AI for Legal: Making Sense of the Hype

John McCarthy, the computer scientist and “father of AI,” defined artificial intelligence as the science and engineering of making intelligent machines, especially intelligent computer programs.

When AI gets mentioned in the context of enterprise legal applications, it is usually referring to “machine learning.” In machine learning, systems learn from outcomes and decisions and improve with experience without being directly programmed to take certain actions or reach specific conclusions. These machines analyze data and discover patterns without significant human intervention, typically requiring only a training dataset.

Machine learning is often confused with rules-based automation, workflows based on pre-programmed “if this then that” algorithms. Legal buyers need to recognize the difference when looking to deploy AI within their departments. If the machine isn’t analyzing and learning from the data but is using pre-programmed, non-evolving rules to automate processes and outcomes, then it’s not AI.

Legal teams usually bring in machine learning to improve efficiency and productivity, as machines can perform tasks faster than humans, freeing legal counsel to do higher-value work. These applications include:

  • Legal Research: reviewing, tagging, and ranking documents relevant to a matter or eDiscovery, highlighting questionable ones that need human review.
  • Contract Review: Identifying and flagging clauses for review, searching for missing clauses, and redlining in bulk and at speed.
  • Invoice Review: Coding, approving, rejecting, or flagging line items and invoices (where rules-based automation isn’t an option.)
  • Data Extraction: This can apply to invoices, contracts, documents, or any requirement where a mass of non-structured data must get organized and classified.
  • Litigation analytics: Analyzing trial data to predict outcomes of litigation.


The above use cases and benefits can transform the legal profession. However, legal departments currently implementing an AI-powered legal solution may be disappointed by the true scope of these tools, especially if they are at the start of their digitalization journey. Buyers do not see the promised benefits and are beginning to question the hype.

The very nature of machine learning is that it needs data to deduce the patterns that help it to evolve and learn. This data doesn’t just need to be abundant in volume; it needs to be complete, accurate, fair, and free of bias. Improved accuracy vs. a human is a benefit often touted, but this is only the case if the data from which the machine is learning is accurate in the first place. Poor or insufficient data will mean the machine does not have enough data to learn from and will not fully deliver the anticipated outcomes and benefits.

Perhaps even more concerning, however, is that the machine will draw partial or incorrect conclusions from a deficient dataset and take the wrong action or reach the erroneous conclusion – thereby creating hidden risk. Ironically, AI can negatively impact productivity if a human must go back over the work, identify issues, and correct them. More severe, though, is if these incorrect conclusions result in damaging actions for the business, even litigation. The reliability of your machine learning needs to be a factor when accounting for legal risk, and legal teams need to understand their role in feeding machine learning tools with quality data and training to avoid these issues; as the saying goes, “you get out what you put in.”

If this sounds paranoid, some examples from other industries will help show why it is critical to be careful when deploying AI. In 2018, Amazon created a tool to review engineering CVs and flag the top ones for an interview. The intention was to automate a time-consuming process. To train the machine, they used the dataset of current Amazon engineering employees plus applications from the last ten years, which happened to be predominantly men. The machine “learned” that ‘more male’ candidates were the best for the role. Amazon soon ditched the tool. Poor data was also at the heart of IBM Watson’s failure to accurately diagnose and treat cancer patients. The data used to train the machine was hypothetical rather than real patient data and frequently gave poor advice. These examples demonstrate not only the importance of complete data for machine learning but the fact that it is hard to predict unexpected consequences before they happen.

The above examples demonstrate the importance of quality and unbiased data, even when the aims are straightforward. AI is not for complex legal work; it speeds up routine tasks, supports better decision-making, and sometimes takes actions based on those decisions. In fact, some of the best examples of AI deployment are where machine learning tools have been combined with rules-based systems to first identify and categorize data and then take defined steps based on that categorization.


Spend management is a legal-specific application using rules-based automation and machine learning together. For example, Onit’s European legal spend management solution BusyLamp uses the following AI functionality for clients and/or law firms that prefer not to use LEDES files:

  • Data extraction: Pulling relevant information from PDF invoices, relieving smaller law firms from the burden of generating complex invoice files.
  • Invoice Reviews: Some law firms struggle to code invoices in a way that clients can understand. BusyLamp AI takes unstructured invoice data and auto-classifies every task to enable automated invoice review.
  • Legal Analytics: Unstructured invoice and matter data can be analyzed to enhance strategic decision-making.
  • Block Billing: English time narratives can be analyzed so that block billing, a practice that usually contravenes billing guidelines, can be identified.


Using point solutions such as the e-billing example above allows legal departments to take advantage of machine learning benefits for gains in specific areas of legal operations. But machine learning is by no means critical to make efficiency and productivity gains; most BusyLamp clients start small and aim big by tackling the issues of collating knowledge, structuring, and cleansing their data sets, and then building automated workflows.

When you gather requirements for your next legal technology project, start by mapping out your current processes, roadblocks, and desired outcomes before looking at any specific technology tool. As you evaluate software vendors, you will discover various solutions and workflows to your problem, which may or may not involve AI.

Remember, you should never use AI for AI’s sake – it is rarely the silver bullet. Almost every legal technology tool uses rules-based (non-AI) automation to relieve the legal team of admin and mundane, repetitive tasks; this will be a fantastic starting point for most teams setting out on their digital journey.

There is no doubt that machine learning is playing a huge role in improving the productivity of the legal profession and will allow in-house teams to take a more pivotal, strategic role in their businesses. But as a profession familiar with risk mitigation, a degree of caution must be applied when looking to reach the machine learning “promised land.” Accurate, high quantities of data alongside a careful selection of technology tools will significantly reduce your exposure to these risks and help you make a success of your team’s digital transformation.

Because AI is so dependent on the data it receives, the real transformational tipping point will not be in using these solutions within the legal function alone but in the enterprise-wide application of machine learning tools. Imagine the insights and outcomes achieved by analyzing documents and data across an entire organization, not just the legal function. This is only achievable with integrated legal and enterprise tech tools and robust, extensive, consistent data.

The “power of AI” and its ability to change the legal profession are beyond question. However, it is essential to proceed with caution and lay the groundwork to ensure that your legal department sees the benefit of machine learning rather than learning that it has been sucked in by the AI hype machine.

Request a demo of BusyLamp today. 

Thank you for subscribing!