Predictive coding is the use of keyword search, filtering and sampling to automate portions of an e-discovery document review. The goal of predictive coding is to reduce the number of irrelevant and non-responsive documents that need to be reviewed manually.
Predictive coding software uses a mathematical model and artificial intelligence programming to scan electronic documents and locate data that is relevant to a legal case. The software, which is capable of learning from its mistakes, first reviews a sample cluster of documents that have been tagged and categorized manually by a human legal team. The predictive coding program is then given a new set of documents and asked to identify which documents are relevant and should be reviewed by humans. The legal team then reviews the software's decisions to determine whether an acceptable level of confidence has been achieved.
Should the software's tagging and categorization fail to demonstrate an acceptable level of confidence, the teaching process is repeated until the software learns what is required. Proponents of predictive coding say it saves time and money because, instead of examining a huge set of electronic files and records manually, the technology allows a much smaller portion of data to be reviewed manually. Because the software speeds up the review process, but still requires human input, predictive coding may also be referred to as technology-assisted review.