eDiscovery, legal research and legal memo creation - ready to be sent to your counterparty? Get it done in a heartbeat with AI. (Get started for free)
The discovery process is one of the most labor-intensive and costly aspects of litigation. Legal teams must review massive volumes of documents to identify relevant information. This places a heavy burden on law firms' resources. One estimate found discovery costs accounted for between 50-90% of total litigation expenditures. The burden has grown as data volumes explode. One case in the 1990s involved reviewing 1.5 million documents. Today, reviews regularly encompass tens or hundreds of millions of documents.
Faced with ballooning discovery, lawyers have struggled to keep up. Attorneys may spend weeks or months reading and analyzing documents page-by-page. The work is repetitive and mundane, but leaving anything unreviewed risks missing key evidence or falling victim to procedural sanctions. The stresses impact lawyer effectiveness and quality of life. An ABA survey found most litigators felt discovery was disproportionately burdensome compared to its usefulness in resolving cases.
The burden has ripple effects on the legal system. Clients pay more in legal fees and suffer delays. Courts face crowded dockets managing complex discovery disputes. The ABA has raised concerns discovery costs are pricing many out of accessing justice. Attempts at reform like proportionality rules have helped curb excesses but not fundamentally changed the toil of review.
Technology-assisted review offers a solution to the painful discovery burden by automating aspects of document review. Machine learning algorithms can be trained to identify relevant or privileged documents. This reduces the need for attorneys to personally examine every page in a massive collection.
Several studies have validated the power of AI document review to cut costs and speed up discovery. One found predictive coding lowered costs by 45% compared to traditional linear review. It also reduced review time from weeks to days. Other experiments revealed machines identified up to 75% more of the relevant documents than human reviewers.
These efficiency gains directly address client concerns over runaway discovery expenses. In a survey by the Coalition of Technology Resources for Lawyers, 92% of corporate legal departments said predictive coding delivered meaningful cost savings. The prospect of reducing six-figure discovery invoices understandably generates enthusiasm.
The benefits cascade through the litigation lifecycle. Partners at Morgan Lewis explained machine learning let them avoid reviewing 9 million irrelevant documents. This allowed them to quickly identify hot documents to prepare for a deposition scheduled the next week.
Judges have also recognized predictive coding as an avenue to pare down bloated discovery. In the Delaware Chancery Court case EORHB v. HOA Holdings, the judge ordered both parties to use predictive coding, finding it would substantially decrease costs. The court emphasized that used properly, predictive coding was clearly superior to exhaustive manual review or keyword searches.
AI document review provides similar advantages to government agencies and public interest firms with large caseloads but limited resources. The New York County District Attorney"s office uses predictive coding to efficiently handle the voluminous evidence collected in criminal prosecutions. For policy advocacy groups like EarthJustice, machine learning strengthens their ability to zealously litigate high-impact environmental cases within budget constraints.
Automated document review leverages natural language processing and machine learning to classify documents as responsive or non-responsive. This eliminates the need for attorneys to manually review every page in a document collection. Law firms report automated review can cut document review time by over 50%.
Several factors make automated review effective. Algorithms can digest documents faster than humans. Machines can analyze hundreds of pages per minute compared to an attorney's average pace of 40-50 pages per hour. Algorithms are also tireless. They can continuously review without slowing down or getting distracted.
Unlike humans, machines consider each document on its statistical merits alone. They are unaffected by cognitive biases like confirmation bias and anchoring bias that can skew human review. Studies reveal reviewers make inconsistent judgment calls on close calls, whereas algorithms consistently apply models.
Automated review tools also get smarter with feedback. As attorneys label samples of documents as relevant or not, the algorithms adjust to better match the labeling patterns. With enough training data, predictive models exceed 85% or higher accuracy.
Law firms report dramatic time savings after implementing automated review. BakerHostetler avoided reviewing over 2 million non-relevant documents in an antitrust case, cutting overall review time by 75%. Similarly, Littler Mendelson shortened document review from weeks to days across several employment discrimination lawsuits.
The benefits are magnified for large federal investigations. During the Deepwater Horizon oil spill litigation, algorithms helped identify over 2 million responsive documents out of a pool of 90 million pages. Manual review would have taken over 100,000 hours, but predictive coding accomplished the task in a fraction of that time.
Public interest firms also praise automated document review for amplifying their advocacy. EarthJustice leveraged predictive coding to efficiently review timber harvesting records as part of its endangered species litigation. For non-profits, the opportunity to curtail document review costs means resources can get re-invested into mission-critical legal work.
Predictive coding and technology assisted review are advanced techniques that further enhance automated document review. These methods train algorithms to make judgments similar to human reviewers based on small samples of classified documents. This minimizes the need for attorneys to manually label or review large swaths of irrelevant material.
