Introduction
The legal industry, previously known for its traditional practices and resistance to change, has entered a period of rapid technological development. Tools intended to streamline and enhance various aspects of legal work have become widespread. A trend that started with eDiscovery, document management, legal research and billing software has now seen the emergence of advanced CLM management programmes, generalist document automation platforms, specific workflow tools, specific industry solutions and much more.
Legal tech companies have struggled over the years with low adoption rates, as lawyers are loath to shift from the familiar duo of Microsoft Word and Outlook. But since the advent of Large Language Models (LLMs), and in particular the launch of ChatGPT, it’s no stretch to say that Legal AI has become one of the most exciting categories in the tech world. As stated in an EDRM article, “there is no doubt that AI platforms and products are a boon for legal professionals that cannot be ignored”. And the stats back it up, too. A recent Clio study found that usage of AI by legal professionals has increased from 19% in 2023 to 79% in 2024. This leap is testament to the waning scepticism of law firms and the vast appetite to experiment in the legal world. This is likely in part triggered by the changing preferences of clients, with an increasingly clear use case for AI in legal (the same Clio study found that almost three quarters of a firm’s billable tasks are potentially exposed to automation by AI). One fact, however, is clear: no lawyers, whether in private practice or in house, want to be seen as late adopters.
Before we go on, it is worth defining just what generative AI is:
Glossary of AI terms:
- Generative AI: AI systems that create new content—like text, images, and music—by learning from patterns in existing data.
- Machine Learning (ML): A subset of AI where algorithms learn from data and improve over time without being explicitly programme
- Large Language Model (LLM): AI trained on vast amounts of text data to understand and generate human-like language. GPT and BERT are examples.
- Training Data: The dataset used to teach an AI model. In generative AI, this data often includes text, images, or audio.
- Fine-tuning: Adapting a pre-trained model to specific tasks by training it on additional, focused data.
- Prompt: Text or input given to a generative AI model to guide the type of content it produces.
- Token: Small pieces of text that make up words, used by language models to process and generate responses.
- Natural Language Processing (NLP): The AI field focused on enabling computers to understand, interpret, and generate human language.
When generative AI first appeared in most of our lives in November 2022, the most obvious use cases revolved around the creation, management and development of contracts. For firms or in house teams without the templates of say, a Magic Circle corporate team, the allure of an AI generated instrument has been undeniable. Slower to adopt have been our learned friends themselves, disputes and in particular, litigation teams.
Litigators typically spend countless hours carrying out the most time-consuming, detail-oriented tasks, ranging from rote document review to drafting court pleadings. Given the complex fact patterns, evolving legal arguments and procedural requirements associated with litigation, generalist AI tools have struggled to cut through with disputes specialists who have often spent years honing their craft.
Recently, however, things have started to develop, with the advent of new tools and existing platforms adding generative AI functionality. The balance is beginning to shift, with even members of the judiciary such as Lord Justice Birss and the Master of the Rolls, Geoffrey Vos, extolling the benefits of this technology to make court processes more efficient.
Lord Justice Birss, having used ChatGPT to help write a judgment, proclaimed that he found it to be “jolly useful”. Vos, while cautioning that such technology must be used carefully, heralded the changing attitudes among the highest echelons of the English court system when he predicted that “AI can and will enable us to develop a digital justice system that is efficient and accessible for all.”
We realise that navigating your way through the morass of tooling can be tricky, so we have penned a two part series: Part One analyses the AI for disputes market, and Part Two sets out a buyer’s guide on how to procure the right tool for your use case.
PART ONE
Early Case Assessment and Case Theory
The very beginning of a litigation, or the moment just before a litigation is commenced, can often be the most challenging time for a legal team. Whether prosecuting or defending, both sides are likely to be mired in documents, new to the facts and hounded for estimates of costs, time and risk.
