Risk tolerance is a concept we frequently refer to, but how often do we have a serious conversation about what it truly means? By its very nature, risk tolerance is not a fixed, static thing. It changes and fluctuates day by day, influenced by a variety of factors. Time and time again, I've witnessed how easy it is for people to take on more risk when markets are on the upswing, and then quickly cut back on that risk when things start to go down. Link pay to profit, and guess what happens? Risk tolerance increases, as if by magic. Persuasion plays a role too. If you have a strong and powerful advocate pushing for taking on more risk, it's amazing how quickly colleagues fall in line and accept that view as well. Accounting also gets a look in. If you run your business in a mark-to-market environment, where everything ends up in the P&L today, it can be a remarkably sobering experience. Is it any wonder that some of the biggest losses I've seen have been in lending? The spreading out of pain over years through the loss of net interest margin seems, for some reason, far more acceptable than taking the hit upfront. But in the end, the net effect is the same - it hits capital. All these factors increase the difficulty of pinning down exactly what risk tolerance truly is. The techniques we use to understand market, credit, liquidity, and operational risk further complicate matters. They are predominantly quantitative measures, but you can't reduce everything to pure numbers - you need to apply subjectivity as well. I've frequently seen limit breaches, which incidentally breach the stated risk tolerance, be overridden by judgment. More evidence that tolerance is a flexible concept. But one of my main concerns is fungibility or lack thereof. Each category of risk has its own limits, and however you look at them, you can't easily compare. Loan exposures cannot be added to liquidity exposures to get anything meaningful, and this is a problem. You can't directly compare individual risks with each other. However, you can get a subjective feel for how much overall risk you are taking - is it low, medium, or high? Specifically, what is the chance that your capital will be significantly depleted from the individual risks you have? Once you boil things down to this very basic common denominator, you are in a much better place to question what you are doing. And there are three important questions to ask: 1.     Why should the market pay us to take this risk? 2.     Can we correctly price this risk? 3.     What experience do we have in managing this risk? Your answers to these questions should ultimately frame your risk tolerance. By constantly revisiting these fundamental issues, you can develop a clearer understanding of your true risk tolerance, beyond the day-to-day fluctuations and influences.

Treasury and dealing have undergone a remarkable change over the past four decades. What was once a field dominated by intuition, paper-based processes, and "barrow boys" has evolved into a highly regulated, technology-driven profession that demands a diverse set of skills from its practitioners. In this blog post, I'll discuss the key changes that have shaped things and how today's professionals can take out some of the tedium to make things fun again. From paper blotters to electronic trading Forty years ago, the treasury world was a far simpler place. Dealers relied on paper blotters to manage positions, trades were conducted over the phone, and the concept of risk management was rudimentary at best. Regulation was virtually non-existent, and dealers were expected to make money based on their gut instincts rather than sophisticated analysis. Today the picture couldn't be more different. Electronic trading platforms have replaced blotters, and the sheer volume and size of trades have grown exponentially. Information is now abundant and instantly accessible, thanks to resources like Bloomberg. Bid-offer spreads have tightened, and the number of major players in the market has shrunk, with most participants now classified as clients. New skills As things have evolved, so too have the skills needed to succeed in this field. Today's dealers are more likely to hold graduate and post-graduate qualifications, and proficiency with spreadsheets and data analysis is essential. Treasury professionals must be able to not only interpret data but also use it to forecast P&L and risk. The modern treasury environment also demands strong communication skills. Dealers must be able to explain complex strategies and concepts to a wide range of people, often without relying on jargon or technical terminology. Report writing, meeting preparation, and the ability to articulate the pros and cons of different approaches are now core competencies for treasury professionals. Moreover, dealers must be aware of the cognitive biases that can cloud judgment, such as recency, confirmation, and overconfidence. These concepts were once seen as virtues but academic studies have shown they aren't. Regulation rules One of the most significant changes in the treasury world has been the rise of regulation. The days of freewheeling, unconstrained dealing are long gone, replaced by a complex web of rules and oversight. Risk management is now at the forefront of every trade and the fear of censure looms large. Despite the challenges posed by regulation, some fundamental truths about markets remain unchanged. They are inherently unpredictable, and even the best-laid plans can unravel at the first sign of trouble. Extreme events, while unlikely, occur more frequently than expected, and we must be prepared to act quickly to adjust to the situation we face. AI makes light of work As the demands continue to grow artificial intelligence (AI) can help. It can be a powerful tool for automating routine tasks, such as report writing and policy updates, freeing up time for more strategic and interesting work. But AI's potential extends far beyond mere automation. It can also help dealers improve their skills in areas where they may be lacking, such as communication and data analysis. By using AI for scenario planning, tail risk assessment, and other nuanced tasks, a deeper understanding can be obtained. Perhaps most importantly, AI can make treasury work more engaging and enjoyable. By eliminating the tedious, repetitive aspects of the job you can focus on work that drew you to the field in the first place. Maybe it's not quite like the old days but the barrow boys were street-smart and saw an opportunity when it presented itself.

