Unlocking the Potential of Generative AI in eDiscovery: A Conversation with Fernando Delgado

Summary: Two AI leaders at Lighthouse discuss how to evaluate GenAI tools and ROI to ensure you are getting the most out of the technology’s potential for your needs.

‍

Generative AI promises to improve eDiscovery in multiple ways, from increasing efficiency to enhancing case planning. But how do you know a generative AI tool is able to deliver on its promise? How do you know it’s a tool or partner you can trust?

Two AI leaders at Lighthouse recently sat down to discuss questions like these and explain how Lighthouse thinks about ROI when we’re developing solutions. This behind-the-scenes look at our process demystifies the AI arms race swirling around you, reveals what it takes to make solutions with true impact, and shows why Lighthouse is a leader in reliable, high-quality AI tools for eDiscovery.
‍

About Fernando Delgado

Fernando is Head of AI and Analytics at Lighthouse. He holds a PhD in Information Science from Cornell University and works at the forefront of implementing AI into eDiscovery workflows for many of our Fortune 500 and Am Law 200 clients.
‍

About Lon Troyer

Lon is the Head of AI and Review at Lighthouse, overseeing the strategic application of advanced technologies to improve client outcomes in eDiscovery processes. Lon works closely with Fernando to bring cutting-edge AI solutions to market.

‍

Lon: Fernando, as the head of our AI solutions team you've spearheaded many exciting developments in how we apply generative AI to eDiscovery. Can you share what you're seeing in terms of how GenAI is affecting our industry?
‍

Fernando: It has transformed workflows already, and the changes will keep coming. Clients and colleagues alike are viewing this transformation with a mix of curiosity and caution. They want to know how to unlock GenAI’s potential while maintaining the appropriate considerations for legal work.
‍

Lon: Let’s talk about the framework your team created to guide the development of GenAI solutions at Lighthouse. First, why did you create it?
‍

Fernando: The framework is a series of steps that guide our thinking at each stage of the development process. We created it to keep us grounded in the reality of what AI can and can’t do in the eDiscovery context. It ensures we don’t get swayed by hype and can stay focused on the ultimate goal: delivering tangible value to our clients.
‍

Lon: It holds you accountable, in a way.
‍

Fernando: Exactly.
‍

Lon: Talk me through how you use the framework. What’s the first step?
‍

Fernando: We start by establishing a clear value hypothesis. What’s the goal? Are we trying to make an existing process more efficient? Innovate and experiment to see what benefits we can unlock?
‍

Lon: Can you give an example?
‍

Fernando: Sure. With our GenAI Privilege Log Solution, we started with a hypothesis of making privilege review more efficient. It’s one of the most arduous and expensive parts of eDiscovery, so we set out to see if GenAI could reduce those burdens.
‍

Lon: What comes next in the framework?
‍

Fernando: Next comes robust problem definition. What specific workflow do we want to improve? What outcomes do we hope to achieve? In the privilege review example, we zeroed in on privilege logging, where a review team may have to summarize tens of thousands of documents, which can lead to inconsistent descriptions. Further, bundling privilege identification and privilege logging can slow down privilege review. We wanted to make that process easier while also improving the quality of the reviewer’s output.
‍

Lon: Okay, we’ve defined the problem and outcomes. Then what?
‍

Fernando: Then we define the technology approach—what exactly we're asking GenAI to do, which large language model (LLM) we'll use, and things like that. For privilege logging, we determined that GenAI would write draft privilege summaries for each privileged document, which addresses one of the most time-intensive aspects of the process. We also determined that we would use complementary natural language processing techniques to ensure consistency and standardization in output where appropriate.
‍

Lon: Earlier you mentioned people wanting to maintain appropriate considerations for legal work. Is that part of the framework?
‍

Fernando: Yes, that's a critical consideration. We look at the role of the user, such as whether the technology requires support staff, or requirements for a user interface. We also analyze what type and volume of data needs to be processed, identify other available data sources that might support the analysis, and determine the best way to organize, prepare, and stage the data. For privilege logs, we typically handle volumes around 6,000 documents, and we ensure the outputs are available in Relativity for attorney review, which integrates with their existing workflow.
‍

Lon: How does your team make sure a solution delivers on the value hypothesis you identified in the beginning?
‍

Fernando: That’s the last part of the framework: development and evaluation. We assess the accuracy and usability of the AI model’s output, and we fine-tune or augment the model as needed.
‍

Lon: What did you learn about the privilege log solution at this stage?
‍

Fernando: You know how GenAI responds to prompts? We learned that our privilege log solution produces the best results when our consultants tailor the prompts with feedback from outside counsel. This means the solution requires our consultants—attorneys don’t use it themselves. But it enables the model to generate extremely accurate content. Using this approach, 15-20% of privilege logs generated by our solution require targeted QC.
‍

Lon: Wow, really following through on that efficiency goal. Let’s zoom out now. Based on your own work and the trends you’re observing, what's your perspective on the future of AI in eDiscovery?
‍

Fernando: There’s no question that AI has unprecedented potential to streamline complex legal workflows—but successful integration is a lot more than chasing trends. It’s about methodically addressing specific challenges with precision and purpose.

That’s why we created our GenAI solution development framework. And it’s why our privilege log solution is so effective. Our commitment to iterative development, rigorous evaluation, and close collaboration with legal teams ensures that AI is a powerful, reliable tool rather than an expensive shiny object or unpredictable black box.
‍

Lon: What’s the next step for someone interested in exploring how solutions like these might help with their own eDiscovery challenges?
‍

Fernando: Check out our AI solutions page or reach out to one of our experts. We're always happy to discuss specific use cases and how our approach to AI can help you with your unique challenges.