Which Tech is Right for Your Review? A Practical Guide to Modern Ediscovery Tools

Here's something you might not expect to hear from a legal tech company known for AI innovation: Generative AI (GenAI) isn't always the right answer for your document review.

The truth is, the best review technology depends on the matter — the timeline, budget, document population, and the stakeholders' comfort with different approaches. Sometimes that's GenAI. Sometimes it's TAR. Sometimes it's a prioritized linear review with human eyes on every document.

In this article, we'll look at four document review methodologies, their strengths, limitations, and ideal use cases. We'll also provide a practical framework for matching the right technology to your next review.

Setting the stage: the modern review challenge

Picture this: you've just been handed 150,000 documents and told you have two weeks to review and produce them. The clock is ticking, the client is anxious, and opposing counsel is watching. What do you do?

This scenario isn't hypothetical. Legal teams regularly face this type of high-pressure situation, navigating tight timelines, demanding client expectations, and budget constraints that leave little room for error. The stakes are high. A missed deadline can derail a case. An inefficient review can balloon costs. 

More importantly, choosing the wrong technology can leave your team scrambling to catch up.

That's why tech selection matters more than ever. Today's ediscovery landscape offers more options than ever before, from traditional linear review to sophisticated AI-driven solutions. But here's the truth that many overlook: The newest technology isn't always the best fit for every case. 

In the industry, we tend to believe that because GenAI is the latest technology, it must be the best. But that's not always the case. Each tool has different purposes and advantages for different reviews.

The key to review success lies not in chasing the latest innovation but in matching the right tool to your specific needs. This guide will walk you through a practical framework for doing exactly that.

💡 Curious about the evolution of GenAI in law? This webinar separates hype from hard-won lessons and highlights the emerging technologies and practices legal professionals need to know.

Step 1: Define your goals and scope the project

Before you can choose the right technology, you need to understand two things: what you're trying to achieve and what constraints you're operating within. The worst thing you can do is spend $100,000 on a case that's worth $1,000, so the first step is always scoping.

Key scoping questions

These five questions help you define your goals, identify your constraints, and determine which approach makes sense.

What kind of documents are you dealing with? 

This has to do with the nature of the data you’re dealing with. Consider volume, complexity, file types, and languages. A 50,000-document set of straightforward business emails presents a very different challenge than 500,000 documents spanning multiple languages with technical terminology and privileged communications. Our guide on complex data types in ediscovery may help.

What does a successful outcome look like? 

It’s important to understand what success looks like for this matter. Are you optimizing for speed to meet an aggressive deadline? Cost efficiency to stay within budget? Risk mitigation to ensure nothing falls through the cracks? The answer will shape which tools make sense. Here’s why general counsels often push for tech-driven litigation strategies.

Who's involved in this review? 

What are the stakeholders’ priorities? Are you working with internal teams, external reviewers, or subject matter experts? Understanding your human resources helps determine how much you can lean on technology versus people. 

Are there any barriers to using advanced technology? 

Some clients or opposing counsel may be skeptical of AI-assisted review. Some courts may have specific requirements. Understanding these constraints early prevents surprises later.

What are your deadlines?

When does production start? When is substantial completion required? Is this a rolling production or a single deadline? Timeline pressure is often the single biggest factor in technology selection.

Determine your must-haves versus nice-to-haves

Once you've answered these questions, categorize your requirements. What features are absolutely essential for this matter? What would be helpful but isn't critical? This distinction helps you avoid over-investing in capabilities you don't need while ensuring you don't skimp on what matters most.

Step 2: Understand your options

The right tech for the right case is never one-size-fits-all. With your scope defined, it's time to evaluate the tools at your disposal. 

Today's review technologies fall into four main categories, each with distinct strengths and trade-offs. Importantly, all of these methodologies leverage AI models in some form and have proven successful in real-world litigation.