Issue at Hand A potential shift in the Bank of England's policy to stop paying interest on commercial bank deposits is becoming a live political debate. There is a growing sentiment that paying interest on reserve accounts is effectively a subsidy to the banking system. This increasing pressure could influence the Bank of England to change its policies, presenting significant challenges for financial institutions. It's important to note that this situation is not the same as a negative interest rate policy; there are viable alternatives available, although their returns will likely be lower than those you are accustomed to. Over-reliance is Risky Relying heavily on a single counterparty, such as the Bank of England, or one method of managing deposits makes you more susceptible to regime change. This is especially true for small banks, which find the simplicity of depositing with the central bank well-suited to their business model. They lack the resources to manage a more diversified portfolio of assets. Larger banks, however, have the resources and capabilities to diversify their asset base, thereby reducing their exposure to changes in central bank policy. Nonetheless, all banks must recognise that heavy reliance on one counterparty leaves them vulnerable to policy shifts If the Bank of England stops paying interest, you will need to consider alternatives such as Treasury bills (T-bills) and repurchase agreements (repos). These options, while liquid, introduce more credit and operational risks and less certain liquidity. With this in mind, it is essential for banks to regularly test alternative strategies to ensure they are viable. This means evaluating and trying out different asset allocations, understanding potential operational risks, and ensuring systems and processes are ready for a seamless transition. For example, you might periodically invest in T-bills or repos to gauge their performance and integration with existing systems. Regular testing helps banks identify and address any operational issues, ensuring that they can quickly adapt to changes in the Bank of England’s policies without significant disruption. But this isn’t all. You need to ask yourself how much of a problem this is. It’s all very well recognising that Net Interest Margin (NIM) will fall but by how much? Readers of my blog will know I like to keep these things simple and here a delta will suffice. If you deposit £200m a year in the Reserve account for every 1% fall in interest there is a cost of £2m from your business. Yes, a zero interest rate policy for reserves is so painful and it will force you to use alternatives. But they too can be affected. Second-Order Effects Changes of this type lead to unpredictable outcomes, for instance, if many banks suddenly shift their investments to T-bills, the increased demand could suppress returns, reducing their attractiveness. Running some type of stress and using the delta approach to give you a feeling of what may happen is worthwhile. As I’ve previously discussed, using AI to give you some scenarios for the effect on both the wholesale and retail markets is worthwhile. While stress testing can provide insights, there’s nothing like being prepared; it’s amazing how long it can take to get credit clearance, ensure your IT systems work and have proper legal documentation in place. Starting now could look fortuitous in the future. In summary, these are the five questions you should be asking: How diversified is your asset portfolio? Are you overly reliant on deposits with the Bank of England, and what are the immediate steps you can take to diversify your assets to reduce fragility? What viable alternatives exist to replace Bank of England Deposits? What are the potential alternative assets, such as T-bills and repos, and how do their liquidity, risk, and return profiles compare to your current holdings? What would a regime change cost you? Are there some simple rules of thumb that help? How could second-order effects change things? How might increased demand for alternatives like T-bills affect their yields, and what would be the financial impact on your business if returns were to fall by 1% or 2%? Are these impacts significant for your overall profitability and stability? How prepared are your systems and processes for a transition? Have you tested your systems and processes to ensure they can handle a shift to alternative assets, and what are the costs and operational implications of such a transition?

One of the podcasts I like is the Rational Reminder: "Sensible investing and financial decision-making, from two Canadians". It provides great insight into personal finance, and a recent episode is well worth listening to if you are an index investor. My views on low-cost investing are best summed up in my blog post from 14th January 2021, "Do less and make more". The inspiration for that post came from the late Jack Bogle's book "The Little Book of Common Sense Investing." In summary: 1. Low-cost investing is essential to maximise returns over the long term. 2. Index investing, as advocated by John C. Bogle, is a simple and effective strategy accessible to everyone. 3. Outperforming the market consistently is nearly impossible, and low-cost index funds have a high probability of beating actively managed funds over the long run. 4. Behavioural biases, such as buying high and selling low, can negatively impact investor returns. 5. Asset allocation dictates investment performance and lower investment costs allow for taking less risk to achieve the same level of return. If this is something that you believe too, I suggest listening to the Rational Reminder 25th April episode (302) with Michael Green, a well-thought-through discussion about the impact of market liquidity on investment strategies. In summary: 1. He argues that the growth of passive index investing has made markets less elastic and more fragile. As more money flows into index funds that buy stocks regardless of valuations, it pushes prices higher in a feedback loop. But this also means prices could fall drastically if there are ever large outflows. 2. This dynamic makes it harder for active managers to outperform passive funds over time. 3. He sees major risks building; he notes the solution is very difficult politically since the index fund narrative is so entrenched. 4. Paradoxically, investors should keep investing in index funds individually, but collectively it creates growing fragility. 5. Regulators and the biggest fund providers can address this, but they are currently incentivised not to. It may take a very severe event to force changes. What are the main differences between Bogle and Green? Mike Green's perspective on the growth of passive investing differs from Jack Bogle's in several key ways: 1. Market impact: Bogle believed that passive investing was a net positive for markets and investors, making investing more accessible and lowering costs. Green argues that the growth of passive investing is making markets less efficient and more fragile. 2. Price discovery: Bogle believed that as long as there are some active managers in the market, prices will be set efficiently. Green suggests that the proportion of active managers is becoming too small to effectively counteract the impact of passive flows. 3. Systemic risk: Bogle didn't see passive investing as a source of systemic risk. Green argues that the growth of passive investing is creating a feedback loop that amplifies market movements and could lead to a severe crash if outflows accelerate. 4. Active management: Bogle believed that active management was largely a losing proposition for investors due to higher costs and the difficulty of consistently outperforming the market. Green suggests that active management is becoming structurally disadvantaged due to the distorting impact of passive flows, rather than just being a matter of skill. 5. Future returns: Bogle believed that passive investing would continue to deliver satisfactory returns for investors over the long term. Green is more sceptical, suggesting that a significant portion of recent market gains may be attributed to the shift towards passive rather than fundamental factors. In essence, while Bogle saw passive investing as a largely benign and beneficial trend for investors, Green sees it as a source of growing systemic risk that is fundamentally altering the way markets function. However, both seem to agree that for most individual investors, passive investing remains the most practical and rational approach, even if their reasons for this conclusion differ. You may, however, want to consider what a drawdown of 85% would do to your portfolio and frame your risk appetite with this in mind. This is why I like the Rational Reminder; some really good insights into personal finance presented by two knowledgeable and thought-provoking Canadians.

