From Documents to Answers: eDiscovery’s Next Advance Isn’t Faster Review
For more than two decades, the eDiscovery industry has devoted extraordinary resources to answering a particular question: How can we review documents faster, more accurately, and at a lower cost?
The advances have been substantial, from predictive coding and sophisticated analytics to generative AI. Yet much of that progress continues to be measured by how efficiently we process, classify, and review information.
When a general counsel learns of potential corporate misconduct, however, the immediate concern is rarely how many documents need to be reviewed. The questions are more fundamental: What happened? Who was involved? What evidence exists? What is the company’s potential exposure?
These are questions about facts, not documents.
What if the next meaningful advancement in eDiscovery comes not from reviewing information more efficiently, but from understanding what happened sooner and letting that understanding guide the work that follows?
Efficiency Is Not the Same as Direction
Modern eDiscovery has made remarkable progress. Deduplication, email threading, communication analysis, early case assessment, and technology-assisted review have reduced unnecessary work and made large matters more manageable.
Experienced investigators have also long relied on interviews, targeted searches, and evolving hypotheses to develop factual understanding. The idea of investigating before reviewing is hardly new.
But much of the industry’s technological innovation has remained document-centric: How many documents can we eliminate, prioritize, classify, or review?
Those are important measures of efficiency, but they do not necessarily tell us whether we are moving toward an understanding of the underlying events. A team can review a document population efficiently and still spend considerable time developing the factual foundation counsel needs to make strategic decisions.
As I argued in Don’t Ask AI to Solve a Data Problem, better AI outcomes begin with better data strategy. But even with the right data and technology, we should ask whether we are organizing our work around the right objective.
Changing the Unit of Analysis
Consider two assignments involving the same collection of information.
The first asks a legal team to identify documents responsive to allegations of improper payments involving a third-party consultant.
The second asks whether improper payments occurred, who authorized them, what services were provided, and what the participants understood at the time.
The underlying evidence may be identical, but the objectives are different. The first emphasizes classifying documents. The second requires reconstructing events, identifying relationships, testing explanations, and determining what the available evidence actually establishes.
Both may require reviewing many of the same materials. The distinction is how the work is organized and what questions guide it.
This is where generative AI offers a meaningful opportunity: the analytical starting point can become a question, event, or relationship rather than an individual document awaiting a coding decision.
Traditional analytics already identify communication patterns, conceptual similarities, and relationships among documents. Generative AI, combined with retrieval and other analytical methods, adds the ability to examine substantive content across multiple sources, connect references that lack obvious search terms, compare explanations, and assemble preliminary narratives supported by underlying evidence.
The important change is not that attorneys can finally ask investigative questions. They always could. It is that AI can substantially reduce the effort required to explore and refine multiple hypotheses across a targeted body of evidence.
Instead of committing significant resources to reviewing a broad population before developing an informed factual picture, teams can investigate several questions in parallel, identify promising leads, and refine their approach as evidence emerges.
The questions become easier to pursue. Establishing reliable answers still requires careful work.
And the capability has boundaries. AI can only analyze the information made available through collection, indexing, and retrieval. Its conclusions remain dependent on the completeness and quality of those processes.
An Investigation That Changes Direction
Consider a hypothetical internal investigation involving allegations that a regional sales executive improperly influenced the selection of a third-party distributor.
The company preserves potentially relevant information and identifies an initial group of custodians. Available materials include email, collaboration messages, contracts, purchasing records, and internal approval documents.
A capable discovery team would already employ searches, interviews, and analytics to understand the allegations. An investigation-led AI workflow would build on those efforts by exploring several factual questions simultaneously.
How was the distributor selected? Who advocated for the decision? Were competing proposals considered? Did anyone raise concerns before approval?
The analysis identifies a Teams exchange referring to “Dave’s group” as the preferred choice, without mentioning the distributor by name. Viewed alongside a meeting invitation and draft vendor scorecard, the communication suggests a connection to the selected company.
Investigators verify that connection against the underlying records.
The system also flags an apparent inconsistency between an approval date and a meeting record. Further examination reveals that the meeting had been rescheduled. The apparent discrepancy is explained, and the lead is discarded.
A plausible AI-generated discrepancy is not evidence until someone verifies it.
Another finding proves more significant. Asked whether anyone questioned the selection before approval, the system identifies a message in a group channel collected from another custodian, sent by a procurement employee who was not among the initial custodians. The employee had raised concerns about the distributor’s qualifications before the executive’s recommendation.
