AI in Action: Improving TAR in Antitrust Matters

Summary: Agility, precision, and the ability to stay ahead of the curve are everything in antitrust. In this Q&A, Lighthouse experts Kamika Brown (Client Success Manager, Antitrust) and Fernando Delgado (Senior Director, AI & Analytics) discuss how large language model-backed AI can help antitrust attorneys stabilize TAR models faster, streamline review, and get to the documents that regulators and opposing counsel care about, earlier.

‍

Kamika: How are antitrust teams using Lighthouse AI to improve TAR in antitrust matters?
‍

Fernando: As you know, deadlines are especially tight in antitrust matters—especially during regulatory reviews like Second Requests. Every day of review counts. Because of this, we’ve embedded LLMs into the TAR workflows that antitrust teams are already comfortable with using during eDiscovery. These models enhance what they already do, by helping them stabilize TAR faster, without asking them to overhaul their process when their time is already tight. It’s a smarter, faster way to review.
‍

Kamika: What makes Lighthouse’s LLMs and linguistic modeling a good fit for improving TAR workflows?
‍

Fernando: What sets the tech apart is the way we use language modeling to find meaningful patterns in the data. Lighthouse AI doesn’t just look at individual keywords or concepts in isolation (the way machine learning tools traditionally used in TAR workflows do); it analyzes language in context.

That means we’re able to surface relevant materials faster and more accurately, which helps teams hit stabilization much earlier in the process.

And again, the workflows attorneys are using stay familiar—they don’t really have to learn new tools or rethink their approach. They just get better results, sooner.
‍

Kamika: Can you share a real-world example of that impact?
‍

Fernando: Absolutely. In one recent antitrust regulatory review—over 30 terabytes of data—the client used our LLM-powered TAR approach. The model hit stabilization after reviewing just over 4,000 documents.

We reached over 85% precision at 76% recall. That meant we could confidently limit the scope of review much earlier than expected, without sacrificing defensibility. It was a game-changer.
‍

Kamika: How does that compare to if they had used traditional TAR tools?
‍

Fernando: If they had used traditional TAR without the enhancements from Lighthouse AI, we calculated that over two million additional irrelevant documents would have remained in scope and require review.

These kinds of results aren’t theoretical anymore. They’re happening right now, and they show the real-world ROI of using modern, case-tuned AI the right way.
‍

Kamika: Why predictive AI and not generative AI? Aren’t we hearing that generative AI is better and faster?
‍

Fernando: That’s a great question, and it’s one we get a lot. It’s all about using the right tool for the right job.

Predictive AI is designed to classify and prioritize documents at scale. It’s built for precision and recall, which is exactly what TAR workflows demand—especially in antitrust regulatory reviews where regulators expect defensible results backed by solid math and proven methods.

Generative AI, by contrast, is not optimized to be a classification engine and isn’t built to deliver the statistical rigor or consistency required for TAR workflows in regulatory reviews.
‍

Kamika: If predictive AI is better in TAR workflows, is your team using generative AI at all in antitrust matters?
‍

Fernando: Absolutely. We’re seeing fantastic results where GenAI is used strategically for other review purposes.

We use GenAI for privilege log generation and names legend creation. These are traditionally repetitive, time-consuming tasks where we use GenAI to accelerate the creation of privilege logs, while keeping humans in the loop.

And our AI image analysis solution uses GenAI to extract content from visuals—think screenshots, charts, and embedded diagrams—that are otherwise missed by traditional TAR. That’s becoming increasingly relevant in antitrust reviews, especially with mixed-format data sources.

We’ve also recently launched our Lighthouse AI Search tool, which uses GenAI to help teams quickly identify critical documents for early case insights, regulatory responses, and deposition prep. Our clients are using it in antitrust reviews to understand themes, key facts, and timelines faster, without sifting through everything manually. They can also use it to replace issue coding (which, in turn, speeds up the document review process).
‍

Kamika: Does bringing in AI increase risk in antitrust regulatory reviews?
‍

Fernando: Actually, we’ve found that using AI, the right way, significantly reduces risk.

Every use of Lighthouse AI in Second Requests and other types of antitrust regulatory reviews is tailored to the case and fully auditable. We tune and test every model with the specifics of the matter in mind.

Our predictive and generative AI solutions have been used in many regulatory reviews without pushback from the DOJ or FTC. That’s because we prioritize explainability. We don’t treat AI as a black box; we partner with counsel to ensure they understand and are confident in how the models work, how decisions are made, and can defend those decisions if ever questioned. In a regulatory environment, that level of transparency matters.
‍

Kamika: What advice would you give to antitrust attorneys who are wary of using AI in TAR workflows in antitrust?
‍

Fernando: From what we’ve seen over the last few years, antitrust regulatory reviews like Second Requests are actually one of the best use cases for trying AI because the scale, speed, and scrutiny demand smarter tools, but also offer the structure that makes AI deployment more predictable and defensible.

You don’t have to change your entire workflow. You can integrate AI backed by LLMs into familiar TAR processes and see measurable improvements, fast. It’s a great proving ground.

Plus, when you partner with an experienced team, you’re not navigating it alone. You get built-in support, model validation, and defensibility, all of which help you build confidence in how AI works, and how to use it to your advantage

‍

Want more practical insights on emerging antitrust workflows?

• Read the Lighthouse 2025 Antitrust Trends Report
• Explore our AI solutions for antitrust