The role of a treasury has changed dramatically over the past four decades. It used to be about making money, which was a lot easier because markets weren’t efficient. There were wide spreads, customers weren’t knowledgeable, and arbitrages paid well. (Hands up who can remember the 10 basis points free lunch between futures and swaps). This was an era when greed was good. Today it’s different. Trading rooms that were full of noise have fallen silent. The easy profit opportunities are gone, regulations are tight, and risk mitigation is now the focus. The choice This means that your expectations of treasury need to be clear. Do you want it to be a profit centre or not? Whilst this isn’t a binary question, there are two distinct approaches. The first is using treasury to efficiently manage the balance sheet without taking much risk. This is where you use the market to hedge and give up the bid-offer spread because you are the customer. The second is having Treasury as a profit centre, but that is easier said than done. A significant challenge is to unravel how money is made. In truth, revenue streams are so mixed up that it’s almost impossible to see what’s going on. Revenue sources It reminds me of the trader who told me that he was going to lose his job because the “system” couldn’t allocate the high coupons he received to his P&L, but at the same time debited funding costs. If simple stuff like this goes on, there is absolutely no way you know what’s happening. Then there’s the dealer who told me how he loved his new job. Everyone had to come to his team to deal with anything under two years; this captive no-risk franchise facilitated bumper profits and bonuses for the desk. No wonder other parts of the bank mistrusted treasury. Maybe proper transfer pricing is the answer, but that’s a can of worms. Every approach I’ve seen is flawed in one way or another. It’s a great theory that doesn’t work in practice. Where's your edge? However, the real risk occurs when you don’t have any reason why a competitive market would pay you to transact. In this situation, if you insist on making money, dealers will oblige, and there’s only one way to go. That’s to push the boat out in terms of risk. Here I’m talking about credit, interest rate, liquidity, and basis risk. With the most stringent limits in place, it is quite amazing how enterprising dealers can become when pushed. Even when you aren’t breaching limits, the risk is there. I recall a Treasury that hedged a Gilt portfolio with swaps and then, at an opportune time, unwound the trade to make a significant gain. The Board was outraged. Why? Because they realised that this profit had come from risk-taking, and none of it showed up in their reports. So, there you have it, the more you push, the greater the risk. This means proper oversight and a curious, questioning approach are essential. If something seems too good to be true, it probably is, underscoring the need for a knowledgeable second line of defence that can tactfully probe what’s going on and get to the bottom of it. The biggest risk, however, is behaviour, not of traders but of management. Business cycles come and go, and so does risk-taking. 1987, 1992, 1998, 2008, and 2020 all have one thing in common – market disruption. It can come from anywhere and can lead to anything, and it’s almost inevitable that when these things happen, you are least prepared because there’s nothing like a period of benign conditions to lull you into a false sense of security. We used to joke about the EM traders increasing their profit and limits year by year. They would go down for the biggest sum ever when the cycle turned, and it did. If you want the treasury to make more, be satisfied you can answer these three things: 1. Why should the market pay you to trade? 2. What is the real source of money? 3. How good is the second line?

After the release of GPT-4o earlier this week, I tried some of the standard tests I conducted on earlier models. The simple conclusion is that if you are not using AI, you will be left behind. . These three tests used single-shot prompts, and the results are sufficient to acknowledge that AI, in many cases, can do what humans can. The difference lies in speed and breadth of knowledge. AI is now much faster, and its ability to cover a wide range of topics at a reasonable depth is unquestionable. In the three examples below, the responses may not be as I would write them, but it is clear that the results are better than those I would have expected from someone with less than five years of experience in treasury. In other words, AI is your helpful junior assistant, willing and able to do things without questioning and with speed. This does not preclude a human in the loop. Indeed, this is highly recommended because AI can get things convincingly wrong. If you are lazy and take the output as given, you could be in for a nasty surprise. As discussed in a previous post, the increase in productivity that can be achieved and the minimal cost incurred means that anyone asking to increase headcount in treasury must surely answer the question: have you tried AI before adding to the payroll? All Chief Executives, please note. Now the examples. Explaining a PowerPoint slide I asked GPT-4 to explain a PowerPoint slide. The prompt is simple, and the response is acceptable. It does this easily, and whilst the narrative isn't quite as I would have put it, it is fast and effective. More Writing for the board In this example, I asked GPT-4o to explain a picture of a red-amber-green report for the board. I then asked it to provide a simple narrative and the action the board should consider. It does this with ease, and it is the direction in which board papers will go in the future. AI can repurpose content for the audience and remove the padding that seems to have accumulated over the past few years. Here's the summary, (GPT 4o won't generate a link for a picture-based prompt): The current financial assessment reveals that credit exposures to Barclays and NatWest are within acceptable limits, though Lloyds exceeds its limit by £1m. Liquidity is generally strong, with high-quality assets exceeding limits and adequate coverage in stressed scenarios, although the "days no action" metric is below the limit by 10 days. Interest rate risk is mostly within limits except for the 1-3 years period, which is slightly above. Basis risk remains within acceptable bounds across all categories, and the investment portfolio shows significant under-utilization in the 0-1 year range, while the 2-5 years range is fully utilized. Actions and Recommendations To address these findings, it is recommended to reduce Lloyds' exposure by £1m to meet its limit. The liquidity buffer for "days no action" should be increased to achieve the required 150 days. The interest rate risk for the 1-3 years period needs adjustment to bring it within the £1m limit. Additionally, increasing investments in the 0-1 year portfolio will optimize returns and ensure better utilization of available funds. Monitoring and adjusting these areas will help maintain compliance and enhance financial performance. Differentiating between banks and companies In this test, I asked GPT-4o to explain how banks and companies hedge their interest rate risk. Could AI differentiate between banks essentially hedging their risk on a floating-floating basis, demonstrating that they are managing margins, with companies focusing on the cost of money and locking in an expense? The result covered more than I expected. I can't fault the AI; it does the job well, even if it's not as I would have written it. That's just something that I will have to get used to. More Why this is important It's simple: AI can perform many tasks in our specialist area quite adequately, and we should be using it in our day-to-day work. If we aren't, there is something wrong. It is no use waiting for a magic bullet to arrive, giving you all the knowledge to integrate AI into your business. You will find that AI is not something you can suddenly switch on; it needs some development, learning, and experimentation. Think of it as trying to run a 5-kilometre race. If you have not run before, you will struggle, but with some practice, the distance will become very manageable. So my challenge to you; if you work in treasury and you're not using AI, unpack it and get going.