To help shape their strategy and provide these estimates, most lawyers will conduct a form of early case assessment (ECA). They will collect, review and analyse data pertaining to the case to determine the key facts, potential approaches and risks. This often involves carrying out extensive and costly document review exercises. However, ECA is instrumental in providing clients with a realistic perspective on their case and informing important aspects of their strategy, in particular the decision on a possible settlement. As such, while it may bear higher upfront costs, ECA often ultimately leads to significant cost savings.
Without conducting proper ECA, there is a risk that important decisions are taken without sufficient regard for the likely consequences. Many high-profile examples serve as a warning against rushing ahead without knowing all of the facts: in the aftermath of scandals such as the BP Deepwater Horizon Oil Spill and Volkswagen Dieselgate, the failure to properly estimate the scope of the damage and consequent liabilities led to protracted legal battles and significant reputational harm.
Given the high volume of paperwork and repetitive, review-based work, ECA is clearly an ideal candidate for AI automation. Some legal tech companies are attempting to build solutions aiming to expedite and streamline data review, predict case outcomes and assess risk. These firms also appear to be seeing results - according to Gartner, lawyers using AI-driven tools have seen a reduction of 50% in the time taken to complete ECA. In theory, the use of AI-powered tools designed for ECA promises smarter strategic decision-making, reduced costs overall and, ultimately, happier clients.
While data review may be a central part of any proper ECA process, there is one crucial step to complete before this: establishing and verifying the key facts of the case. A solid understanding of the relevant actors, dates, events and times is the bedrock of building any useful search terms or convincing case theory.
Wexler provides a comprehensive fact management platform, first extracting facts from documents and then providing an analysis of their significance. Critically, Wexler only looks at the documents you give it, and has been trained extensively on what constitutes facts in the context of legal disputes.
Once Wexler has identified the key facts, it structures them into a table, from where lawyers can build chronologies. As a barrister’s old adage goes: ‘understand the chronology, understand the case’. You can create dramatis personae, run targeted natural language searches and extract data to create reports. For example, ‘extract every time John Smith is criticised, and who by, in table format’. At the heart of Wexler’s approach to ECA is Kim, the specialised AI agent for multi step tasks. Users can instruct Kim to answer complex questions, create documents and much more.
A key part of ECA is of course testing your hypotheses and building case theories. Wexler’s functionalities are uniquely well-suited to assist with this. A good case theory helps a legal team structure arguments and dictates how the team will go on to present their client’s case, acting like a road map for the whole matter, all the way to trial (and beyond). Traditionally, constructing a case theory requires extensive document review, legal research, and factual investigation, which can take significant time and resources. Using Wexler helps get a grip on the facts and build the narrative, ultimately putting you in the best position versus the other side.
Legal Research
Every junior lawyer and/or paralegal in a litigation team is likely to spend a portion of their time conducting legal research. Particularly in common law jurisdictions, staying on top of case law developments throughout the span of a case is important to presenting a competent claim or defence. Cases are decided either on a point of fact or a point of law, and though the legal question is usually secondary to the factual one, having a grip of the case law is still vital.
Historically, this has meant trawling through case law databases and other dispersed resources, scanning lengthy, complex judgments to find relevant paragraphs and translating wordy legalese into client-friendly language. Legal research is usually left within the purview of junior team members on the basis that more senior lawyers don’t have the time. However, identifying relevant case law, understanding complex judgments and explaining research findings in a useful way is no mean feat for inexperienced lawyers, and can have a significant negative impact on a case if not properly undertaken.
The technological evolution of legal research is well underway, with platforms such as Thomson Reuters’ Westlaw Edge leading the charge. Westlaw has now incorporated AI algorithms into its case law search platform to deliver "WestSearch Plus," which provides suggested search results as lawyers type their questions in natural language. Other platforms offering a similar service include Lexis Advance and Casetext, which has now been folded into TR's offering since its acquisition.
The benefits of using these tools are already apparent - according to Casetext, lawyers using CARA reported a 20-30% reduction in research time compared to standard legal databases. Similarly, Thomson Reuters reports that law firms using Westlaw Edge experience a 24% faster completion of research tasks, highlighting the gains from AI-driven research.
