Classification

Label legal documents zero-shot, with no training data

Overview

Classification

Universal classification, or zero-shot classification, is the process of determining whether a statement expressed about a document (e.g. “this is a confidentiality clause”) is supported by that document. Unlike traditional classifiers, universal classifiers do not require examples beforehand. Simply write out your classification criteria.

Use cases

  • Zero-shot labeling

    Determine whether a statement about a document is supported by that document with no training data, finetuning, or bespoke taxonomy. Use natural language.

  • Clause and provision detection

    Pull particular types of clauses from contracts, flag privileged material, or spot force majeure provisions.

  • Retrieval and ranking

    Score search relevance, pre-filter documents for downstream models, or evaluate statements across thousands of documents in seconds.

  • Benchmark-leading accuracy

    The world’s most accurate universal legal classifiers of their size.

    Despite their compact size, Kanon Universal Classifiers punch far above their weight — up to 12% better than their closest general-purpose counterparts.

    They outperform every universal classifier benchmarked, including DeBERTa v3 large, despite being 17% smaller.

  • Zero-shot at scale

    Write your classification criteria in natural language without examples or finetuning.

    Evaluate a statement against thousands of documents in seconds and get accurate confidence scores.

    Teams can iterate on labels as quickly as they iterate on product ideas — without collecting examples or maintaining a separate model for every workflow.

  • Purpose-built over generic

    Legal AI models trained on legal data outperform generic text models at operations on legal documents.

    Legal teams classify constantly: termination rights, privileged material, factual propositions. Classification turns those decisions into scalable product primitives.

    Use confidence scores to build review tools, triage systems, retrieval filters, and workflows that route, rank, and act on documents automatically.