I've been reading Ethan Mollick's new book "Co-intelligence: Living and Working with AI" It's an excellent read for anyone interested in the topic; there are some gems in it. One of them is how creative AI is. That's something I've seen, for example, when I drew a picture based on a poem by Wilfred Owen. Brainstorming with AI Ethan Mollick suggests, "From a practical standpoint, the AI should be invited to any brainstorming session you hold," and using this as inspiration, I asked AI to give me 30 names for a one-year savings bond for the fictitious "Leicester Building Society". Here are the results: Charnwood Saver Bond Guildhall Growth Bond Foxton Finance Bond Belvoir Benefit Bond Iron Age Investment Bond Leicester Legacy Bond Ashby Advantage Bond Richard III Return Bond Bosworth Battlefield Bond Loughborough Loyalty Bond Leicester Growth Bond Leicester Secure Saver Future Growth Bond Leicester Nest Egg Bond Secure Horizon Bond Leicester Annual Yield Bond TrustPlus Savings Bond Leicester Premier Bond Golden Year Bond Prosperity Bond by Leicester Leicester Boost Bond Foundation Year Bond Capital Gain Bond Leicester Wealth Builder Solid Future Bond Pathway Savings Bond Yearly Yield Bond Steady Growth Bond Leicester Financial Pillar Bond Annual Advantage Bond Some are great, others mediocre, and some could make savers run a mile. As Mollick suggests, the AI comes up with a few pearls in the output — something that would then need humans to decide on what's best. I like the "Bosworth Battlefield Bond"; the "Richard III Return Bond" might also attract certain types of customers; there is no accounting for taste! Synthetic Polling Which name is best is a matter of choice, and maybe AI can help us here too. In an earlier blog, I explained how AI can create synthetic responses. Now, I adopt a similar approach to ask AI to imagine it is asking 100 customers which name they prefer. With a bit of to and fro, I got the poll done, here is the result: Synthesised Feedback Generation: Example Respondent: Jane Doe, 45, Teacher, from Market Harborough. Selected Name: Leicester Legacy Bond. Reasoning: Prefers names that evoke a sense of history and trust. Leicester Legacy Bond - 14 votes Charnwood Saver Bond - 12 votes Belvoir Benefit Bond - 11 votes Secure Horizon Bond - 9 votes Golden Year Bond - 8 votes Foxton Finance Bond - 7 votes Leicester Secure Saver - 6 votes Future Growth Bond - 6 votes Leicester Growth Bond - 5 votes Iron Age Investment Bond - 5 votes Richard III Return Bond - 4 votes Bosworth Battlefield Bond - 4 votes Loughborough Loyalty Bond - 3 votes Ashby Advantage Bond - 3 votes Leicester Premier Bond - 2 votes TrustPlus Savings Bond - 2 votes Prosperity Bond by Leicester - 2 votes Other names received 1 or 2 votes each, reflecting a diverse but less concentrated interest. (Note, this doesn't add up to 100, AI seems bad at doing some basic things, but here I'm interested in sentiment rather than absolutes, so I'll go with it). Now a word of caution — I'm not certain how it's done this, it could be making the whole thing up. But this is an experiment, and with more prompting, I could get individual detailed responses and use sentiment analysis all within AI. More work needs to be done here but if AI can replicate demographics and mimic behaviour this appears to be a cheap, fast and viable approach to sentiment analysis. Copywriting Finally, I asked the AI to imagine it is a copywriter for an advert for our bond, and it came up with some fairly good copy you could tweak. However, more worryingly, I then ask it to forget all the regulations, and it enthusiastically agrees. The copy is completely unethical and breaks FCA guidance. Not that regulated institutions would do this, but we know AI can be extremely persuasive and with AI models reaching human-level ability and soon passing it, then bad actors can encourage people to do all manner of things. Let's Have More Fun at Work What I've learned from this exercise is that Mollick is right — AI should be at the brainstorming session. It's fast and creates some gems, but for now, it takes humans to spot them; AI has taken out the hard bit and left you with the fun bit — judging the best. My synthetic poll is a bit of fun too, but it could save both time and money to get rapid feedback and also challenge you on what you perceive is the best name or outcome. The copy is good, but it could be better, but this was a one-shot round and if I spent more time on it, I could get close to something very acceptable. I've not included the copy here but the full transcript of my conversation is here. Potential Risks The persuasive power of AI can only get better, and in financial services, regulators will have to look out for its use and its potential to mislead investors. But what is more worrying is that in the future many of us will be persuaded to do things without even realising we are being manipulated. What message you trust will be very hard for many people to assimilate. This means those who own the big models, whether AI companies or governments, have the power to alter thoughts and at the same time, those with extreme views now have the potential to generate disruptive narratives. I'm not certain this is what we call X-risk, but it's a dystopian future. As an optimist, I prefer to see the advantages AI brings, but we must have our eyes open and be realistic. Practical Application Finally, my exercise here shows that one person, with some AI knowledge, can find a perfectly suitable brand or product name, test it on an audience, and write copy. A further step could be to ask AI to write a memo seeking approval for a product launch; a day's work can be done in an hour.