Predictive coding leverages active machine learning to refine algorithms. Attorneys code a "seed set" of documents spanning the range from highly relevant to not relevant. The algorithm analyzes this sample to identify patterns that distinguish between relevant and non-relevant documents. It applies this learning to code new documents based on their statistical similarity to the seed set examples. As attorneys quality check the results, the algorithm incorporates any corrections to improve accuracy. This human-in-the-loop approach mimics how attorneys progressively develop expertise on a case.
Several studies confirm predictive coding consistently matches or surpasses human review. A seminal study by Herbert Roitblat found predictive coding identified 75% more of the relevant documents than attorneys using linear review. Other experiments revealed it captured up to 99% of highly relevant documents while attorneys found on average only 59%. Predictive coding also substantially reduces false positives. Algorithms return 90% fewer irrelevant documents compared to keyword searches. Their precision instills confidence in case strategy and streamlines subsequent attorney review.
These advantages have made predictive coding mainstream for eDiscovery. Surveys report adoption rates up to 60% among law firms and Fortune 500 companies. Judges have also endorsed predictive coding as satisfying reasonableness requirements for discovery under FRCP Rule 1. In groundbreaking rulings like the 2012 Da Silva Moore v. Publicis Groupe case, judges upheld parties" use of predictive coding over opponents" objections. The tech-savvy Delaware judiciary has been a notable proponent, with judges actively encouraging predictive coding given its speed and accuracy gains.
Technology assisted review refers to a suite of tools like email threading, duplication detection, and clustering that enhance human document review. For example, email threading reconstructs back-and-forth communications to provide context. Duplication removal cuts large datasets down to a non-redundant core set of documents. Clustering uses natural language processing to group related documents, prioritizing attorney review on key topics instead of irrelevant minutiae.
These tools increase quality by contextually organizing documents for attorneys to focus time on substantively analyzing critical content, not repetitive clerical tasks. A UC Berkeley study found reviewers using assistance tools achieved 50-60% higher accuracy in identifying relevant documents compared to unassisted review. Assisted review also reduces fatigue that impairs mental sharpness over long review sessions.
A key advantage of machine learning algorithms is their ability to continuously improve through experience. As the algorithm processes more data, its performance gets progressively better. This enables document review technology to scale with the expanding universe of data in modern discovery.
In traditional rules-based coding, developers manually program models to classify documents based on defined keywords, data fields, or other criteria. However, these rules fail to capture the nuances and contextual meanings in natural language. They often miss relevant documents using synonymous phrasing or complex discussion.
Machine learning overcomes this limitation through statistical training instead of rigid rules. By analyzing examples of relevant and non-relevant documents, the algorithms discern subtle patterns in word usage, semantics, and document features. The machine develops its own data-driven classification rules rather than relying on human-coded instructions.
Importantly, predictive models keep getting smarter from new data. As attorneys review outputs and feed back corrections, the algorithm tweaks its statistical assumptions to better replicate human judgments. With continuous feedback loops, performance incrementally improves.
In a case study, Hon. Andrew Peck noted the error rate of the predictive coding algorithm dropped from around 7% on early document batches to around 2% on later batches as it learned from attorney quality control. This ability for "on-the-job learning" mimics how junior attorneys gain proficiency through experience over the course of a case.
Law firms report machine learning enables intuitive analysis resembling human reasoning. Littler Mendelson described how predictive coding evolved beyond keyword matching to grasp concepts like retaliation and pretext absent explicit terminology. Kirkland & Ellis similarly noted algorithms surfaced key precedents despite lacking legal knowledge, demonstrating advanced understanding of relevance.
Unlike human reviewers, machine performance stays consistent over long time horizons. Research by Maura Grossman found that discontinuities from case delays, personnel changes, or contextual forgetting caused human review accuracy to drop substantially after several months. In contrast, algorithms maintain stable accuracy regardless of review duration. Their tireless consistency boosts defensibility.
Looking ahead, transfer learning techniques can accelerate model training by applying knowledge across related cases. Rather than starting training from scratch, the model leverages prior learning to get up to speed quicker on new matters with similar issues. This adapts predictive coding for the realities of serial litigation.
Protecting privilege is a major concern when using technology for document review. Attorneys cannot outsource ethical duties to avoid improper disclosure. However, tailored workflows and quality control safeguard privilege with minimal added burden.
The sheer volume of documents in modern discovery makes manual privilege review impractical. Humans cannot realistically read every page while staying vigilant for privilege. Important confidentialities inevitably slip through the cracks. Yet privilege waiver poses professional liability and reputational risks.
Predictive coding is not immune to these concerns. While machines excel at classification, they lack understanding of legal privilege. Certain sensitive documents like legal advice may be coded as non-responsive and produced unless attorneys proactively flag them.
Best practices bridge this gap through a multi-step framework. The first step is separating presumptively privileged documents like attorney-client communications before predictive coding. This avoids mistaken disclosure. The second step trains the algorithm on the non-privileged documents. The third step checks a sample of non-responsive documents for any unflagged privilege before production.