However, legal research has been beset by problems, particularly around the infamous ‘H’ word: hallucinations. In June 2024, a group of researchers at Stanford University found high hallucination rates in two market leading legal research generative AI tools, LexisAI+ and Westlaw Precision. Rearchers found that Westlaw's tool hallucinated nearly twice as often as Lexis+ AI—with Lexis+ AI hallucinating 17% of the time, and Westlaw hallucinating 33% of the time. Clearly in high stakes litigation work, even 1% of the time is not acceptable, so a form of Ask Jeeves for legal research may be some way off. To be clear, there were rebuttals from Thomson Reuters and LexisNexis about the methodology of the study, seeking to clarify the rate was much lower. What is apparent, however, as has been pointed out by legal AI experts like Richard Tromans, is that we need benchmarks, standards and baselines to compare legal tools, especially when applied over such vast databases as the corpus of case law of an entire country.
E-Discovery
As one of the most time-consuming (and crucially for clients, most eye-wateringly expensive) parts of a typical litigation, e-discovery was among the first clear use cases for legal tech. The initial disruptors in this space, such as Relativity, emerged to help law firms manage the increasing scale of e-discovery, offering search capabilities, document tagging and review workflows. They also introduced metadata analysis and basic deduplication features, helping to reduce the time spent reviewing redundant files.
A step forward was taken when technology-assisted review and predictive coding were introduced. These tools allowed algorithms to "learn" from human reviewers, helping them to identify relevant documents by applying machine learning to predict document relevance. They have steadily improved over time, without becoming second nature in many instances. Indeed, there have been many cases which resort back to human review after attempting to review with tools like these. These tools do use AI, but the traditional, ‘rules-based’ Machine Learning rather than generative AI.
Since the introduction of generative AI, however, e-discovery practices are starting to look different. Most significantly, perhaps, is the ability to now automate large portions of the document review process. Companies like Relativity, Everlaw and Disco offer tools that can quickly analyse vast datasets and automatically categorise, tag and prioritise documents. AI-driven e-discovery tools have been found to reduce the volume of documents requiring human review by 60-80%, drastically lowering costs for law firms. Unlike traditional keyword searches, more advanced NLP currently being used enables the AI to grasp the context and meaning behind the words in a document, surfacing documents that might not contain obvious search terms but are still highly relevant due to their content or the intent behind the language used. AI systems can also identify duplicate documents and near-duplicates, reducing the volume of data to be reviewed. Now a common feature in e-discovery platforms, continuous active learning (CAL) enables the AI to continuously refine its understanding of what constitutes a relevant document as the review progresses. As reviewers tag more documents, the AI's accuracy improves in real time, making the review process faster and more precise. Gartner estimates that AI-driven document review practices can cut 70% off the time spent reviewing documents, helping legal teams prepare for trial faster and more cost-effectively.
But these platforms are stymied in that they merely prepare documents for human review. It is hard to rip up the rulebook when the entire business model is built around (AI-assisted) review, conducted by humans. There is an opportunity for an gen AI native platform to ride the early adopter wave and launch an ‘agentic’ framework for the disclosure/discovery process. Most e-discovery experts do not believe that the current technology allows for end-to-end automation of the process, but with the advancements that allow systems to actually take actions independently, or with some oversight, rather than just returning content (in the co-pilot model), there is a huge chance for a new platform to build an ‘auto-pilot’ for the process, which plans out each step it will take, as a human does, before executing. This would look like an end-to-end flow starting with triage, categorisation and prioritisation of documents, filling out disclosure forms and then conducting the detailed second-level review, too.
Proving Statements
After pleadings are filed, or before the claimant files their claim, the parties to a litigation must determine how best to support their case theories using available evidence. For example, once a claim has been filed, the defending party must quickly assess, analyse and construct a response to each assertion made by the claimant. This is a painstaking task, requiring detailed analysis of the claim, extraction of the different statements in each paragraph, review of the supporting evidence and a slowly build out of responses based on the relevant evidence and interpretation. It is effectively a matchmaking process, going through each element of the claim to determine which piece of evidence within your possession or control best counters it.
