Predictive Coding | Practical Law

Predictive Coding | Practical Law

Predictive Coding

Predictive Coding

Practical Law Glossary Item 8-520-0523 (Approx. 3 pages)

Glossary

Predictive Coding

Predictive coding is a type of technology-assisted review (TAR) and an alternative to document-by-document manual review of electronically stored information (ESI). When properly leveraged, predictive coding can drastically lower the costs of electronic discovery by reducing the amount of time attorneys spend manually reviewing documents.
Predictive coding tools vary, but the process is generally as follows:
  • One or more attorneys with broad knowledge of the case review and code a subset (seed set) of ESI as either relevant or irrelevant.
  • Using the seed set coding as guidance, the predictive coding software reviews and codes documents for relevance in a manner that is consistent with counsel's seed set.
  • Counsel reviews the software's automated coding and makes revisions as appropriate.
  • The software incorporates counsel's feedback and repeats its review and coding.
The above process continues until counsel is confident that the software's relevance coding is consistent with counsel's judgment. Before making such a determination and allowing the software's coding predictions to stand, counsel should:
  • Thoroughly sample the software's coding, to ensure that:
    • the ESI coded as relevant does not contain an unreasonable number of irrelevant documents; and
    • the ESI codes as irrelevant does not contain an unreasonable number of relevant documents.
  • Document their sampling efforts and any other quality checks they perform.
Once counsel is confident in the software's relevance coding, counsel moves on to the next step of the document review process, which may involve reviewing all ESI coded as relevant to:
  • Confirm relevance.
  • Identify the discovery request to which it is responsive, if any.
  • Locate and withhold (or redact) privileged or protected content.
For more information about technology-assisted review, see Practice Note, Continuous Active Learning for TAR.