• Home
  • Analytics & TAR Services

Analytics & TAR Services

Analytics & TAR Services

Analytics and Technology Assisted Review (TAR) are the preeminent factors driving productivity, quality, and cost-savings in litigation today. This is productivity in the Henry Ford sense of the word, deriving more output per labor unit. By combining leading edge applications administered by seasoned Litigation Technologists with the skilled and experienced Document Review Professional, the outcome is one of undeniable efficiency and productivity. Ultimately, the effective use of Analytics will drive the litigation, compliance, and information governance aspects of the Litigation Support Services industry for the next several years, with Innovative Discovery as a recognized leader in the forefront.

Some of our Solutions include:

  • Clustering: An organizational method whereby documents are segregated into mutually exclusive groups, or “clusters,” of conceptually similar documents based on similar text patterns within the documents. Clusters can be created without requiring any user input and can be implemented as soon as the documents are indexed.
  • Categorization: A search and organizational method whereby documents are categorized based on an example document or a set of example documents. This method requires user input in that the user must identify the exemplar document(s) before the categorization can take place.
  • Predictive Coding: A TAR process whereby a universe of documents is categorized based on a “training set” of documents, commonly referred to as the “Seed Set.”

Predictive Coding

A predictive coding workflow can be utilized in several ways, as follows:

1. As a replacement for human review

Predictive Coding as a Replacement for Human Review should be considered in the following situations:

  • Large review universe (e.g., over 100,000 documents).
  • Review universe is not expected to be rich in responsive documents.
  • Aggressive timelines have been established

2. As an organizational tool

Predictive Coding as a Replacement for Human Review may not be a feasible option in the following scenarios:

  • Review universe is too small to justify the up-front investment required
  • The matter is a high profile matter where human review is mandated by the client
  • There is no up-front SME time available
  • The law firm/client is not familiar with predictive coding and is just starting to vet it

Predictive Coding can still be useful as an organizational tool. For example, you can assign the presumptively responsive documents to reviewers with more subject matter expertise (e.g., staff attorneys, lower-level associates), while assigning the presumptively non-responsive documents to reviewers without subject matter expertise (e.g., contract attorneys).

Using predictive coding as an organizational tool also has the following benefits:

  • Can increase review metrics
  • Increases consistency of coding
  • Identifies responsive documents more quickly (especially advantageous for aggressive timelines and rolling productions)

3. As a quality assurance tool

Basic Categorization can be used as an additional quality assurance measure for documents coded as privileged.

Example Workflows Using TAR as an Organizational Tool

EXAMPLE #1: Clustering Followed by Categorization

NOTE: Since this example begins with Clustering – which does not require any user input – this option means the Review Team can start the review as soon as the Review Universe has been loaded and indexed.
  1. Cluster the Review Universe

  2. The Review Universe will be organized into “clusters” of conceptually similar documents.
  3. Assign clusters to the Review Team

  4. You can prioritize the clusters by having the Review Team review clusters containing “hot” search terms first.
  5. During Quality Assurance (“QA”), identify documents that would be good exemplar documents to “back into” a Seed Set.

  6. One of Innovative’s recommended Best Practices is to never waste eyes-on review that happens during QA. In other words, since the QA Reviewer is already assessing a document for QA, they may as well take the extra step to make a determination of whether or not the document itself is a good exemplar document.
  7. Once you have a representative Seed Set identified, categorize the Review Universe into presumptively Responsive and Non-Responsive buckets. Batch the presumptively Responsive documents and their families for review.

  8. NOTE: You can also cluster the presumptively Responsive documents which will further increase review efficiency.

  9. When review of the presumptively Responsive documents and their families has been completed, the Review Team will then move on to the presumptively Non-Responsive documents.

  10. NOTE: As QA continues throughout this process, more documents can be added to improve the Seed Set and the Review Universe can be re-categorized to improve the quality of the presumptively Responsive and Non-Responsive buckets.

EXAMPLE #2: Predictive Coding/Categorization Followed by Clustering

Predictive Coding as a Replacement for Human Review may not be a feasible option in the following scenarios:

  • Review universe is too small to justify the up-front investment required
  • The matter is a high profile matter where human review is mandated by the client
  • There is no up-front SME time available
  • The law firm/client is not familiar with predictive coding and is just starting to vet it

Predictive Coding can still be useful as an organizational tool. For example, you can assign the presumptively responsive documents to reviewers with more subject matter expertise (e.g., staff attorneys, lower-level associates), while assigning the presumptively non-responsive documents to reviewers without subject matter expertise (e.g., contract attorneys).

Using predictive coding as an organizational tool also has the following benefits:

  • Can increase review metrics
  • Increases consistency of coding
  • Identifies responsive documents more quickly (especially advantageous for aggressive timelines and rolling productions)