That discovery changes the direction of the investigation.
Counsel may now need to interview the procurement employee, determine who received the warning, examine subsequent approval decisions, and consider additional custodians or information sources.
An experienced investigative team could uncover these same facts through conventional methods. The distinction is timing. Identifying the connections early allows the findings to influence the interviews, collection decisions, and review priorities that follow.
The result is not necessarily a smaller investigation. It is a better-informed one, potentially broader where the evidence warrants further inquiry.
The Hardest Question: What Did We Miss?
There is a critical distinction between verifying an individual finding and establishing whether an investigation is sufficiently complete.
A sourced communication may establish that someone raised concerns about a transaction. It cannot, by itself, establish that no one else expressed reservations, that contradictory evidence does not exist, or that every relevant source was examined.
An AI-generated explanation can be accurate within the information it references while remaining materially incomplete.
Technology-assisted review developed established approaches to evaluating classification performance, including statistical sampling and recall estimation. Question-driven AI investigations present a different validation challenge. Confirming that a system accurately cites supporting documents does not establish whether its retrieval process overlooked other important evidence.
There is no single measure that resolves completeness across every open-ended investigative question.
A defensible methodology therefore requires examining the sources included in the analysis, testing alternative searches and competing hypotheses, evaluating information outside the AI-identified population where appropriate, and documenting material gaps.
Investigators must also recognize the limitations of the available evidence. Important conversations may reside in uncollected messaging applications, personal devices, or communications that were never preserved electronically.
The distinction is straightforward: A verified finding establishes something the available evidence supports. A defensible conclusion requires understanding what else the evidence might show, including what remains unknown.
This is not merely a technological limitation. It is a fundamental principle of sound investigation, one that becomes even more important when AI can generate persuasive narratives from incomplete information.
Investigation Is Not Discovery
An investigation-led methodology must also recognize that understanding the apparent facts of a matter is not equivalent to satisfying legal discovery obligations.
In an internal investigation, earlier factual understanding can help counsel prioritize interviews, identify additional evidence, assess exposure, and determine which leads warrant further attention. Appropriate safeguards for privilege, confidentiality, and AI-generated investigative records remain essential.
When litigation or regulatory demands are involved, additional requirements govern the process.
Investigative findings may inform custodian selection, reasonable search planning, and discussions about scope and proportionality. They do not replace preservation obligations, responsiveness and privilege determinations, or required productions.
Federal Rule of Civil Procedure 26(b)(1) establishes a proportionality framework for federal civil discovery. Better factual understanding can help inform decisions and discussions about an appropriate discovery scope, but it cannot justify narrowing an otherwise reasonable search simply to conform to an initial investigative theory.
A Different Measure of Progress
The opportunity presented by generative AI is not to replace established eDiscovery practices. It is to integrate investigations, early case assessment, analytics, and document review more deliberately, allowing factual understanding to guide the work as it unfolds.
Document review will remain essential. Legal obligations will continue to determine what must be preserved, searched, reviewed, and produced. Human judgment will remain indispensable to evaluating evidence and testing competing explanations.
But the ability to investigate questions earlier, connect evidence across sources, and develop verified findings before committing to extensive downstream review creates an opportunity to rethink how we measure progress.
The next evolution of eDiscovery won’t be defined by how quickly we can get through the documents, but by how much sooner we can get to the answers.
__
About Author
Marco Nasca is the Executive Vice President of Sales at Lineal and a 2001 graduate of DePaul University College of Law. For more than two decades, he has worked at the forefront of eDiscovery and legal technology, advising corporations and law firms on the defensible and practical application of technology in complex litigation, investigations, and regulatory matters. His work focuses on structured data, emerging technologies, artificial intelligence, and the practical implications of evolving judicial doctrine.
__
About Lineal
Lineal is an innovative eDiscovery and legal technology solutions company that empowers law firms and corporations with modern data management and review strategies. Established in 2009, Lineal specializes in comprehensive eDiscovery services, leveraging its proprietary technology suite, AmplifyAI™ to enhance efficiency and accuracy in handling large volumes of electronic data. With a global presence and a team of experienced professionals, Lineal is dedicated to delivering custom-tailored solutions that drive optimal legal outcomes for its clients. For more information, visit lineal.com
Inhaltsverzeichnis
Abonnieren Sie unseren Newsletter
Vielen Dank für Ihr Abonnement.
Sie erhalten praxisnahe Einblicke, Produktupdates und Inhalte, die Ihr Team tatsächlich nutzen kann.

