The Economic Argument for Using LLMs in Treasury & Risk

Large Language Models (LLMs) in Treasury and Risk

LLMs, or Large Language Models, are advanced artificial intelligence systems designed to understand, generate, and interact with human language, mimicking natural human communication. They are a subset of machine learning models within the field of natural language processing (NLP). In previous articles, I have discussed their use in various tasks, including:

• Streamlining Communication

• Writing for the recipient

• Undertaking research

• Stress and scenario creation

• Preparing for committees & analysing risk

• Adding narrative to presentations and record-keeping

These tasks, which we regularly undertake, are time-consuming and, in some cases, repetitive. LLMs excel here, thereby freeing up our time and thought, allowing us to concentrate more on what is important and productive.

Even if no further progress were made in developing these models, their impact on treasury and risk management, and indeed the whole firm, would be significant. We are only just learning how to use them in business.

Key Findings from the UK Department for Education Report

In November of last year, the UK Department for Education published a report titled "The Impact of AI on UK Jobs and Training." This report outlines AI and LLMs' influence on the UK labour market. The key findings are:

1. Occupational Exposure: Professional occupations, especially in finance, law, business management, and teaching, are more exposed to AI. Large language models are particularly relevant in these fields.

2. Sectoral Impact: The finance and insurance sector is most exposed to AI, followed by the information and communication, professional services, property, public administration, defence, and education sectors.

3. Geographic Variations: Workers in London and the South East have the highest AI exposure, reflecting the concentration of professional occupations. The North East has the least exposure, but overall geographic variation is smaller than that observed across occupations or industries.

4. Qualifications and Training: Employees with advanced qualifications, especially in accounting, finance, and economics, are typically in jobs more exposed to AI. Conversely, those with qualifications in building, construction, and transportation operations are least exposed.

The analysis emphasises that AI is likely to augment rather than replace many jobs, with varying degrees of exposure across different occupations, sectors, and geographic regions.

Upon first reading, I was surprised at the level of impact AI could have on us. However, after reflection, LLMs allow us to produce more output from the same resources or the same output from fewer resources. The implications are profound in terms of the workplace and more widely society.

A Ridiculously High Return on Investment

My own experience with LLMs has been more of an experiment in what they can do, particularly in the field I am familiar with.

A simple conclusion I have come to is that the $20 subscription to GPT-4 is probably the best investment a bank or building society can make for its employees in Treasury and Risk. At first glance, this may seem extravagant, but a closer look makes it hard to ignore. One of the highest costs for financial firms is employing staff. With this in mind, let's look at things further.

I've made some broad assumptions, so what follows is an estimate, a conservative one at that, because I have only considered the use of LLMs for document preparation.

I’ve assumed an hourly rate for a skilled person in this area of £40/hour (based on a 46-week year, working five days a week, eight hours a day, totalling 1,840 hours per year, with a remuneration package of £75,000).

Let’s also assume that 10 hours a week are spent on writing and processing documents, a common scenario given the workload for internal meetings and management reporting. From my experience, GPT-4 can easily halve the time taken for this, saving 5 hours per week.

Over the year, this amounts to 230 saved hours with a value of approximately £9,200 or 12% of the salary bill. Even if you deduct the 20-25 hours needed to get proficient in using LLMs in your business (a process that can be significantly accelerated with the appropriate training). The first-year cost saving is over £8,000 or 10%.

A More Positive Outlook

You can now see why the report from the Department of Education highlights the implications of AI on banking and finance, especially in the southeast of England. While this may be true, I see a more positive impact on the workplace.

Colleagues working in treasury and risk are constantly under pressure to deliver and things fall behind. The solution, hiring more staff, can be tricky, and many firms have headcount limits. Spending a small amount on LLMs will go a long way to alleviate some of these problems without breaking any headcount restrictions. It's also that much easier to sign up for an LLM than to go through the process of hiring.