Predictive Coding Without the Workflow

At LightSpeed, we are big fans of predictive coding* for its ability to jump start document assessment and enhance – or even replace – traditional reviews. However, many of our clients are not ready to take the leap. For those who want the benefits of predictive coding with the option to follow their own workflow, we offer Presumptive Designation.

Through LightSpeed’s passive/active learning process, Presumptive Designation predicts responsiveness, privilege and spam for all documents in your collection. We describe it as passive/active because there are no required user steps. Users simply designate documents as usual while Presumptive Designation actively reads and analyzes the documents to determine what conceptual patterns reliably indicate whether a document is privileged, responsive or spam. These “presumptive designations” are placed in a separate field and actively updated as more documents are designated by users.

So, what can we do when we presume?

  • Tally searches to group your results into likely responsive, non-responsive, privileged or unknown.
  • Create review batches so likely responsive documents are reviewed first. Likely privileged documents can be batched separately.
  • View a summary of related presumptive designations each time media is added to our platform.
  • Compare user designations to presumptive designations as a quality control step. [See diagram]
  • Compare the percentage of deviation between each user and presumptive designations to quickly highlight users who are inconsistent with the group and the related inconsistent documents.
  • Identify responsive documents that have not been promoted for review.
  • Cull documents faster with more accuracy and confidence.

Presumptive Designations is included with all LightSpeed cases hosted in Eclipse in 2017. If you have any questions about how to use this feature, please contact your project management team. If you are curious about this “baby step” to predictive coding and would like to learn more, please contact us.

*AKA Technology Assisted Review (TAR) or Technology Assisted Learning. Predictive coding is a machine learning process that uses software to take logic entered by people, for the purpose of finding responsive documents, and applies it to much larger datasets to reduce the number of irrelevant and non-responsive documents that need to be reviewed manually.

This example highlights 10 documents that users coded as responsive, but Presumptive Designation has flagged them as privileged.