What is eDiscovery in law?

What Is eDiscovery in Law? The Direct Answer

eDiscovery — short for electronic discovery, and sometimes written e-discovery or ediscovery — is the process of identifying, collecting, preserving, reviewing, and producing electronically stored information (ESI) during litigation, government investigations, or regulatory proceedings. It is the digital counterpart to traditional paper discovery, the pretrial phase in common-law civil procedure where each party obtains evidence from the opposing side. Under the Federal Rules of Civil Procedure in the United States, particularly Rule 26 and Rule 34 as amended in December 2006 to explicitly address ESI, parties are obligated to produce relevant documents in the format in which they are ordinarily maintained. That means emails, text messages, Slack threads, database records, cloud files, chat logs, voicemails, metadata, and even data from IoT devices can all fall within the scope of discovery.

Also worth reading: How do AI clawback provisions function within protective orders in modern eDiscovery? · How does Technology Assisted Review (TAR) compare to Large Language Model (LLM) document review in modern eDiscovery? · What are the specific risks of waiving attorney-client privilege when using AI tools for eDiscovery and legal document drafting?

The scale involved is what separates eDiscovery from ordinary document exchange. A single mid-sized commercial dispute routinely involves hundreds of thousands to millions of documents. The Electronic Discovery Reference Model (EDRM), first drafted in 2005 by George Socha and Tom Gelbmann, became the industry-standard framework precisely because handling that volume required a repeatable, defensible workflow rather than ad hoc printing and reading. In 2026, eDiscovery is no longer a niche specialty: it is a core competency for litigators, an entire vendor ecosystem worth billions of dollars annually, and one of the fastest-moving areas for artificial intelligence adoption in legal practice.

Why eDiscovery Exists: The Legal Obligations Behind It

eDiscovery exists because modern evidence is born digital. When Rule 34 was amended in 2006, the drafters recognized that requiring parties to print electronic records would destroy metadata, lose searchability, and misrepresent how information actually exists inside organizations. Courts have since sanctioned parties harshly for failing to preserve ESI. The landmark case Zubulake v. UBS Warburg (2003–2004) established cost-shifting principles and the duty to preserve; Pension Committee of the University of Montreal v. Banc of America Securities (2010) reinforced that gross negligence in preservation can support sanctions; and more recent decisions continue to penalize spoliation — the destruction or alteration of evidence — with adverse inference instructions, monetary fines, and in extreme cases default judgment.

The duty to preserve typically attaches when litigation is 'reasonably anticipated,' which can be months or years before a complaint is filed. This triggers a legal hold: a formal directive suspending routine document-destruction policies (such as automatic email purging after 90 days) for custodians relevant to the matter. Failure to issue and enforce a legal hold is one of the most common sources of sanctions. For corporate defendants, this means eDiscovery obligations begin at the first hint of a dispute, not when a lawsuit arrives, and the consequences of getting it wrong are financial and reputational rather than theoretical.

The Nine Stages of the EDRM Workflow

The EDRM breaks eDiscovery into nine overlapping stages, and understanding them clarifies why the discipline requires both legal judgment and technical infrastructure. First, Information Governance: organizations manage data proactively so future discovery is cheaper. Second, Identification: counsel determines which custodians, systems, and date ranges might contain relevant ESI. Third, Preservation: legal holds freeze potentially relevant data. Fourth, Collection: forensic specialists copy data in a defensible manner, preserving metadata such as creation dates, authors, and modification histories without altering originals.

Fifth, Processing: raw data is de-duplicated (removing exact copies), filtered by date range and file type, and converted into reviewable formats. Sixth, Review: attorneys examine documents for responsiveness, privilege, and confidentiality — historically the most expensive stage, often consuming 60–80% of total eDiscovery budgets. Seventh, Analysis: reviewers identify patterns, key players, and communication themes. Eighth, Production: responsive documents are delivered to opposing counsel, frequently Bates-stamped with unique sequential identifiers (a practice dating to pre-digital eras that eDiscovery software now applies electronically). Ninth, Presentation: documents are used in depositions and trial. Each stage has quality-control checkpoints, because an error early in the chain — a missed custodian, a broken collection — propagates through everything downstream.

How AI Has Transformed eDiscovery by 2026

Artificial intelligence reshaped eDiscovery in three waves. The first was Technology-Assisted Review (TAR), also called predictive coding, which emerged around 2011 following Da Silva Moore v. Publicis Groupe, the first case where a court expressly approved TAR. TAR uses machine-learning classification trained on attorney-reviewed sample documents to prioritize or auto-classify the remainder, cutting review costs dramatically — studies have shown reductions of 50% or more compared to linear manual review, while often achieving equal or better recall (the percentage of truly responsive documents found).

The second wave brought large language models. By 2026, generative AI can summarize thousands of documents, draft deposition outlines from reviewed material, identify privilege issues, and answer natural-language questions across a document corpus. Major platforms have embedded these capabilities directly: OpenText markets its eDiscovery Aviator agents, Casepoint partnered with Elastic to modernize its platform, HaystackID acquired the AI company eDiscovery AI, and Reveal announced integration with Thomson Reuters to connect reviewed evidence directly into AI research and drafting tools. Anthropic expanded Claude-based offerings for law firms, and ILTA partnered with BARBRI to expand eDiscovery education and certification — signals that AI-assisted discovery is becoming institutionalized rather than experimental.

