[Podcast] HaystackID® in the EDRM Illumination Zone: Jim Sullivan

AI isn’t the Differentiator in eDiscovery—Execution Is

When collecting hundreds of thousands of documents, legal teams are sifting through everything from emails to Slack messages to internal communications to financial records. They’re often sorting through years of activity spanning multiple custodians. They have a finite amount of time (and resources), and there’s a lot that they don’t know at this stage, like:

• What’s in the data.
• Where risk sits.
• Which facts will shape the case.

What they do know is that they have to process and review a wealth of information while racing to find answers fast enough to meet deadlines.

Actionable insights often come later in the game, often after substantial time and cost have already been spent, which means teams may have already committed significant resources before knowing whether they are focused on the right issues. By the time patterns emerge that can inform how the team will proceed, executing on the strategy is already underway.

In a recent episode of the EDRM Illumination Zone podcast, Jim Sullivan, founder and CEO of eDiscovery AI, shared how legal teams can get insights earlier using AI and how this technology is solving issues that have persisted in our industry for years.

There’s a New Starting Point in eDiscovery

After working in eDiscovery for nearly a decade, Sullivan decided to pursue new territory and build automation solutions for small businesses. However, he didn’t stay away for long. What drew Sullivan back into the field wasn’t a shiny new problem plaguing legal professionals; it was the technology available to solve them, and the speed at which that could happen.

For Sullivan, the opportunity wasn’t to introduce something entirely new, but to revisit long-standing challenges with tools that could finally address them. That shift doesn’t change the structure of eDiscovery—it places more emphasis on the earlier stages of the process.

Because early missteps can be so expensive to unwind, teams are reconsidering where analysis belongs in the workflow. As those capabilities improve, more of the work that historically happened during review is moving further left on the EDRM. Analysis, classification, and early insight no longer need to wait until documents are fully processed and queued for review. Legal teams can begin to understand their data earlier, before time and costs accumulate.

“As computers can provide more and more analysis, it really just seems completely normal to say we should do that analysis earlier on,” Sullivan explained. “There’s no need to wait to get understanding about your data if a computer’s going to do that review.”

This is where the change becomes practical. Moving left isn’t about altering the process—it’s about changing when key decisions are informed.

That approach is also reflected in how these capabilities are being deployed. HaystackID’s acquisition of eDiscovery AI represents a shift from partnership to full integration, bringing proven GenAI workflows directly into our broader discovery and investigation ecosystem. Rather than requiring teams to adopt new infrastructure, the technology operates as an intelligence layer within existing environments, surfacing patterns, relationships, and risks across large datasets while maintaining the precision and recall required for high-stakes matters.

The Defensibility Standard Hasn’t Changed

There’s less conversation amongst legal teams about whether they should use AI; it’s how the technology will hold up under scrutiny. However, when it comes to AI, the industry isn’t starting from scratch.

“The validation piece around data classification, we really feel like has already been litigated, established, and proven to be defensible in tens of thousands of cases,” he said.

For classification tasks, the same validation methods that have supported technology-assisted review (TAR) for years still apply. Sampling, control sets, and precision and recall calculations remain intact. While the technology has changed, the underlying framework has not.

But validation isn’t the only concern for legal teams. Where AI changes the conversation is in the type of output it produces. GenAI introduces new forms of work product—summaries, timelines, and responses generated through chat-based interactions—that do not fit as neatly into existing validation models. In some cases, approaches to validating these outputs are still evolving, depending on the use case.

That shift does not remove responsibility from legal teams.

“You have an obligation to confirm that all the citations you have are legitimate and true and correct,” Sullivan said. “If you sign your name on something, you need to make sure that that’s accurate and proven.”

It doesn’t matter how the content was created, whether a legal team relied on traditional research, got insight from a colleague, or used AI. That information has to be verified before it is used to support legal arguments or filings.

AI can accelerate analysis and surface insight more efficiently, but it does not change the standard for defensibility. It changes how that standard is met—requiring legal teams to validate outputs in context and tie them back to source data before relying on them.

Where AI is Doing the Work

A year ago, many of the use cases now gaining traction with AI were not even part of the conversation. Privilege review was one of them. The idea of using GenAI to support privilege logs was met with hesitation—too important, too sensitive, too much risk. Today, that same workflow is seeing significant activity, not because the stakes have changed, but because the results have.

“We’re seeing a very, very high appetite for using generative AI to generate privilege log entries to speed up the review process,” Sullivan said.

Privilege workflows have long required significant manual effort to generate log entries at scale, maintain consistency, and meet production timelines. AI is now being applied directly to that work—shifting effort away from drafting entries and toward validating and refining outputs before production.

Teams are also extending these tools into adjacent areas. Early case assessment workflows are being used to support deposition preparation, surface key themes, and identify gaps in the record before the strategy is finalized.

That impact becomes more pronounced in complex matters. In one internal corporate investigation involving more than 2.2 million documents, legal teams needed to quickly understand the facts behind a rapidly developing issue with limited information at the outset. Using HaystackID® Core Intelligence AI Case Insight™, critical communications, timelines, and decision points surfaced within 48 hours, enabling counsel to establish a factual foundation early and act more quickly.

What begins as a targeted application becomes part of how work gets done. AI is not replacing legal workflows; it is increasingly being applied where manual effort has historically defined the process.

The Great Divide Isn’t the Technology

If you’ve been in this industry for twenty years or five, we’ve all heard the same talk track around growing data volumes, limited resources, and the need to validate. What’s changing in eDiscovery isn’t driven by the technology itself; it’s driven by how teams decide to use it.

More forward-looking teams are putting AI to work on active matters, using it to pressure-test their approach, surface inconsistencies, and adjust in real time. Others are holding off, uncertain what “good” looks like before they commit. The difference often comes down to how clearly the technology fits into the work itself, whether it solves a defined problem and can be applied without adding friction.

“Solving problems is the number one piece, but then second is just making it as easy as possible to get people to adopt it,” Sullivan said.

The teams that tend to gain the most ground are identifying where work slows down and targeting those points specifically. The AI they deploy is measurable, consistent, and aligns with how their reviewers already operate.

As a result, the real competitive gap is not who has access to AI, but who can turn it into a reliable, repeatable practice. The gap between teams is less about capability and more about execution. And as insight continues to move earlier in the process, that disconnect will only become more visible, shaping not just how quickly teams can respond, but how effectively they can act when it matters most.