system how to find more relevant documents on their behalf. For more information on how predictive coding works within the Everlaw platform, download our white paper “Tactical Review with Predictive Coding” today.