The Calculator Analogy Back in the 1970s when I was at school, one of the prized possessions for many students was a Sinclair calculator. It cost a lot of money and it could do simple arithmetic. When it first came out, I can remember that it was banned from use in the classroom. The argument was that boys needed to understand maths by doing things longhand and that calculators damaged this process. It puzzled me why this didn't apply to log tables and slide rules, but it didn't. But of course, anyone could use a calculator for homework and within a few years, the world was recognised for what it was, where new tools help you progress and the ban was lifted. What's Important? This leads to an important question: what do we teach students now that AI is so freely available? It's easy to put an assignment into AI and get a credible answer. What's more, it is doubtful that AI detectors work, particularly if there have been some changes to the generated text. Furthermore, there is good evidence that students writing in a second language are more prone to getting flagged as using AI. As the world becomes a much more complex place, it is almost impossible for an individual to assimilate all the knowledge that has gone before. AI therefore helps us get to a point where we can build on knowledge and be more creative sooner than would otherwise have been possible. AI is here to stay and we should integrate it into what we do to improve our educational experience and understanding of the world. Core Skills There is something else: because I was forced to learn my multiplication tables and do simple arithmetic, I have a reasonable ability to approximate – a useful life skill and something I wouldn't have gained from using a calculator. What core skills children need in the future is something educators need to decide. Mental Models and Soft Skills When I was younger, I wish that I had been made aware that information collected from different disciplines can be connected to give you insight into the way that the world works. Mental models are valuable; if you have an appreciation of compounding in finance, you can understand the implications for infection rates for viruses, something that was largely misunderstood in COVID. Whilst scientific skills are important to the country, a good grounding in all sorts of fields is valuable to the individual; some experts believe that talking with AI as a person gets more from it and that people with humanities backgrounds can excel. It is a powerful argument; having an articulate discussion with AI is a way to get it to do more. Critical Thinking When spreadsheets were first introduced, there was blind faith in the print-out. This danger also applies to AI. It likes to please us and the answers it generates are echoes of the prompts used; it's easy to be fooled. What's more, those who hold the model hold the truth, so the ability to question what you see and cross-reference it to what you know is an important skill. Flipped Learning and Continual Challenge Last week i mentioned that I had read Ethan Mollick's new book "Co-intelligence: Living and Working with AI". One of the things he mentions is the idea of flipped learning, where students engage at home using AI to prepare their answers, and rather than hand in a piece of coursework, they then discuss their solution in the classroom. This is a reworking of the way we educate and it is more like the workplace; a problem-solving process that ends with a discussion. It also improves social skills and leverages the role of the tutor as a facilitator. I like it. Something else that Ethan Mollick mentions is the need for continual challenge. It is much easier to do the same thing repeatedly and confuse it with progress. If you want to drive forward challenge yourself and make the task harder. AI is good at this as it can adjust its response to the level of the student, thereby improving the learning experience. It is truly a tiered experience. I liken it to the idea that we can all have access to an expert coach. If you haven't exercised for years, then a very moderate programme of improvement will get you on the first steps, and to improve from there means upping the difficulty to higher levels. Painful but effective. Practical Applications Learning in the workplace is time-consuming and expensive, for it to be genuinely helpful, it needs to apply to the circumstances employees face. This is where GPTs come in, (custom prompts you can build in GPT4). Any business that has important processes and procedures can assimilate them into a GPT and develop a powerful knowledge base that can be used by all. I would urge you to give it a go.

