Insights Articles

Our experts recently hosted Array's Review Intelligence Masterclass, Accelerating Review Without Sacrificing Defensibility: What Works in Practice.

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:

• Matt Buser; Senior Director, Managed Review
• Jeff Grobart; Associate Director, Client Solutions
• Jessica Lockett; Vice President, eDiscovery
• Umer Moolla; Director; Review Services 

Matt Buser: One key insight participants often overlook is that the defensibility of AI-assisted review comes from the validation process, not the technology itself.

At scale, the workflow is systematically tested through structured sampling and quality checks, including statistically valid samples from both responsive and excluded populations.

The goal is not to assume the technology is correct, but to prove it through validation, so teams can confirm AI is performing as expected and that responsive documents are not being missed.

2. Does your AI recognized mixed language data, and can it translate it into English?

Jeff Grobart: Depending on the solution, many AI tools can recognize and process multiple languages, and prompts in the source language often improve results.

Where language expertise is limited, translated text can help standardize the dataset, similar to transcription for audio or video.

In practice, this is not a one-step process and typically involves:

• Identifying non-English data
• Translating relevant content
• Indexing translations into the workflow
• Applying AI classification to the combined dataset
•  

3. What makes Array stand out from other providers?

Jessica Lockett: What differentiates Array is our people and how we show up for clients and each other.

Our teams bring deep expertise in eDiscovery technology and processes, but also take time to understand each client's matter, risk profile, and priorities, leading to strong, collaborative relationships.

We operate as an extension of the client team, with a high level of accountability, responsiveness, and trust. Internally, our cross-functional approach combining legal, technical, and operational expertise and collaboration allows us to solve problems efficiently and continuously improve our processes.

4. What safeguards ensure defensibility as AI adoption increases?

Matt Buser: To maintain defensibility, the focus is on building a process that is transparent, measurable, and repeatable.

Four safeguards are critical: 

• Documented Protocols: Clearly define how AI is used, including decision logic and review criteria.
• Statistical Validation: Use sampling to confirm responsive documents are captured and excluded populations are appropriate.
• Human Oversight: Maintain active legal involvement in training, calibration, and edge cases.
• Auditability: Preserve decision logs, model iterations, and metrics so workflows can be reconstructed.

Ultimately, defensibility is about demonstrating that the process was reasonable, consistent, and validated, not relying on the tool alone.

5. Even well-drafted prompts can produce different results. How do you handle that?

Jeff Grobart: Running the same prompt more than once can product slight variations in output, and this is expected.

In practice, these differences are typically minor, which is why we rely on statistical sampling and validation to confirm results fall within an expected range.

Where models are updated, workflows are re-tested and validated prior to implementation, particularly in long-running matters or where data has evolved.

6. If AI does the heavy lifting, what does the reviewer process look like?

Matt Buser: At scale, where AI performs initial classification, the role of reviewers shifts to training, validation, and second-level review.

Senior reviewers support prompt development or model training using sample documents, ensuring alignment with case-specific criteria.

As the workflow progresses, reviewers validate outputs through statistically meaningful sampling, confirming accuracy across both responsive and non-responsive populations.

Once validated, AI separates the dataset, and reviewers focus on issue coding, privilege analysis, redactions, and legal decision-making.

In most cases, AI handles the majority of classification, with human reviewers focused on oversight and substantive analysis.

7. What platforms support Array AI Review?

Jeff Grobart: Array's Managed Document Review team is platform-agnostic, though most tools are optimized for Relativity.

We also support AI and TAR workflows in Everlaw, DISCO, eDiscovery AI, and Casepoint, depending on the environment and permissions available.

8. How do you address concerns that GenAI feels like a "black box"?

Jessica Lockett: Concerns about GenAI acting as a "black box" typically relate to lack of visibility into the workflow, not the technology itself.

At Array, GenAI is embedded within a structured, fully documented process where prompts, inputs, and outputs are transparent and traceable, allowing every step to be revisited.

We apply the same defensibility standards as traditional review, including rigorous testing, statistical sampling, and human-in-the-loop validation, ensuring the process remains clear, measurable, and defensible.

9. Are lawyers allowed to use AI without disclosure to opposing counsel?

Matt Buser: Legal decision-making remains with legal teams, who provide instruction and guidance on review decisions and validation of outputs in both CAL and generative workflows.

The use of AI or TAR does not automatically trigger a disclosure obligation. Courts focus on whether the process is reasonable, proportionate, and well-documented.

While methodologies may be discussed if challenged, there is generally no requirement to disclosure or seek approval solely for using AI-assisted review, though local rules should always be considered.

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