Prioritized linear review: AI analysis, human review

In prioritized linear review, AI analyzes the full document set and assigns each document a relevance score, but humans are responsible for the actual review decisions. 

This workflow keeps humans in control while using AI to determine what should be reviewed first — and, where appropriate, when it is defensible to stop reviewing the lowest‑value material.

AI assists by prioritizing documents based on signals such as keywords, concepts, metadata, and communication patterns, then sorting them by score so reviewers always see the most likely responsive documents next. As reviewers tag documents, the AI learns from those decisions and continuously re‑prioritizes the remaining population. 

In this workflow, AI does not replace human judgment. It amplifies it. By constantly pushing the strongest candidates to the front of the queue, AI helps reviewers spend more time on documents that matter and less time slogging through obvious noise. Over the life of a matter, this continuous feedback loop improves speed, consistency, and quality while still giving legal teams clear insight into how documents were scored and reviewed.

Advantages: This approach surfaces important documents early, which is valuable for case strategy and settlement discussions. Reviewers see the relevant documents first, before fatigue sets in. The methodology is simple to explain to clients and courts, and it's accessible for teams new to AI-assisted review. 

For stakeholders who aren't comfortable with documents being excluded from human review, prioritized linear offers a way to leverage AI while maintaining complete coverage.

Disadvantages: Because every document still requires manual review, this approach can be slow and labor-intensive for large datasets. Overall review costs tend to be higher than alternatives that reduce the review population. Human inconsistency can also creep in over long review periods.

TAR 1.0: Trained AI with cutoff-based review

In TAR 1.0 (Technology-Assisted Review), human reviewers first train the system on a carefully selected “seed set” and “training rounds” of documents that includes examples of both responsive and non-responsive material. The model learns from those coding decisions and uses them to score the rest of the population for likely responsiveness. 

Teams typically run several training and validation rounds, refining the model until it hits agreed precision/recall targets and behaves reliably on held‑out samples (“Control Set”).

Once the model is validated, it ranks the full document set, and you draw a defensible cutoff line on the score curve. Documents above the cutoff are presumptively responsive and are either human-reviewed or QC’d, while documents below the cutoff are presumptive not responsive and are generally not reviewed at all, subject to sampling and quality checks. 

The core purpose of TAR 1.0 is to shrink the eyes‑on population by using that trained model and cutoff, not just to reorder the queue.

How TAR 1.0 differs from prioritized linear

• Prioritized linear review: The system continuously reprioritizes documents so humans always see the “most likely to be relevant” remaining docs next, but the conceptual default is “we’ll keep reviewing until we’ve reviewed every document.” The emphasis is on order of review, not on a hard statistical cutoff.
• TAR 1.0: There is a distinct training phase and an explicit classification/cutoff decision. Once trained, the model is used primarily to separate the corpus into “responsive slice” vs “not responsive slice,” backed by sampling and validation.

You can think of it as: prioritized linear makes review smarter, TAR 1.0 makes the review set smaller in a formal, testable way.

Advantages: TAR 1.0 can reduce human review volumes to roughly 10,000–25,000 documents, even when the starting corpus contains hundreds of thousands or millions of files, delivering significant cost and time savings over linear approaches. Once trained, the model applies the same criteria across the population and doesn’t fatigue or drift, helping smooth out differences among reviewers. It’s also statistically defensible with well-established legal precedent, including the landmark Da Silva Moore decision, which first validated TAR in federal court.

Disadvantages: TAR 1.0 requires a well-curated seed set, which takes time and expertise to develop. If your seed set isn't representative of the full document population, the model may not perform well at the outset and may require additional training rounds. Even at the cutoff point, the model may require sacrificing precision to meet a defensible recall threshold. For some stakeholders, the model can feel like a "black box" that's difficult to explain, particularly when justifying why documents below the cutoff weren't reviewed.

Case study

A civil liability defense firm faced a challenging matter involving a sensitive topic. With 1.6 million documents and a two-month production deadline, the team leveraged DISCO AI to create an efficient, defensible workflow.

