Insights Articles

Array recently sponsored a webinar in partnership with the Association of E-Discovery Specialists (ACEDS) on The Role of AI in eDiscovery & Document Review.

The session sparked a wide range of thoughtful audience questions. Below, our speakers address the most common topics raised during the discussion, including several questions we weren't able to answer live. 

Session Panelists:

• Jessica Lockett; Vice President, eDiscovery
• Julia Helmer; Vice President, Advisory & Analytics

Julia Helmer: Over-culling is a legitimate concern and one of the reasons AI tools should be viewed as supporting human decisions rather than a replacement for legal judgment. Additionally, it is important to select purpose-built tools at each step of the case lifecycle. All defensible AI workflows must include validation throughout the process. Before data is excluded, legal teams should test results against representative samples, evaluate recall and precision metrics where appropriate, and conduct targeted quality control reviews, ensuring that testing is done on both the included as well as excluded populations. In the ECA phase, recall should be prioritized over precision. Prompts should be drafted more broadly with the understanding that precision refinement is best handled during the review phase. Additionally, culling can be treated as an iterative process. As the case team learn more about the nuances of the project and data, it is important to leave time and opportunity to “go back to the well”, if needed. Human expertise remains critical for refining criteria, assessing context, and validating outcomes. 

The most effective approach combines purpose-built AI-driven efficiency with iterative testing and human oversight, helping legal teams reduce data volumes while minimizing the risk of excluding potentially relevant information. 

2. How is AI changing the role of legal professionals, and what security considerations should organizations evaluate when adopting AI-powered solutions?  

Jessica Lockett: As AI adoption continues to grow across the legal industry, the role of legal professionals is evolving alongside it. Rather than replacing legal expertise, AI can automate and accelerate certain tasks while placing greater emphasis on strategic workflows, oversight, validation, and defensibility. Human input remains essential for evaluating outputs, managing risk, addressing privilege and confidentiality, and ensuring that technology is used appropriately within the context of a particular matter.

From a security perspective, organizations should apply the same level of diligence to AI-powered solutions that they would to any technology processing sensitive or confidential information. Key considerations include where data is stored and processed, how it is retained and protected, whether customer data is used to train models, who has access to it, and whether the solution meets applicable privacy, security, regulatory, and client requirements.

Not all AI solutions are designed in the same way or provide the same protections. Legal teams should understand not only what a solution can do, but also how their information is handled and what safeguards are in place, with the level of diligence reflecting the sensitivity of the data and the nature of the proposed use. 

3. Can exclusion criteria be incorporated into AI-assisted review workflows to identify non-responsive or irrelevant documents? 

Julia Helmer: Absolutely. In most cases, exclusion criteria can be just as useful as inclusion criteria. 

For example, teams may define categories of documents that are clearly non-responsive, administrative, personal, duplicative, or otherwise outside the scope of the matter. Providing both inclusion and exclusion criteria often helps AI tools make more consistent decisions and can improve review prioritization. 

As with any AI-assisted workflow, all prompts should be validated through sampling and quality control before substantive decisions are made. The best results typically come from combining clear inclusion criteria, exclusion criteria, detail and examples via natural language. 

4. How does Array's technology support a hybrid approach to eDiscovery and document review? 

Jessica Lockett: Array's approach combines advanced eDiscovery technology, AI-enabled workflows, analytics, and human expertise rather than relying on any single methodology or technology.

Depending on the matter, we may leverage technologies such as advanced analytics, Technology-Assisted Review (TAR) including Continuous Active Learning (CAL) and generative AI (GenAI), as well as automation alongside experienced project managers, review lawyers, and subject matter experts. The appropriate approach depends on the objectives of the matter, the characteristics of the data, the associated risks, and the needs of the legal team.

Technology can help accelerate workflows, prioritize information, reduce the volume of material requiring manual review, and improve consistency. At the same time, human expertise remains central to workflow design, validation, strategic decision-making, privilege considerations, and quality assurance.  

Ultimately, there is no one-size-fits-all approach to ediscovery. The most effective workflows match the technology to the particular review problem and combine the speed and scale of technology with the input and expertise of legal and review professionals, improving efficiency while maintaining the oversight and quality control necessary for defensible outcomes. 

Missed the webinar? View the full on-demand recording to hear the full discussion and key takeways from our experts.