When is technology-assisted review (predictive coding) defensible to use instead of linear human document review?

Technology-assisted review (TAR), often called predictive coding, uses machine learning to prioritize or classify documents for responsiveness based on human reviewers’ coding decisions on a training set. Rather than having attorneys read every document in order (linear review), TAR lets the system extend human judgment across a large collection. It is part of the broader e-discovery process of identifying and producing relevant electronically stored information.

When TAR Is Generally Defensible

TAR is most defensible when its use is reasonable and proportional to the matter. Key conditions include:

• Large or complex collections. When the volume of electronically stored information makes page-by-page review impractical or disproportionately expensive, TAR is a sound, well-recognized choice.
• A sound, documented process. Defensibility depends less on the tool than on the workflow: how training was conducted, who supervised coding, and how quality was measured.
• Measurable quality. Validation through sampling and metrics such as recall (how much relevant material was found) and precision supports a showing that the process was adequate.
• Cooperation and transparency. Courts and commentators favor parties who discuss methodology, scope, and search approaches with opposing counsel early, consistent with proportionality principles in the Federal Rules of Civil Procedure.

When Linear Review May Be Preferable

• Small collections, where review by hand is faster and cheaper than building and validating a model.
• Highly nuanced issues (for example, privilege calls or subtle factual distinctions) that still require careful human reading, often as a layer on top of TAR.
• Poor-quality inputs, such as image-only files without reliable text extraction, which limit what the algorithm can learn from.

Practical Guidance

There is no universal rule that one method is always required; the governing standard is reasonableness, not perfection. Document your decisions, validate results, and be prepared to explain your methodology if challenged. Requirements also differ by jurisdiction — state courts, federal courts, and other countries apply different rules and expectations — so confirm the standards that apply to your matter.

The Sedona Conference offers widely cited, consensus-based guidance on search, review, and proportionality that practitioners use to support defensible TAR workflows.