DISCO's AI Consulting team identified responsive documents for AI training, while the managed review team built workflows for predicted responsive documents covering relevancy, issue tags, and redaction. The three-stage review process began with an AI-driven workflow processing all 1.6 million documents, followed by a second-level review of 202,000 documents, and finally manual redaction of 62,000 documents.

The results were impressive: 35,000 documents human-reviewed in one week with 82% recall and 87% precision, metrics that represent a high bar for any review. The firm successfully reduced its review population from 1.6 million to 250,000 documents while maintaining quality through minimal human review.

🔎 Dive deeper: Here’s a quick analysis of TAR workflows and how they save time and reduce costs while improving the quality of the review process.

TAR 2.0: Continuous learning TAR protocol

TAR 2.0 (also called Continuous Active Learning or CAL) takes the same basic idea as prioritized linear review — reviewers code documents while AI keeps pushing likely-responsive items to the top of the queue — but turns it into a formal TAR protocol with no separate training phase. Instead of curating a seed set as in TAR 1.0, reviewers simply begin their work, often from search results or a small starter set, and every coding decision immediately feeds back into the model.

As review progresses, the system continuously re-ranks the remaining population, selecting the next-most-likely-responsive documents for human eyes, then updating its predictions again based on how those documents are coded. Over time, this loop both prioritizes review (like a sophisticated prioritized linear workflow) and generates the performance metrics and sampling needed to decide when you’ve reached a defensible stopping point — that is, when further review of the low-scoring tail is unlikely to uncover many additional responsive documents.

Advantages: TAR 2.0 gets you to relevant documents faster because there's no waiting for seed set development and training rounds, and the process typically allows the review team to cut off a significant portion of the review population from human review.

Disadvantages: The continuous nature of TAR 2.0 requires skilled reviewers and project managers who understand how to work with the technology. Poor coding decisions early in the review can send the model in the wrong direction, so quality control is critical throughout. 

Determining when to stop reviewing — the "stopping point" — requires statistical validation, which can be hard to explain to stakeholders. Some perceive it as more complex to defend than a traditional review, though court acceptance is now well-established.

Case study: Kennedys

When the London-based team at Kennedys Law LLP received a construction dispute with 1.4 million documents and just four weeks until the production deadline, they turned to DISCO Ediscovery. Using DISCO's high-speed uploader, the team ingested and processed the first 1.2 million documents in just eight hours, with an additional 254,000 documents processed in under two hours the following evening.

Using DISCO's powerful search tools, Kennedys analyzed the data and identified relevant date ranges, senders, and recipients. By demonstrating their methodology to opposing counsel, they successfully negotiated to reduce the 1.4-million document universe to a review population of 123,000 documents. The team then implemented an AI-prioritized review, achieving a responsiveness rate of over 80%, nearly four times higher than the general document population.

The results: By reviewing just 1.85% of the initial 1.4 million documents collected, Kennedys met their tight production deadline with confidence, completing the review and production in just 11 days.

Michael Hogg, Partner at Kennedys, observed, "DISCO provides cutting-edge software, giving clients a compelling commercial solution for data-heavy litigation. It helps us deliver comprehensive advice early in the life of a complex dispute, while reducing upfront costs and overall legal spend."

GenAI: AI-first review

Generative AI takes a fundamentally different approach from TAR and prioritized linear workflows. Instead of learning from reviewers’ coded examples, a large language model applies your natural-language tag instructions (“prompts”) directly to the document set, making the initial responsiveness and issue calls itself.

In practice, the review team describes what they’re looking for in plain English, the model reads and interprets each document against those prompts, and then both humans and the model review sample sets of documents to assess the effectiveness of those prompts. That means the human team can move straight into a quality‑control phase where humans audit and refine the AI’s work instead of doing traditional first‑pass review.

