AI is a transformational technology that’s continuing to gain momentum in the legal, financial and professional services sectors. But many firms don’t yet have the internal knowledge or training to fully unlock its potential.
We’ve found the best way to ensure client success is to educate and build up experience inside the firm about how AI works and how to apply it to a broad spectrum of business problems, which is why we launched AI University (AIU) in mid-2018.
AIU is a multiday course available in two formats: an in person offering delivered at the customer site over two full days, and a newly introduced live, virtual model delivered on-line during three half-day sessions. With the virtual offering, legal and financial services professionals can actively participate in project-driven, best-practice, remote AI workshops that use their own, realworld data to address specific business issues – which is particularly relevant now, as work-from-home becomes the new norm worldwide.
Training and learning to go beyond the basics
While early entrants in the legal AI market focused on out-of-the-box models to address specific use cases like reviewing employment contracts or lease terms, we find these point solutions by themselves very limited. They may serve as a starting point for engaging the technology but are narrowly focused; they can’t easily accommodate more nuanced and unique content, or allow an organisation to address a broader range of issues.
For those that want this option, we do offer a large number of pre-packaged models that can be used straight away, but our main focus is on providing a highly trainable AI engine and teaching transferable skills. As individuals continue to apply and hone these learned AI skills, they are able to take a more innovative approach to efficiently solving business issues and driving business transformation. The knowledge gained on an initial AI project compounds on itself and becomes additive. Firms can focus on more fine-grained results in subsequent projects, or delve deeper and use the models they’ve developed to apply to other document types and business problems.
The models that clients build themselves – using their own data – are often superior to out-of-the-box models that are developed and trained using publicly available data from sources like Edgar, which is one of the reasons we encourage customers to take this step. Additionally, custom models allow customers to capture the data points that are most important or relevant to them and that reflect the particular contracts, leases, share purchase agreements, or other documents that they’re working with, providing a ‘tailor fit’ that out-of-the-box models can’t provide. Moreover, we don’t think that technology companies should simply throw a bunch of pre-packaged tools at customers and then walk away expecting customers to figure out how to leverage AI to its full power. Transformational results are achieved through a deeper dive.
That’s where AIU comes into play – either the in person or virtual offering. We tackle a client’s first project right alongside them – like a due diligence review or some other initial application of AI to get their feet wet. We use ‘project one’ as a live training exercise to help them develop best practices and repeatable processes that will allow them to implement the technology across their organisation for ‘project two’ and well beyond. Using their own data and solving for a real business issue helps to spur ideas of how, where and for what else they can leverage the technology.
Part of AIU is familiarising customers with the product – what the AI solution is, how it works, what it can do. But an even larger part of AIU is teaching customers how to think about AI in general and how to approach an implementation in their organisation. Along the way, we dispel a few myths and share plenty of best practices.
AI Myths
Myth 1: AI is a magic wand.
AI is a very powerful technology, but for many scenarios and use cases, it’s not realistic to expect to ‘wave the AI magic wand’ and instantly get perfect results. Some advance planning and preparation are typically required – but it’s a straightforward process, and certainly nothing for firms to be intimidated by.
Myth 2: The robots are coming to take our jobs.
Fear not: AI is not going to be taking anybody’s job. Instead, it’s going to give lawyers a new tool to do their jobs more efficiently – and to gain a competitive edge over their competitors.
This second myth speaks to the fact that AI is a relatively new technology and that there is learning to be done about how this emerging technology fits into the legal industry and what role it will play. The objective of AIU is to help firms establish AI centres of excellence, understand what AI is and is not well-suited for, learn how to train their AI engine with precision and accuracy for best machine learning results, and determine how to leverage the right mix of AI methods to achieve their objectives. We want to make sure customers are armed with the tools and training to put AI to work across their data stores and documents, so that they can help their organisations unlock critical knowledge and insights in a repeatable way across the enterprise. Our approach, you might say, is to teach a customer to fish, not give them a fish every day.
So, how best to make this actually happen?
At AIU, AI instructors with deep technology and legal expertise work with clients in advance to help identify use cases for the session and to facilitate the most effective approach to extraction techniques for client projects. The daily curriculum includes demonstrations with user data and individual and group exercises to evaluate and deepen user skills. Notably, we work with the customer’s actual data, which results in a richer learning experience.
Approaching AI the right way: best practices
In a typical AIU, we focus on some key best practices that help set customers up for success with AI in general:
- Focus on the problem and find the right people
An AI team should include a mix of customer stakeholders, including data scientists, knowledge managers, lawyers, partners, contract specialists, and trained legal staff. It’s important to have a subject matter expert – preferably someone at the senior partner level –who really understands the use case that is going to be tackled with AI.
This means that they can really drill down on questions like: what is the business problem they’re trying to solve? What sort of documents are they dealing with? What are the data points they’re looking to extract, and how can they tease those data points apart if they’re embedded in documents in a fairly complex way?
While that senior level person is fundamental to making sure things are done properly at the outset, he or she might not want to be the one using the AI on a day-to-day basis afterwards. It’s important, then, to ensure that the people who are actually going to be using the tool on a day-to-day basis are also in the room. Amongst them, make sure you have a technology based or knowledge-based person who can answer questions like: where are the documents coming from? Who manages those databases? Is there someone who will be in charge of uploading those documents? These are all important questions that warrant careful consideration.
