Technology-assisted review applies supervised machine learning to document review. Attorneys code a set of documents, the model learns from those decisions, and it then ranks or classifies the remaining population by likely responsiveness. Modern workflows use continuous active learning, in which the model retrains constantly and continually surfaces the most likely responsive documents.
Courts have accepted the methodology since Da Silva Moore in 2012, and it is now routine in large matters. The defensibility question has shifted from whether TAR may be used to whether the specific protocol, training, and validation were reasonable.
Alternative Names:
Predictive Coding, Computer-Assisted Review, TAR
Why it Matters?
Document review is the single largest cost in complex discovery. TAR can reduce reviewable volume by a large margin while improving consistency over linear review. In matters with millions of documents, it is often the only proportional approach, and negotiating the protocol early prevents disputes over recall targets and validation later.
Frequently Confused with
Related terms
Frequently asked questions
Is technology-assisted review defensible in court?
Does TAR eliminate attorney review?