Because GenAI is handling the first‑pass decisions rather than just re‑ranking a human review queue, the human role shifts from “primary reviewer” to “expert checker.” 

Teams focus their time on validating recall and precision, spot‑checking edge cases, and tuning prompts, and resolving nuanced calls like privilege. Once the prompts perform above the necessary quality level, the model can then apply the optimized prompts across the full population at machine speed.

Advantages: The speed potential is dramatic — large document sets that would take weeks or months with human reviewers can be processed in hours once the prompts are optimized. Because you're writing prompts rather than coding training examples, you can quickly adjust your criteria or add new issue tags without retraining a model. GenAI also provides explanations for its decisions on every document for every tag, which can help reviewers understand why a document was tagged a certain way. For lean teams facing tight deadlines, it allows senior attorneys to focus their expertise on QC and judgment calls rather than first-pass review.

Disadvantages: As the newest approach, GenAI has less established legal precedent than TAR methodologies, though this is evolving rapidly. (For example, The Sedona Conference published a draft of their “Primer on GenAI in Discovery” stating that "For GenAI-assisted review, the validation methodologies and metrics used in TAR offers a vetted framework that can be applied ... to validate that the results meet requirements of reasonableness and proportionality.")

The quality of results depends heavily on prompt optimization; poorly written prompts produce poor results. GenAI also requires thoughtful validation to ensure the model is performing as expected, particularly for nuanced issues. Depending on pricing models and document volume, it may not always be the most cost-effective option for smaller or more straightforward matters.

⚙️Dig a little deeper: Learn how to use Generative AI for document review in this guide.

Case study

When a Fortune 500 company faced a high-stakes breach of contract lawsuit, a top litigation boutique found themselves with a compressed timeline after the court unexpectedly ordered substantial completion of discovery in just five weeks. The lean team faced 45,824 documents with minimal time for quality control.

The firm partnered with DISCO to implement Auto Review, a generative AI solution. The case team collaborated closely with DISCO's AI Consulting team to fine-tune prompts for responsiveness and seven issue tags. Using the optimized prompts, the review was completed in just 2 hours and 12 minutes, achieving 95% recall and 91% precision, metrics that significantly exceeded typical human review benchmarks.

As the senior associate leading the case explained: "No matter how good a review team is, there's no substitute for the lawyers closest to the case — but it's not feasible for my team to review 50,000 documents in a month. With Auto Review, we could more effectively harness our expertise, working directly with DISCO's AI consultants to hone the tag prompts and focus our time on QC."

Your options: The big picture

All four workflows live on the same spectrum. What changes is who does first-pass review and how formally you decide what documents don’t get human eyes.

Prioritized linear review

AI scores the whole corpus and serves documents to humans in score order, so reviewers always see the most likely responsive items next. The workflow is still fundamentally review‑driven: teams keep reviewing until they have reviewed all of the documents.

TAR 1.0

Reviewers train a model in a separate, front‑loaded phase, then use the validated model and a fixed cutoff score to decide which slice of the corpus will ever be reviewed. Above the cutoff, humans review or QC. Below the cutoff, documents are typically not reviewed at all, based on the model’s performance and sampling.

TAR 2.0 (CAL)

TAR protocol with no distinct training phase: reviewers begin coding, the model learns from each decision, and AI continuously reprioritizes the queue as you go. Over time, the same continuous learning that drives prioritization also produces the metrics and sampling needed to define a defensible stopping point and to identify the low‑scoring tail that will not receive human review.

GenAI review

Large language models perform AI‑first review by directly applying responsiveness and issue tags (with natural‑language explanations) across the corpus, based on your written instructions. Humans shift into a QC and exception‑handling role, validating recall and precision and resolving edge cases, rather than coding document by document for first pass.

📚Additional Reading: TAR in the Age of GenAI walks you through the evolution of document review technology, from linear review to GenAI.