While legal tech tools exist to address select parts of this process, there is currently no end-to-end solution through which lawyers can upload a counterparty’s pleading and quickly retrieve evidence they could leverage to contradict it. Wexler’s new feature, Wexler Proof fulfils exactly this need. Once a legal team has uploaded their case documents, the other side’s pleading and the other side’s supporting evidence to the Wexler platform, they will be able to use the Proof feature to analyse the arguments in the pleading, determine their strength based on all available evidence and retrieve evidence that can be leveraged to defend against those arguments. Ditto Witness Statements (Depositions), litigators will be able to extract claims and corroborate them with supporting evidence, or disprove them if evidence suggests to the contrary (or doesn’t exist at all).
Trial preparation
As every lawyer knows, the period just before a trial begins can often be one of the most hectic and paperwork-intensive moments in litigation management. On top of preparing witness evidence and skeleton arguments, legal teams must prepare and thoroughly check bundles containing thousands of pages of evidence from the entirety of the case. Bundle-prepping, while typically the kind of mundane and rote task entrusted to trainees, can cause significant embarrassment if done badly (a notorious example of this occurred during Gina Miller’s Brexit case, when the government's legal team mistakenly included duplicate documents in the trial bundle. This led to confusion during the hearing and prompted the court to point out the errors, highlighting the importance of meticulous bundle preparation in high-stakes litigation).
The first wave of legal tech tools designed to tackle some of these pain points will now be very familiar names for many in the industry - for example, Opus 2, which offers transcription services and the digitisation of trial bundles. Other companies, like XBundle, also addressed the evident inefficiency of bundle preparation. However, more players have been entering the space in recent years, and with the introduction of more advanced AI, the focus appears to be shifting.
One of the key growth areas is in predictive analytics - a number of legal tech companies now offer predictive litigation analytics services, which they provide by assessing a range of variables to predict case outcomes. For example, Lex Machina uses NLP and machine learning to analyse court rulings, identifying patterns in judicial decisions and lawyer performance. Ravel Law (part of LexisNexis) assesses judicial rulings and case trends in order to understand how judges rule on particular issues, thereby allowing lawyers to adjust their litigation strategies. Other key names in this space include Solomonic, CaseCrunch and Premonition AI.
Wexler is built for use throughout the life of a case. Its features, including help creating chronologies of key facts, dramatis personae, and a new feature, Wexler Extract, are used frequently in trial preparation. With Extract, litigators can extract searches and prepare reports. For example, you could create a report of every time your client has been criticised in witness statements, and use these to troubleshoot potential lines of attack by the other side. Advocacy is about knowing as much as you can about your documents, so there are no ‘unexploded bombs’ lurking which you don’t know about. Wexler is built for just that.
Conclusion
As the legal industry continues to evolve, the use of AI in litigation is becoming increasingly essential for modern legal practice. From ECA and e-discovery to case theory development and trial preparation, AI tools are revolutionising the way legal teams handle litigation, allowing them to work more efficiently, reduce costs, and improve their strategic decision-making. With platforms like Wexler, Everlaw, Relativity, and more, legal professionals can now leverage AI to streamline document review, assess case risks, build strong legal arguments, and prepare for trial in a more data-driven and efficient manner. In time, legal work is sure to be done to a higher standard and at a faster pace. As Erik Brynjolfsson, Director of Stanford's Digital Economy Lab predicts: “lawyers working with AI will replace lawyers who don’t work with AI.”
And, increasingly, generalist AI tools for lawyers won’t cut it in the high stakes fields of litigation. Litigators have skills that require years of training to perfect - and the same must be true for the tools. As Judge Artigliere observes: “In the legal realm, precision and specialised knowledge are essential, and lawyers are trained to make fine distinctions about legal precedent and laws”. These lawyers need the best quality AI to match their extremely challenging, specialist work. In the coming weeks, months and years, the stage is perfectly set for an AI native disruptor to dominate the litigation landscape, help law firms win more cases, drive more profits and increase access to justice by reducing overall costs.