eDiscovery AI: How Technology-Assisted Review Reduces Cost and Risk
Document review has always been the most expensive phase of litigation. In large matters, it is not unusual for review to consume the majority of the total discovery budget, attorneys working through thousands of documents, one at a time, billing by the hour. The math has never worked in anyone's favor. And for years, the legal industry largely accepted it as the cost of doing business.
eDiscovery AI changed that calculus. Technology-Assisted Review, Continuous Active Learning, and automated first-pass review have fundamentally altered what is possible, not just in terms of speed, but in terms of accuracy, defensibility, and cost control. This post explains how each of these technologies works, what they actually do to your review costs, and why the risk calculus shifts meaningfully when you deploy them correctly.
Why Document Review Is Still the Biggest Cost in Discovery
Before getting into the technology, it helps to understand the scale of the problem.
In 2024, document review spending reached an estimated $10.81 billion, accounting for 64% of total eDiscovery expenditures. That number reflects something most legal professionals already know from experience: review is where budgets go. It is labor-intensive and time-consuming, and in a linear manual review, it is almost entirely dependent on attorney throughput.
Traditional first-pass review worked like this: a project manager batched documents into sets of 500 and distributed them to contract reviewers, billing at $35 to $60 per hour. At an average throughput of 50 to 75 documents per hour, the all-in cost per document ranged from $1.50 to $3.00, before quality control, privilege screening, or any escalations. On a matter with 200,000 documents, the first pass alone ran $300,000 to $600,000.
That is the baseline. And for matters with document populations in the millions, large antitrust cases, complex commercial disputes, and regulatory investigations, the numbers scale accordingly. A million-document review at those rates is not a hypothetical. It is a real scenario legal teams face regularly, and it is the exact scenario that eDiscovery AI and TAR eDiscovery tools were built to address.
What Is Technology-Assisted Review in eDiscovery?
Technology-assisted review (TAR) in eDiscovery refers to the use of machine learning algorithms to analyze documents and predict their relevance to a matter. Rather than having every attorney review every document, TAR trains a model on a smaller set of human-coded documents and then applies its predictions across the broader population, prioritizing likely-relevant documents for review and pushing likely-irrelevant ones to the bottom of the queue.
The result is that reviewers spend their time on documents that are likely to matter, rather than wading through the full universe of collected data in sequential order.
TAR has undergone significant evolution since its introduction. TAR 1.0, also called Simple Active Learning or predictive coding, required careful creation of a training set, a determination of when to stop training, and statistical validation of the model before review could begin in earnest. It worked, but the setup process was significant.
Continuous Active Learning, also known as TAR 2.0, refined TAR workflows by eliminating the need for a control set and reducing the amount of training and statistical analysis required. Unlike the two-step training and review process used in TAR 1.0, document reviews that leverage CAL can begin almost immediately because the CAL algorithm continuously ranks documents based on the decisions of the human review team and serves the highest-ranked documents to reviewers first.
That distinction matters practically. In a fast-moving investigation or a matter with a tight production deadline, waiting weeks to complete a training phase before review can begin is not always an option. CAL removes that bottleneck.
Continuous Active Learning Review: How It Works
Continuous Active Learning is an advanced machine learning protocol used in eDiscovery to identify and prioritize relevant documents. The system analyzes the text, metadata, and context of every coded document to score the remaining unreviewed population, constantly re-sorting the review queue so the most likely responsive documents are always served next. This results in a cycle where the model effectively pushes non-responsive documents to the bottom of the pile, surfacing the signal through the noise.
The practical effect of this is significant. In a high-stakes antitrust matter with one million documents, a linear review might not surface the most critical documents until reviewers are deep into the population, potentially weeks or months into the project. With CAL, you might find your first smoking-gun document on day one rather than day 45. The most relevant material rises to the top immediately and stays there as the model learns.
CAL uses attorney decisions to continuously train the software to identify which documents are likely responsive, serving up the most promising material first. The model updates automatically as the review proceeds, stopping once a certain threshold of non-responsive documents is reached. TAR 2.0 has quickly become the default technology for prioritizing and identifying responsive electronically stored information at scale.
One important nuance: continuous active learning review does not eliminate human judgment; it directs it more efficiently. Attorneys are still making the relevant calls. The model learns from those calls and uses them to make the next batch of decisions more precise. The quality of the output is directly tied to the consistency and accuracy of the reviewers doing the coding, which is why calibration and reviewer training remain critical even in a fully AI-assisted workflow.
Automated First-Pass Review: What It Does to Your Cost Structure
Automated first-pass review takes this a step further. Where CAL prioritizes documents for human review, automated first-pass review uses AI classifiers to determine initial responsiveness across the full document population, flagging documents as responsive or non-responsive, often with a reasoning trace explaining which facts in the document drove the classification.