The third wave is agentic workflow automation, where AI systems execute multi-step tasks like drafting legal-hold notices, tracking custodian acknowledgments, and generating production QC reports with human oversight. The caveat deserves emphasis: courts increasingly expect disclosure of AI use in some contexts, and hallucination risks mean AI outputs require verification. AI accelerates eDiscovery; it does not eliminate attorney responsibility under Rule 26(g), which requires a certification that productions are complete and correct after reasonable inquiry.

Comparing Your eDiscovery Options: In-House, Vendor, and Software Platforms

Organizations facing discovery generally choose among three models, each with distinct trade-offs in cost, control, and defensibility.

Full-service vendors suit high-stakes matters where defensibility outweighs cost. Self-service platforms fit law firms and corporate legal departments handling recurring mid-sized matters who want predictable pricing and direct control. Pure manual approaches remain viable only for tiny matters — a single-employee wage claim, perhaps — and even then, exporting email without preserving metadata creates avoidable risk. Many organizations adopt a hybrid model: self-service software for routine disputes, vendors for bet-the-company litigation. Pricing varies widely, so procurement teams should benchmark per-gigabyte processing rates, monthly hosting per gigabyte, and reviewer seat costs before committing.

Common Mistakes That Create Sanctions Risk

The most expensive eDiscovery errors happen before anyone opens a review platform. Waiting until a complaint is filed to suspend document destruction is the classic failure — courts evaluate whether litigation was reasonably anticipated, and routine auto-deletion running after that point constitutes spoliation regardless of intent. Overly narrow legal holds are a close second: limiting custodians to named executives while ignoring the employees who actually exchanged relevant messages leaves gaps opposing counsel will exploit. Third, many teams over-collect out of fear, pulling every mailbox for five years, which inflates processing and hosting costs enormously; a targeted approach based on custodian interviews and system mapping is both cheaper and more defensible.

During review, the recurring mistakes are underestimating privilege review (producing privileged documents waives protection unless clawback provisions under FRE 502(d) are secured), skipping metadata validation, and relying on keyword searches alone without testing their recall. Keyword-only culling routinely misses responsive documents because people describe the same events in unpredictable language. Finally, treating AI outputs as final answers without sampling and validation invites both accuracy problems and judicial skepticism. Documenting every decision — why custodians were chosen, why date ranges were set, how search terms were tested — is what turns a defensible process into a demonstrably defensible one when challenged at an evidentiary hearing.

When to Act: Timing, Deadlines, and Practical Steps

eDiscovery obligations operate on a clock that starts earlier than most non-lawyers expect. The trigger is reasonable anticipation of litigation — a demand letter, a serious internal complaint, a regulator's inquiry. From that moment, preservation must begin immediately, ideally within days. Once a lawsuit is filed in federal court, Rule 26(f) requires the parties to confer about discovery within 21 days after any defendant appears, and to submit a discovery plan 14 days later; initial disclosures under Rule 26(a)(1) follow within specific deadlines set by the scheduling order. Missing these milestones narrows your strategic options and hands leverage to the other side.

Practically, the sequence looks like this: issue the legal hold within the first week of anticipation; interview key custodians within two weeks to map data sources; complete forensic collections before data ages out of retention windows or employees leave; negotiate ESI protocols with opposing counsel covering formats, search methodology, and privilege handling; then run processing and rolling productions on the schedule the court sets. Rolling productions — delivering documents in batches rather than all at once — keep matters moving and spread review workload. If you are a business without in-house capability, engage counsel or a platform provider at the anticipation stage, not after the Rule 16 conference, because retrofitting preservation after gaps exist is far harder than maintaining it from day one.

Costs, Budgets, and Where the Money Actually Goes

eDiscovery spending concentrates unevenly across the workflow. Industry analyses consistently show document review consuming the majority of matter budgets — commonly cited figures range from 60% to 80% — because it is labor-intensive attorney time. Data processing typically runs tens of dollars per gigabyte, hosting runs a few dollars per gigabyte per month, and collections by forensic professionals can range from a few thousand dollars for simple single-custodian jobs to six figures for multi-national matters involving mobile devices, chat platforms, and structured databases. A contested commercial case with 500,000 documents can easily generate $200,000 to $1 million in total eDiscovery spend depending on complexity and review staffing.

AI changes this arithmetic meaningfully. TAR and LLM-assisted first-pass review reduce the human hours needed, shifting spend toward technology fees and away from contract attorney line items — which is why G2's 2026 roundups of eDiscovery software emphasize AI features as primary selection criteria. Cost-shifting motions under Rule 26(b)(1) proportionality limits also matter: courts may order the requesting party to bear some costs when requests are disproportionate to the needs of the case. Smart budgeting means scoping early, negotiating flat-rate or capped processing fees, using analytics to shrink review populations before humans read anything, and reserving premium forensic services for the custodians and systems that genuinely warrant them.