Products
Use cases
Industries
Resources
Company
New Study from Reveal’s Onna Finds Collaboration Data Drains 26 Hours Per Matter as 80% of Organizations Face Cost Overruns.
Measuring and Validating the Effectiveness of Reveal's GenAI Review System
An independent empirical evaluation of aji, Reveal's GenAI-powered document review engine, conducted by Bennett B. Borden, Founder and CEO of Clarion AI Partners and former Chief Data Scientist at DLA Piper.
Download the Evaluation
Why this study matters
GenAI has arrived in legal document review, but adoption requires more than excitement. Every law firm deploying AI-assisted review must answer three questions before committing: Is the process accurate? Is it reliable at scale? Is it explainable and defensible? This independent evaluation puts aji to the test.
1
Independently validated accuracy - Conducted by Bennett B. Borden, a 25-year litigator and one of the world's first lawyer/data scientists, with full control over methodology, analysis, and conclusions.
2
Tested on real matter data at scale - The study ran against 33,134 documents from an actual matter, not a synthetic dataset.
3
End-to-end process coverage - the evaluation examined every stage of the aji workflow, from initial Definition creation through full-scale GenAI review, not just the output
96.95%
Human-AI agreement rate on 380-document validation sample
33K+
Documents processed in under 12 hours of machine time
15 min
To create and refine an initial Definition with the AI Advisor
What the evaluation covers
A definition, calibration, and validation workflow
An end-to-end walkthrough of aji's five-stage process (Define, Calibrate, Validate, Run, Utilize) and how the AI Definition Advisor helps lawyers sharpen relevance criteria before touching a single document. The study ran four calibration cycles and achieved an average agreement rate of 91.60% before a final validation of 96.95%.
Accuracy, reliability, and explainability results
A rigorous analysis of true positives, true negatives, false positives, and false negatives across calibration and validation stages. The evaluator scrutinized not just ratings but aji's natural-language reasons and in-text citations, the same transparency that makes review defensible when challenged.
What this means for legal teams in practice
Beyond the numbers, the evaluation addresses how aji changes the work lawyers actually do, from faster time-to-accuracy and early actionable insights during Definition development to an audit trail that can withstand scrutiny from any stakeholder. The evaluator's conclusion: aji provides a repeatable, testable, well-defined workflow ready for use in actual matters.
Download the Evaluation
"aji provides a repeatable, testable, well-defined workflow ready for use in actual matters. It enables users to quickly and cost-effectively conduct document reviews and deliver high-quality results.”
Get the independent evidence your evaluation requires
The complete empirical evaluation for legal technology leaders, eDiscovery professionals, and law firm partners assessing whether Reveal's aji meets the bar for production use in active matters. This whitepaper covers:
Three-question evaluation framework covering accuracy of the calibration process, reliability of ratings at scale, and transparency of outputs
Step-by-step testing methodology across all five aji stages with real matter data
Full statistical results including agreement rates, precision and recall metrics, and true/false positive breakdown
Evaluator analysis of aji's reasons, inline citations, and audit reporting capabilities
Practical guidance on what defensible AI review looks like and how aji delivers it
Trusted by thousands of modern legal teams globally
%20(1).png)
%201.svg)
.svg)