A large language model reads each document against a written responsiveness definition and returns a binary classification, responsive or non-responsive, along with a reasoning trace explaining which facts in the document drove the decision. The attorney reviewing the output sees not just the label but the model's reasoning, which makes quality control tractable and the workflow auditable. Platforms running this approach are reaching F1 accuracy of 0.86 at $0.017 per document, compared to $1.50 to $3.00 per document for traditional contract review.
That is not a marginal improvement. In a 200,000-document matter, the difference between $0.017 per document and $2.00 per document is between a $3,400 first-pass cost and a $400,000 one. The savings at scale are material enough to change how cases are staffed, how budgets are structured, and, in some cases, how settlement decisions are made.
Privilege detection runs on a similar model. Modern eDiscovery platforms use machine learning models trained on privilege determinations to flag potentially protected documents for human review. TAR and predictive coding tools can dramatically reduce the volume of documents requiring attorney eyes, cutting costs and compressing timelines that once stretched on for months.
This matters from both a risk and a cost standpoint. Inadvertent privilege waiver, producing a document you should have withheld, is one of the most damaging things that can happen in discovery. Automated privilege flagging does not eliminate the risk, but it applies consistent criteria across the entire document population in a way that human reviewers working under time pressure and fatigue cannot reliably match.
The Risk Side of the Equation
Cost reduction is the headline, but risk reduction is equally important, and often underappreciated.
Manual review is inconsistent by nature. Different reviewers make different calls on similar documents. Reviewers working late on a long project code differently than they did on day one. When a production decision is later challenged, reconstructing and defending the review methodology requires documentation that many manual review workflows never generated in the first place.
Technology-assisted document review creates a documented, auditable process. Every decision feeds into the model. The training history, validation statistics, and stopping criteria are all recordable. When opposing counsel or a court asks how the review was conducted, and this question is increasingly common, a team using TAR with proper documentation can answer it in detail. A team that conducted linear manual review often cannot.
Courts have consistently supported TAR when deployed with appropriate transparency and validation. The Sedona Conference has published guidance on TAR protocols, and case law in both federal and state courts has affirmed that TAR-assisted production can meet or exceed the reliability of traditional manual review. The key requirement is that the process is documented and defensible, which a well-implemented TAR workflow, by its nature, tends to be.
Real-World Scenarios Where AI in eDiscovery Changes the Outcome
Large document populations with tight timelines. A regulatory investigation that drops 800,000 documents into your lap with a 60-day production window is not manageable with linear manual review. With CAL and automated first-pass, the most relevant documents surface in the first days, privilege gets flagged systematically, and the team can focus its attorney hours on the documents that actually require legal judgment rather than spending them on obvious non-responsive material.
Rolling productions in complex litigation. In cases where new data keeps coming in as discovery proceeds, TAR 1.0's training-and-stop model creates problems; it was trained on earlier data and may not handle new document types or custodians well. CAL adapts continuously as new documents are coded, which makes it significantly better suited to rolling production workflows.
Matters where budget certainty matters. When a client asks what review is going to cost before you start, an automated first-pass review gives you a number you can actually defend. The cost per document is predictable, the timeline is compressible, and the QC process is systematic rather than dependent on reviewer count and throughput variability.
Investigations where early case assessment drives strategy. One underutilized advantage of eDiscovery AI is that it surfaces the most relevant documents early. Before a full production is complete, the model has already identified the documents most likely to matter, which means case strategy conversations can happen earlier, with more information, than they could in a traditional linear review.
How Premier Legal Technologies Deploys TAR and CAL
Understanding what these technologies can do is straightforward. Deploying them correctly, configuring the model, setting the responsiveness definition, calibrating reviewers, validating outputs, and managing the stopping criteria, is where expertise matters.
Premier Legal Technologies implements AI in eDiscovery through Reveal and Logikcull, two platforms that bring TAR, CAL, and automated first-pass review into a unified workflow. Reveal's Brainspace analytics and generative AI tools power advanced visual analytics and conceptual clustering, accelerating early case assessment. Logikcull's AI-driven culling and suggested tagging tools handle the speed and simplicity end of the spectrum, ideal for fast-moving matters where time is the primary constraint.
What makes the difference is not just the platform, but how the platform is configured and calibrated for each specific matter. Issue tags get defined clearly. Privilege rules get locked down before review starts. Reviewer training ensures that the decisions feeding the model are consistent enough for the model to learn from reliably. Sampling and validation reports are built into the workflow, not retrofitted at the end.
The result is a review process that is faster, cheaper, and more defensible than linear manual review, and one that holds up when needed.
If your team is evaluating whether TAR or automated first-pass review makes sense for an upcoming matter, the conversation starts with understanding your document population, your timeline, and your budget. Premier Legal Technologies works with legal teams to configure, calibrate, and manage AI-assisted review from start to finish. Contact us to schedule a consultation and find out which approach is the right fit for your next matter.
