AL Interview: Ravel and the AI Revolution in Legal Research

Finding legal data on court cases and judges’ opinions used to be the preserve of voluminous law libraries. If you wanted to get into a faster-paced search you could use the digitised resources of one of the massive legal publishers such as LexisNexis and Westlaw.

But, there were limits. The manual search in libraries was clearly out-dated even two decades ago. While the key word search type queries for the traditional legal data monopolies can provide a cross section of material to examine, but can leave the reader with many hours, if not days of work, to find what they are really looking for.

Now there is a new wave of legal data pioneers who see that by using better analytical technology, such as Natural Language Processing (NLP) and machine learning, they can provide legal researchers with a new approach that could hone in exactly on what they want to find. But, not only that, it could help to find nuances in legal opinions and judges’ behaviours that before would have taken weeks to investigate, or simply would never have been found due to the complexity of doing that manually.

One such company is US-based Ravel Law, which in 2012 spun out of Stanford University’s Law School, Computer Science Department, and the Institute of Design, with the support of CodeX (Stanford’s Center for Legal Informatics).

Dan Lewis, Co-Founder, Ravel Law

Artificial Lawyer caught up with its San Francisco, California-based CEO and co-founder, Daniel Lewis, to learn more about what Ravel had achieved so far.

Lewis explains that the idea to do something about legal research came to him after attending Stanford Law School. He also came from a legal family. From what he could see the law had fallen far behind the current level of what technology was capable of doing, especially in relation to legal research, which was after all just data.

‘So I reached out to computer science professors at Stanford because I didn’t have qualifications in that area. We then put together a team to look at legal research from a more data-driven approach,’ says Lewis.

But, there was one small issue that had to be overcome first: to find meaning in legal data you first had to have access to enough of it to be meaningful. And that was a problem because there are literally millions of US court decisions and access to them could mean having to pay for access to digital data companies before the project had even got started.

The team started off making use of whatever resources they had on hand, but the real breakthrough came recently in October 2016, after partnering with Harvard Law Library to digitize all 10 million court decisions it had a copy of. This was a massive project, and Ravel raised millions of dollars to fuel its growth.

Harvard Law School Library

So far so good, but you might ask: so what? Why bother with digitising the whole of the US legal corpus all over again when the big legal publishers already had done this? What value was Ravel going to add?

The short answer is: detail and the ability to derive meaning from that added level of detail. Not to mention a huge leap forward in time saving.

At present Ravel approaches US law from three perspectives: case law, judges’ behaviour and court variation.

Case analytics helps a researcher spot trends and the development of legal positions over time. It shows how often a particular case has been cited in other court filings, as well as on a page by page basis for each particular citation. It also shows examples of the language used and the wider context where following cases have cited certain parts of an earlier case. It can do this because of its use of NLP, rather than relying on key words.

Judge analytics helps to show how specific judges ‘think, write and rule’. Each US judge has a profile and this can be explored via keywords or motion type. You can then look at all relevant cases where the judge has given an opinion and analyse certain aspects quite precisely, such as the wording they tend to use in relation to, for example, motions to dismiss a case. In addition, Ravel can tell you what percentage of the time a judge grants or denies over 90 different motions. It even has compiled a complete CV of each judge that you can also search. This kind of granular analysis, that can isolate the language used in certain motions is delivered by machine learning that over time has taught the system what to recognise.

A new facility, Court Analytics, which was released last month (Dec 2016), compares forums and helps to assess possible outcomes by showing at a very granular level exactly what has taken place in each court in the US and the pattern of certain types of actions, covering over 90 different types of motion judges in those courts have passed.

This allows, in combination with the other applications, for lawyers to have a very real and pragmatic picture of the likelihood of certain legal arguments being successful in certain courts, and before certain judges, relative to the type of case. It is not a silver bullet, but it could reduce much of the uncertainty and reliance on a ‘gut feeling’ when litigating for a client.

screen-shot-2017-01-22-at-16-28-35Could all of this be done ‘manually’? Could all of this have been found and cross-referenced using traditional digital libraries? Yes. But, it would have taken much, much longer. In effect, Ravel is speeding up the analysis process of huge volumes of complex and unstructured legal data to Google speeds, which in turn allows a lawyer to spend time looking for more subtle discoveries that would normally not be attempted, such the success rate of a particular type of argument before a certain judge over the last 20 years.

But, what about the Holy Grail of litigation analytics: outcome prediction? Any insight into prior cases, use of language, motion patterns and judge behaviours gives a lawyer a degree of advance knowledge that improves one’s prediction of how the case will end. How far can this go? Could Ravel outright predict the results of a case even before the case has started?

Lewis is sceptical of total case prediction. ‘What Ravel can do is help a lawyer to build up a picture via multiple micro-predictions, for example with regard to a particular motion occurring based on past trends, or a specific argument being cited. But, it’s hard to predict the outcome of a big case,’ Lewis concludes.

One aspect of the Ravel interface at work

In fact, Lewis, despite his role as a co-founder of a legal tech company, is not wholly wedded to the idea that technology can answer every question in the law. He notes that the law is still in part an art and has not become 100% science that can be wholly reduced to patterns and logic.

‘For me, the law is not all about art, or all about science, it’s both,’ says Lewis and adds that the issue is perhaps about law catching up with science and technology, rather than law being replaced with technology.

So, what now? Lewis says that they will continue to sell the software to law firms and they already have a roster of Top 50 US law firms as clients. Ravel will also not remove the need for Westlaw and LexisNexis, which provide a very broad collection of legal information, but instead be an additional tool in a lawyer’s armoury for when they need additional insight.

The future it would seem promises further growth. According to crunchbase, Ravel has received over $9.2m in funding since its launch in 2012, with $8.1m of that coming in 2014. And although it does not make public its prices, as the software’s renown grows and more law firms use it, the more it will be able fund further growth and new innovations to its applications.

Given how massive the US litigation market is, which could be around half of the total US market’s $437 billion value, it would appear that Ravel is onto ‘a good thing.’ As Lewis adds: ‘About 30% of [litigation] lawyers’ time is spent doing legal research, but clients increasingly don’t want to be billed for this time and that means it often has to be written off.’

Cue Ravel, not just to meet the hope of providing clients with better litigation outcomes, but to also protect profit margins at a time of continued pressure on fees.