The Challenge My mum loves reading, but she has read all the books in her bookcase at least twice. At 97, it's simply not feasible for her to visit a library or bookshop to find more. One of her favourite authors is R.F. Delderfield, whose books depict England in the past*. However, without investing a significant amount of time to grasp the intricacies of his writing, it's challenging for me to identify similar authors. Perhaps this is a task for Gen AI? Turning to Generative AI What if I asked Gen AI to recommend similar books written by different authors that would be suitable for a 97-year-old? The result was a pleasant surprise; some of the authors were already familiar to my mother, while others were new**. I made a random selection and, to keep costs down, discovered that purchasing second-hand books on Amazon is the most cost-effective approach. I can say that my mum enjoyed some of the books more than others, but curating a diverse selection like this would have been a challenge and incredibly time-consuming. It's a novel way in which AI is helping us improve our daily lives. But that's not all. Comparing AI with Traditional Search Years ago, I would have visited a library or a bookshop and browsed, purchasing a selection of books that I thought might fit the bill. This process would have been quite time-consuming. Perhaps I would have sought assistance from the librarian or bookshop owner along the way. Then, over the last 20 years, search engines emerged, providing an alternative method of browsing the Internet with Google to find titles that might have been suitable. Alternatively, I could have utilised Amazon, possibly focusing on purchases made by people who had already bought a book by R.F. Delderfield. The Future of AI in Everyday Life These options are still available, but now I can also utilise Gen AI. It's clear to me that there is a distinct difference between a Google search and using AI. Search is excellent if you know precisely what you want and are simply trying to locate it on the Internet. However, if you require more background information and knowledge about a topic, it's not as effective. Even if you do find information, it tends to be in a wiki page format, which may or may not answer the specific question you are seeking. Consequently, since the advent of Gen AI, my use of Google has significantly decreased, and I prefer to use platforms like Perplexity to guide me in the right direction. These Gen AI-based searches also have a less tacky feel to them; while I may still be subjected to marketing, it's not as overt. I've heard quite a few commentators remark on how the old search engines often feature paid advertisements at the top, pushing them above free information due to the commercial nature of the model. There's something else, too. Learning Lessons Learning how to use AI is something that many people will have to do, and when faced with a blank prompt window, it's amazing how complex it can seem. That's why some basic training to gain confidence in integrating Gen AI into your everyday life is helpful. I've heard some commentators mention that, in their opinion, simply learning prompts from other people is no better than trying to learn phrases from a book when learning a foreign language, and I tend to agree. A smattering of phrases might get you a beer or a coffee, and you may be able to buy a railway ticket and say thank you, but it's not truly a conversation. That's the difference between using AI and using Google. With Google, you input a prompt and sift through the answers, whereas using Gen AI tends to be more of a dialogue and discussion, or at least that's how you get the most out of Gen AI; it's not just a fact-generating machine. However, I do think that seeing what other people do with Gen AI is very helpful. The example I've given above with my mother may not be immediately obvious, but if you copy, adopt, and adapt, you may find that a slightly different approach can help you benefit from AI. For example, if you've read a book and want to learn from that experience, you can now have a constructive dialogue with AI about the content of that book, reinforcing your impression of the written work. My mother continues to read, and now she has discovered some new authors and titles. If you know an elderly person who shares a similar passion for reading, why not try Gen AI to expand their reading list? ChatGPT: * Ronald Frederick Delderfield was a popular English novelist and playwright, best known for his historical novels and family sagas that vividly depict English life from the Victorian era through the mid-20th century. His works are characterized by their detailed narrative, accessible style, and warm humanism. Delderfield's novels often explore themes of societal change and the impact of historical events on personal lives, resonating with a broad readership. Some of Delderfield's most notable works include: The "Avenue" series ("A Horseman Riding By", "Theirs Was the Kingdom", and "The Green Gauntlet") - This trilogy covers the life and times of the inhabitants of a Devonshire community from the early 1900s through World War II, illustrating the social and economic changes affecting the characters. The "Swann" series ("God Is an Englishman", "Their Sons' Great Glory", and "The Avenue Goes to War") - This series follows the Swann family as they rise from modest beginnings to prominence through the transport industry, detailing their personal trials and tribulations across generations. "To Serve Them All My Days" - Perhaps one of his most beloved works, this novel follows the life of a shell-shocked World War I veteran who becomes a teacher at a boys' boarding school in England. It's a poignant exploration of healing and service, set against the backdrop of the interwar years and World War II. Delderfield's writing is notable for its affectionate and detailed portrayals of English countryside and the decency of ordinary people. His books often provide a comforting retreat into a world where values like integrity and duty are upheld, and they have been adapted into several successful television series, bringing his stories to an even wider audience. ** R.F. Delderfield is known for his sweeping historical novels and engaging family sagas, which often explore themes of English rural life, societal changes, and the human spirit. If your mother enjoys Delderfield's style and themes, here are some authors and books she might also enjoy: Maeve Binchy - Known for her warm and engaging novels set in Ireland, Binchy's books often explore interpersonal relationships and community ties. Some of her most popular works include "Circle of Friends" and "Tara Road." Rosamunde Pilcher - Pilcher's novels frequently delve into family dynamics and life in the British countryside. "The Shell Seekers" and "Coming Home" are two of her beloved novels that offer rich storytelling similar to Delderfield's. Elizabeth Jane Howard - Her series "The Cazalet Chronicles" begins with "The Light Years" and is an absorbing family saga set before, during, and after World War II. It provides a detailed look at the lives and transformations of the Cazalet family. Winston Graham - Best known for the "Poldark" series, Graham weaves compelling narratives of romance, adventure, and family ties against the backdrop of Cornwall. The series offers a blend of historical detail and drama. John Galsworthy - His series "The Forsyte Saga" is a classic exploration of upper-middle-class English family life spanning three generations. The series begins with "The Man of Property" and offers a critique of Victorian and Edwardian society. Nevil Shute - While often categorized under historical romance and adventure, Shute's novels like "A Town Like Alice" and "On the Beach" provide compelling narratives with a strong sense of place and era. These authors provide a similar sense of depth in their character development and historical settings, which should resonate with a fan of R.F. Delderfield.