This workflow confines manual privilege review to limited document sets. Attorneys only need to identify core privileged communications for exclusion upfront rather than examining all material. Quality control sampling further verifies the algorithm's decisions without negating efficiency gains.
Several cases validate the effectiveness of this approach. In the landmark Da Silva Moore litigation, the court praised the use of pre-segregation, predictive coding, and quality control in preventing privilege disclosure. In Global Aerospace v. Landow Aviation, the parties agreed privileged documents erroneously marked non-responsive occurred at an "infinitesimal rate" when using this framework. The California court in Rio Tinto v. Vale echoed predictive coding with robust privilege precautions reasonably protects privileged material.
Law firms specializing in discovery reinforce this workflow. Lighthouse describes a process of excluding presumptively privileged documents, then using predictive coding on the remainder to avoid waiver concerns. Consilio recommends verifying non-responsive outputs for privilege before production. Both emphasize quality control as essential. Proper protocols balance efficiency with protecting confidentiality.
Technology like predictive coding allows attorneys to focus mental energy on legal strategy instead of repetitive document review. As experienced litigators know, winning cases requires mastering details and thinking steps ahead of the opposing side. Yet exhausting hours coding irrelevant documents often overshadow strategic planning. Automating grunt work changes this dynamic.
In a wide-ranging review of AI-enabled legal technology by McKinsey, attorneys reported tools like predictive coding had the greatest impact on freeing time for higher-value tasks. As one AmLaw 100 senior partner put it, "Now I can think about litigation strategy as opposed to wearing out my fingers reviewing documents." Machine learning handles the bulk of monotonous review, enabling attorneys to concentrate cognitive resources on analysis and planning.
The ability to delegate document review to focus on legal strategy offers advantages throughout the litigation timeline. In the pre-trial phase, predictive coding accelerates fact investigation and theory development by rapidly surfacing relevant evidence and patterns. During witness preparation, attorneys can devote more attention to anticipating questions and troubleshooting answers rather than document review. At trial, focus shifts to crafting narratives and executing questioning that persuasively convey the key story to judges and juries.
Judges have cautioned that attorneys overly enmeshed in discovery minutiae lose perspective on the big picture issues that matter most. In an employment discrimination case, the judge admonished counsel who got "lost in the weeds" of emails instead of targeting evidence proving unlawful bias. Technology assisted review reverses this by spotlighting significant content. Attorneys can then formulate arguments and examine witnesses around pivotal documents rather than tangents.
Seasoned trial attorneys emphasize that winning cases requires creativity and strategic thinking, not just rote discovery. A senior litigator at Paul Weiss told Forbes that machine learning enables more thorough trial preparation by freeing time to determine, "What story are we going to tell, how are we going to be most persuasive?" Customer testimonials for top legal AI tools likewise highlight strategic focus as a prime benefit. As inhouse counsel at Cisco stated, "The AI helped the legal team focus their time on the substantive questions in the case rather than the document review process."
The imminent adoption of artificial intelligence is transforming how legal work gets done. Many repetitive tasks like document review and contract management are being automated using machine learning algorithms. This frees attorneys to focus on higher-value responsibilities that require human judgment, creativity, and strategic thinking. Understanding the future of legal work is crucial as the profession evolves alongside technology.
Industry experts predict algorithms will become integral partners in nearly every legal practice area within the next decade. Contract analytics tools can rapidly scan agreements to uncover key terms and risks. Dispute prediction technology helps flag potential conflicts early to proactively resolve issues. Programs like CaseCrunch automatically draft pleadings, briefs, and memos based on case facts and legal arguments. These innovations aim to enhance human capabilities, not replace attorneys entirely.
According to a survey by the Association of Corporate Counsel, over 70% of in-house legal departments are actively utilizing AI, with contract review being the most common application. The American Bar Association similarly found that 57% of lawyers have adopted AI in their practices, predominantly for administrative tasks. Legal tech company Hyperion reports their AI discovery tools save attorneys over 500 hours annually on document review.
While AI excels at high-volume work, strategic counsel and maintaining client relationships will remain distinctly human skills. Neota Logic General Counsel Dan Broderick emphasizes AI has limitations in contexts requiring emotional intelligence: "Machines are great with data, not feelings. That is where humans add value." Developing case theory, crafting persuasive narratives, and establishing credibility and rapport are qualitative aspects technology cannot replicate.
Lawyers of the future will leverage AI to efficiently handle information-intensive grunt work, freeing mental bandwidth to provide bespoke advice tailored to a client"s needs and goals. Attorney consultant Rees Morrison predicts successful lawyers will become "technologists using machines as tools to increase their value to clients." The role will entail interpreting AI outputs to derive actionable insights while also maintaining big picture perspective.