Aid4Mail AI Classification Benchmark Report

This benchmark report evaluates 11 AI models, including 5 commercial cloud models and 6 offline models, focusing on their performance in digital forensics and eDiscovery email classification tasks. The report outlines the methodology used in the evaluation, which consists of five tests and a large-scale Production Pilot. The tests assess various metrics such as classification accuracy, multi-category discrimination, multilingual performance, and operational throughput. The document aims to assist Aid4Mail users in understanding the performance of different models on realistic tasks, comparing AI classification with traditional keyword search and Technology-Assisted Review (TAR), and selecting the most suitable model based on specific workflow requirements. Additionally, the report provides methodological caveats and emphasizes that the performance of classifiers can vary based on several factors, including the nature of the corpus and the model's suitability for the task. The findings are supported by detailed results and analyses of individual models.