- Give the data the attention it deserves
Too often, people don’t consider the time and effort required to make a good model. Another common mistake in AI implementations is wanting to skip straight to the capture stage. This leads to inconsistency and – ultimately – to inferior models. An upfront investment in data curation will result in better and more accurate models. These models will provide a greater return in the long term.
To guide this data curation process, you’ll also want to create a design document that serves as a ‘playbook’ for the entire team to refer to. This ensures everyone is tagging data points consistently.
- Understand the different tools in your toolbox
There’s more than one way to get your hands on the data points you’re seeking to extract – and it’s important for users to know what the different tools are, and how to use one or more of them in combination. Questions most customers have are: when do I use which method? And when and how do I combine methods for the best outcome?
Sometimes the answer comes down to volume. Let’s say you’re going to review 10 share purchase agreements. Even if those documents are 100 pages long, it might make more sense to review those manually than to try to train a model to identify clauses in those documents. For starters, you might not have enough samples to tag – and by the time you’ve tagged five samples, you may as well just go ahead and review the other five manually as well. If you had 1,000 documents to review, obviously that would be a different story and would favour using machine learning.
AIU familiarises customers with two different types of machine learning. The quick learning algorithm – as the name suggests – only requires you to tag a few examples of what you’re looking to extract before you can run it across the remaining documents. The advanced learning algorithm, meanwhile, requires you to tag at least 30 samples of the data point before you can run it across the remaining data set. The advanced learning algorithm is a more nuanced tool, but it requires more front-end training.
Sometimes, the way a document is laid out is consistent enough that you don’t need an algorithm at all – instead, you can take a rulesbased approach. Think of rules like the Boolean terms you’d use to find messages in your inbox – for example “Show me all messages where from is ‘Jane Montague’ and received is ‘2019’.” Rules are quicker to write than an algorithm, and quicker to run across a data set. As the saying goes, sometimes you just need a flyswatter, not a cannon.
By the end of AIU, customers will be able to approach a business problem and ask themselves: is this a manual review situation, or is AI going to be applicable here? If AI is applicable, which approach do I want to use – machine learning or rules? And within machine learning, does this seem more like a quick learning algorithm or an advanced learning algorithm situation? Creating a review form that is capable of drawing on both rules and machine learning allows firms to take a ‘combo’ approach and use different methods to pull out different pieces of data from their documents.
With this knowledge and training under their belts, companies are well positioned to start leveraging AI in transformative ways.
Using AI to solve real world problems
Companies that are seeking to innovate with AI can see the value that training like AIU delivers by looking at companies that are successfully using the technology to solve real world business problems.
AI in Action: UK law firm makes accurate predictions around insurance claims
Challenge: A leading insurance risk and commercial law firm based in the UK and Ireland needed to capture data from its documents to analyse and make accurate predictions around claims outcomes.
Benefit: Built models that can quickly and accurately extract information from largely unstructured documents for use in analysing claim costs and likely outcomes, allowing firm to provide better advice for its clients while reducing claims processing time.
AI in Action: large toy manufacturer solves NDA review
Challenge: Lack of corporate legal department resources to respond to numerous requests for information contained within contracts.
Benefit: Processed 6,000+ nondisclosure and influencer agreements within 40 minutes. Now expanding usage of AI to licensing and distributorship agreements.
AI in Action: global financial services company tackles LIBOR
Challenge: In advance of upcoming LIBOR transition, the firm needed to review over 1,000 documents (including mortgages, promissory notes, and mortgage deeds of trust) for 16 data points.
Benefit: Cut 50% off the expected review time to identify LIBOR documents with duplicate document detection. 1,500 hours saved from the overall review. When an additional 220 documents showed up for review at the last minute, a simple dragand-drop into the AI engine allowed timely processing in a matter of hours, rather than an associate having to cancel her weekend plans to review the documents.
Once they’ve gone through AIU, organisations are well positioned to start using their AI engine for these types of more advanced use cases. The knowledge and expertise they gain also opens the door for them to make use of other AIpowered technologies, which can help them find, analyse and identify organisation information across disparate systems and unlock key insights. One step at a time, organisations can start to create their own AI center of excellence.
In this way, AI serves as a platform that – when deployed correctly – can create transformational results. While some projects will remain ‘push button,’ many projects will require some degree of advance preparation, set-up, and understanding around when to best use which tool for what task. Until organisations start hiring people who natively have these AI skill sets, everyone will need to be ‘coached up’ a little bit. Fortunately, AIU delivers this type of deep knowledge, providing a foundational understanding that can help firms fully deliver on the promise of AI.
Three best practices to apply to any AI implementation
Don’t Forget the Daily Users. In addition to including a senior level person who’s defining the business problem that AI will tackle, make sure your AI team also includes the people who will actually be using the tool on a day-today basis.
Garbage In, Garbage Out. Invest the time upfront in making sure everyone is tagging data the same way – otherwise, the accuracy of the model will suffer. Create a playbook everyone can refer to, to keep everybody on the same page.
Know When the Juice is Worth the Squeeze. There are several different tools for extracting data. For a low volume project, manual review might be the most practical way forward. But for larger jobs, AI is worth exploring – and machine learning, rules, or a mix of both offer their own advantages.
Bryan Bach is the US AI University Coordinator at iManage
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
.jpg)