The Bank of England commissioned Dr. Bernanke's report, "Forecasting for Monetary Policy Making and Communication at the Bank of England: A Review." We use AI to get a summary, try it for yourself and do a Q&A; Banks and Building Societies can use this as a catalyst to enhance their own forecasting, particularly for managing interest rate risk; AI can provide hypothetical advice on balance sheet management, such as using forward rates, and stress-testing unhedged positions; I ask Claude to write like Ben Bernanke - a novel way of using AI to think like an expert; AI in this persona can facilitate valuable discussions on an institution's readiness for rate shocks by generating extreme tail risk scenarios and their potential impact on interest rates; Through dialogue, AI is changing what we do and it’s a lot more interesting to work like this. As I'm sure you are aware, the Bank of England has been concerned about its ability to model the future. Whilst I'm of the school that much modelling is guesswork and that forward rates are the wisdom of crowds, many financial institutions prefer to put their interpretation on what may or may not happen. In this article, AI provides a very brief summary of Dr. Bernanke's report, "Forecasting for Monetary Policy Making and Communication at the Bank of England: A Review." For an in-depth read, you need the PDF from the Bank of England website, and if you are feeling experimental, upload it to a large language model and engage in a Q&A session. As I write this, it reminds me that the world has changed. Before AI, lengthy documents had to be read to get insight. Now, there are faster and arguably better alternatives using AI, and long documents can be repurposed for the reader. But that’s not all; it's far easier to take some of the suggestions contained in the report and then apply them to your own business. All you have to do is prompt AI, and this is much more engaging than having to do all the work yourself; in fact, working with AI is like having a conversation, and for many people, that can be fun. Let's have a look at what GenAI makes of Dr. Bernanke's report. I asked ClaudeOpus for a summary: "The key points of the report "Forecasting for Monetary Policy Making and Communication at the Bank of England: A Review" by Dr. Ben Bernanke are: The report reviews economic forecasting at the Bank of England, focusing on how forecasting can better support policymaking and communication during times of high uncertainty. The report makes recommendations to: Improve and maintain the Bank's forecasting infrastructure, including data management, software, and economic models. Support effective policymaking by equipping the MPC and staff to learn from past forecast errors, identify risks, and deal with uncertainty. Help the MPC better communicate its economic outlook, risks, and the basis for policy choices. Serious problems were found with the Bank's outdated forecasting infrastructure, which limits analysis. Upgrading data management is underway, but economic models also need maintenance and updating. The forecast process should highlight significant errors and evaluate models, judgements, and assumptions. Alternative scenarios should compare policy choices and assess risks. The central forecast is emphasized in the Bank's communications, but has drawbacks. Publishing selected alternative scenarios would improve communication of the outlook, risks, and policy rationale. Despite deficiencies, the Bank's overall forecasting accuracy in recent tumultuous years was comparable to peer central banks and external forecasters. But the experience provides motivation to review forecasting processes. In summary, the report recommends infrastructure upgrades, expanded use of alternative scenarios in policymaking and communication, and reconsidering the emphasis on the central forecast to strengthen the Bank's economic analysis and transparency". Using the Review as a Catalyst If the Bank of England has felt it worthwhile to overhaul this area of their work, I wonder whether some of the insights from the review can be used by banks and building societies to enhance their own forecasting models in times of uncertainty. I put this to Claude, and the response is: "The bank may want to consider conducting a comprehensive assessment of its current forecasting infrastructure. This could involve identifying potential areas for enhancement and investment related to data management, software tools, and economic models. Another idea is to develop a structured approach for periodically evaluating the bank's forecast performance, underlying models, key judgements, and assumptions. Insights gleaned from this review process could then be incorporated into future decision-making. To better assess risks and inform strategic choices, the bank might explore integrating alternative scenarios into its forecasting and planning processes. This can provide a more robust view of potential economic outcomes. The bank may also want to look at ways to enhance its stakeholder communications. Offering a clearer and more detailed perspective on the economic outlook, risks, and rationale behind decisions can foster greater understanding and engagement. Finally, the bank could benefit from keeping abreast of evolving best practices in the forecasting field. Maintaining an active dialogue with industry peers to exchange ideas and experiences may also yield valuable insights for the bank to consider". Balance Sheet Positioning I want to explore a specific area - positioning the balance sheet for interest rate risk. This is often done by examining forward rates and then overlaying management judgement, which is used in the hedging strategy. Simply put, the balance sheet is managed based on the judgement of future rates and within the risk tolerances considered acceptable. It sounds like science, but it is far from it. Wouldn’t it be interesting to get Dr. Bernanke’s view on this? One of the things I've learned when using GenAI is that you can prompt it to adopt a persona; just ask it to imagine it's Warren Buffett, Jim Slater, or someone else, and it will. It's as if the essence of the individual and their thinking is concentrated in the name. I suspect this is due to the way GenAI has been trained, and the name acts as a sort of magnet for a particular line of thinking, which is then used in its output. Let's now ask Claude to assume the persona of Ben Bernanke and give us some advice on managing the balance sheet; here is part of the result: "Adopting forward rates as the base model and then stress-testing unhedged positions for tail events is indeed a prudent approach to managing your bank's interest rate risk within your risk appetite. Let me elaborate on this: Adopt forward rates as the base model: Use market-implied forward rates as the foundation for your interest rate modeling and forecasting. Regularly update your forward rate curves to reflect the most current market expectations. Incorporate forward rates into your hedging strategies and mortgage pricing decisions. Rationale: Forward rates provide an objective, market-based view of future interest rate expectations. By aligning your base model with market consensus, you can make more informed decisions and reduce the potential impact of individual biases. 