Use of AI in finance transactions – how can it help?

Briefing

Background

AI is the current buzzword on the mouth of every tech-minded person, from huge corporations to small entrepreneurs.

The (relatively) recent accessibility of open AI (for example, Chat GPT) has brought this into the homes of the everyday user. Chat GPT, a form of generative AI (or Gen AI), has the ability to create text that is intended to imitate any kind of existing writing. It uses data provided to it and processes it in such a way to create further data (ie words and text), which follows the same style and logic but is completely new. This is also done at an incredibly high speed, far faster than what a human could generate in the same amount of time.

Using AI to generate a leveraged finance facility agreement

It has been reported that, currently, an AI can analyse the data source and create an output of 75,000 words each minute. This means that, in terms of the pure number of words created, this AI could generate the full LMA leveraged finance facility agreement, with all optionality and drafting notes included in under two minutes – obviously, the challenge is that the content of the AI-generated document is not (yet) of the same quality as that of the LMA documents. That being said, it is clear to see the potential of Gen AIs in law firms working on finance transactions, even at this early stage.

How does it differ from automation products currently used?

Law firms today do use some forms of automation in order to streamline processes. These are, usually, carefully created specific software that use logic to create documents. Typically, there would be a form of list where the lawyer would input specific information which the software would read and, after completing several pre-coded logical steps, the document would be generated based on that input. There are versions of AI that are also used, but these are still in their infancy. An automatic tool (ie automation) does not use data from external sources and is only able to generate content based on the specific information fed to it and the outcome is, therefore, relatively predictable (which is the entire point of the software). These are intended to be a first step to aid the lawyer's efficiency and the generated document would then be reviewed and amended by a lawyer before it is ever seen by the end client. However, this is not true Gen AI and it is clear to see the potential benefits of a fully functioning Gen AI for legal documentation.

Limitations to using AI in this context currently

Words generated may not have their intended meaning

It is important to note that the AI tools currently available are not sophisticated enough to produce full documentation. The current AI tools are based on words and language and they interpret and generate text based on the words used around or sequential to certain words.
An example given in What does AI mean for professional services business models? (Butterworths Journal of International Banking and Financial Law – December 2023) is that "references in a document to "interest" are interpreted by the system to refer to a rate of return on a loan because they are grouped with references to “lenders” and “per annum”, among other like terms". Due to this style of reading and writing, the machine can create words in a way that makes sense but will not necessarily have the meaning that such clause is intended to have – for example, an AI is not able (in its current form) to create a facility agreement if it is "fed" a term sheet.

Standardisation of templates may not be possible

Lawyers, usually, prefer all the documents they produce to look the same (think something as simple as "house style" as well as client specific templates). This results in a document being drafted that looks and feels the way the lawyer (and maybe the client) prefers. There is, therefore, a low (if not zero) chance of uniqueness in these documents.
A current AI would not be able to create text in such a specific form and if you asked 10 AIs to produce a document with the same inputs, they would all be slightly different, with a degree of uniqueness built in.
Reinforced learning systems (ie trial and error and feedback) can be integrated, which means the machine would "learn" what is required. However, this requires a huge amount of data.
One train of thought is that if every large law firm in the UK provided all of their facility agreements to a machine, this would still not be enough data for the machine to learn what is required in order to produce a facility agreement that is ready to go. This is also operationally unviable – law firms (and clients) are protective of their documents and would not be willing to provide them. These doesn’t even take in to account the huge operation that would be required to redact the documentation beforehand to ensure that there is zero risk to any breach of client confidentiality.

Given the above, at this current stage, full AI documentation is not viable using current products although it is worth noting that the speed of AI development is staggering therefore progress in this area is likely to be fast.

How can AI be used currently in putting together finance documents?

At its current stage, AI is more used as a tool and can be used by business services to generate models rather then be used for full legal documentation. Clients are more likely to generate more useful results from AI (in how to streamline and operate their businesses) than law firms are. There is a level of expertise that is required by a lawyer (and each lawyer is different – the expertise for a banking lawyer is different from those that are required by a consumer protection lawyer) and the AI cannot, current, operate at such a level.

AI can pass exams – Reuters recently reported that GPT-4 (Microsoft AI) has passed the New York Bar exam and scored highly enough to practice law in most US states. However, there is a vast difference between passing an exam and operating as a lawyer.

It is worth noting that all professional services are regulated by a regulator and, therefore, have to adhere to a strict set of rules. Regulators are, by nature (and particularly after the 2008 financial crisis), conservative and risk adverse and are slow to adopt new technologies (especially those that are not fully understood). Regulators will need to be on-board with new technologies before they can be fully utilised by professionals. This may result in new regulators being created to supervise specific tech products.

Looking ahead!

Despite the current limitations, the potential for AI is enormous – lawyers, in particular, still work on analogue tasks (eg reviewing & analysing term sheet, drafting document based on term sheet, refining document). In an increasing digital world, this may not be the long-term solution. Law firms will need to keep up with the technology and adapt to the digital world or else they may be left behind.

History shows that technological advances have the potential for market disruption for a specific sector (eg the wide adaptation of emails – if a law firm was slow to implement email communication, they may well have been left behind) and, provided AI continues to grow at its current rate, it has the potential to do so here.

It will be important for professional services providers to manage (and stay on top of) this disruption to ensure they are not pushed out of the market by more tech savvy firms (or start-ups). It looks like AI is here to stay, but there is a long way to go before lawyers are out of a job!