Step 3: Map costs and resources

Understanding the true cost of each approach requires looking beyond licensing fees. For each tool being considered, teams need to account for licensing and hosting fees, reviewer hours at various levels (first-pass, senior review, QC), training and ramp-up time, validation and quality control processes, and potential re-work if issues arise.

The critical calculation is total cost of review versus cost of the tool. 

A more expensive technology solution that dramatically reduces reviewer hours may deliver significant overall savings. Conversely, a sophisticated AI tool may be overkill for a straightforward matter with a modest document population.

Consider the Kennedys example. By investing in DISCO's technology and reducing their review population to just 1.85% of collected documents, they achieved cost efficiency that would have been impossible with traditional approaches. Similarly, the litigation boutique using Auto Review was able to complete in hours what would have taken their lean team weeks, a trade-off that made sense given their timeline pressure.

💡Did you know? This article reviews four ways an outdated review process can inadvertently drain your budget and impact profitability.

Step 4: Plan your timeline

Different tools impact your timeline in different ways. Prioritized linear and TAR 2.0 allow you to start reviewing almost immediately, with the AI learning as you go. TAR 1.0 requires more upfront investment in seed set development and training rounds before the model becomes effective. GenAI requires prompt development and testing, but once optimized, can process documents at remarkable speed.

When planning your timeline, consider: 

• Ramp-up time for each approach
• Expected review velocity once fully operational
• Time needed for quality control and validation

Then buffer for unexpected issues or scope changes.

The right balance between speed and accuracy depends on your specific situation. A matter with an immovable court deadline may require prioritizing speed, accepting some trade-offs in precision. A high-stakes investigation where missing a single document could be catastrophic may warrant a more thorough, slower approach.

Final checklist: Choosing the right tool for the job

Key questions to ask before choosing

1. What is our document volume, and how does it compare to our available reviewer resources?
2. What is the nature of the documents? Do we expect a meaningful percentage of responsive documents in documents that are not text-based (e.g., images or videos).
3. Do we anticipate a lot of foreign-language documents?
4. How aggressive is our timeline, and which approaches can realistically meet our deadlines?
5. What is our budget, and what's the total cost of review (not just technology cost) for each option?
6. Are there any stakeholder concerns about AI-assisted review that we need to address?
7. Do the opposing counsel or the court disfavor using advanced technology for review?
8. What level of defensibility documentation do we need?
9. Do we have the internal expertise to manage this review, or do we need partner support?

Building a flexible toolkit

The most successful legal teams don't commit to a single methodology. They build familiarity with multiple approaches and select the right tool based on each matter's unique requirements. This framework provides a repeatable process you can apply to every matter.

As Kristin Zmrhal, DISCO’s Vice President of Product Strategy, advises: "Don't forget that there's a people part of this. Find the right people, the right champion at your firms, and bring it together with the expertise. While the technology is great, you have to know how to leverage it."

Scoping smarter for review success

There is no one-size-fits-all solution for document review — and that's precisely the point.

It's tempting to chase the newest technology, especially when the legal industry is buzzing about generative AI. But the firms achieving the best outcomes aren't defaulting to any single approach. They're scoping each matter thoughtfully, weighing their options honestly, and selecting the methodology that fits their specific goals.

Sometimes that means leveraging GenAI to complete a 45,000-document review. Sometimes it means using TAR 2.0 to reduce 1.4 million documents to a manageable population. And sometimes it means prioritized linear review to satisfy a client who needs human eyes on everything.

The technology is only as good as the strategy behind it. By following this framework — scoping thoroughly, understanding your options, mapping costs and resources, and planning your timeline — you can approach each matter with confidence, knowing you're choosing the right tool for the job, not just the newest.

Ready to find the right review approach for your next matter? DISCO's AI Consulting and Professional Services teams can help you evaluate your options and build a review strategy tailored to your specific needs. 

Learn more about Cecilia AI or schedule a demo to see DISCO in action.