2. Explore tail events on unhedged positions: Identify and quantify your bank's unhedged interest rate exposures, particularly in your mortgage portfolio. Conduct stress tests and scenario analyses to assess the potential impact of extreme interest rate movements on these unhedged positions. Evaluate the resulting downside risks and determine if they fall within your bank's risk appetite. Rationale: While hedging can mitigate interest rate risk, it's crucial to understand the potential consequences of tail events on any remaining unhedged exposures. By stress-testing these positions, you can gauge the potential downside risk and make informed decisions about whether to accept, mitigate, or transfer that risk. 3. Ensure clear understanding and alignment with risk appetite: Clearly define and communicate your bank's risk appetite, particularly with regard to interest rate risk. Regularly assess whether your interest rate risk exposure, as determined by your stress tests and scenario analyses, aligns with your stated risk appetite. Make adjustments to your hedging strategies, product mix, or pricing as necessary to maintain alignment with your risk appetite. Rationale: A clear understanding of your bank's risk appetite is essential for effective risk management. By consistently ensuring that your interest rate risk exposure remains within your defined risk appetite, you can protect your bank's financial stability and make more informed strategic decisions. By adopting forward rates as your base model, exploring tail events on unhedged positions, and ensuring clear alignment with your risk appetite, you can create a robust framework for managing interest rate risk within your bank's constraints. This approach will help you make better-informed decisions, protect against downside risks, and allocate your limited resources effectively". The tail risk problem is very relevant; after all, it is the tail risk that will potentially get us into trouble, and this is something we need to know about. One way we can explore this is by building scenarios. Again, I ask the Bernanke Claude persona for guidance; here is part of the response: "To help you better understand and prepare for tail risks, I've designed three extreme events that could lead to significant movements in interest rates. For each event, I'll provide a brief description and an estimate of where the Bank of England's Bank Rate might stand in the aftermath, given its current level of 5.25%. Severe Global Economic Recession: Description: A severe and prolonged global economic downturn, triggered by a combination of factors such as a major financial crisis, geopolitical tensions, and a collapse in consumer and business confidence. The recession leads to a sharp contraction in global trade, rising unemployment, and deflationary pressures. Estimated Bank Rate: In this scenario, the Bank of England would likely cut the Bank Rate aggressively to stimulate the economy and combat deflationary risks. The Bank Rate could potentially fall to 0.5% or even lower, depending on the severity and duration of the recession. Unexpectedly High and Persistent Inflation: Description: A scenario in which inflation surges well above the Bank of England's 2% target and remains persistently high due to a combination of factors such as supply chain disruptions, energy price shocks, and a significant expansion in fiscal spending. The high inflation expectations become entrenched, leading to a wage-price spiral. Estimated Bank Rate: To combat the inflationary pressures, the Bank of England would likely raise the Bank Rate significantly. In this scenario, the Bank Rate could potentially rise to 8% or higher, depending on the level of inflation and the effectiveness of monetary policy in anchoring inflation expectations. Major Geopolitical Conflict: Description: A significant escalation in geopolitical tensions, leading to a large-scale military conflict involving major global powers. The conflict disrupts global trade, energy supplies, and financial markets, causing a surge in risk aversion and a flight to safe-haven assets. Estimated Bank Rate: The impact on the Bank Rate would depend on the balance of recessionary and inflationary pressures resulting from the conflict. If the conflict primarily leads to a global economic downturn and financial market stress, the Bank of England might cut the Bank Rate to support the economy, potentially to 2% or lower. However, if the conflict also results in supply-side disruptions and inflationary pressures, the Bank Rate might remain unchanged or even increase moderately to balance growth and inflation risks". These events are based on opinion rather than fact, and indeed, that's what stress testing is about; it's not a science, nor is it an art—it's a mixture of both. For any institution interested in examining its tail exposures, using GenAI in this way to explain some of the nuances that could unfold is a starting point for a valuable discussion about how rates and liquidity could move and whether the institution is capable of handling such disruption in a short period. It would be very easy to continue our discussion with GenAI and build on what we have so far. Not so long ago, if we had the time, we would have read Dr. Bernanke’s report and left it at that. Now, you can get the key information and drill down as you like. AI simply acts as your research assistant. But that’s just a start. You can now consider using the report as a catalyst for change in your business. AI can make suggestions that you can follow up on. You can also marshal the wisdom of the persona to improve the way you work. From my experience, this can be very powerful. It captures the essence of the expert and gives us insights we would never have been able to obtain otherwise. Why not try it out? We live in interesting times. If you've enjoyed reading this and find it helpful, I'll be pleased to hear from you.

It's a well-known but seldom-discussed fact: Board directors of financial institutions rarely read all the documentation originated by the FCA. The challenge lies not only in the sheer volume of guidance and practice emanating from the regulator but also in their complex and often dry content. This is where generative AI, with models like ChatGPT and Claude, comes into play—offering a means to quickly produce nuanced summaries of lengthy documents. Consider the dilemma further exacerbated by the nature of regulatory publications: not only are they extensive and tedious, but they also adopt a one-size-fits-all approach. This necessitates interpretation specific to the regulated entities, which in turn, weakens the board's oversight capabilities. Effective governance demands a comprehensive understanding of guidelines and policy statements. Without grasping the nuances of regulation, boards are unlikely to pose the critical questions needed to stimulate debate and challenge the business, ensuring that regulation is effectively enforced. This scenario highlights a communication gap between regulators and boards—a situation that, while blameless, leads to potential lapses in oversight and broader implications for stakeholders and the public at large. So, how can generative AI improve this? Take, for example, the FCA’s issuance of FG24/1: Finalised guidance on financial promotions on social media. Despite being one of the "lighter" documents at 45 pages, it's unrealistic to expect board members to engage with such material thoroughly. Traditional solutions have involved condensing these documents to a more manageable length, tailored to the business's specific needs. Yet, these summaries often end up either too technical or misaligned with the board's high-level needs. Generative AI changes this dynamic. With minimal effort, an entire document can be fed into an AI model, which then produces a summary closely aligned with a business's needs, suitable for inclusion in a board pack. As a demonstration, let's explore a simple AI-generated (Claude Opus) summary of FG24/1, this is for a Board of a small financial institution and it is written as a paper: Dear Board Members, I am writing to bring to your attention the Financial Conduct Authority's (FCA) recently published guidance on financial promotions on social media (FG24/1). As a small bank/building society, it is crucial that we fully understand and adhere to this guidance to ensure compliance, protect our customers, and safeguard our reputation. This paper outlines the key aspects of the guidance and how it impacts our responsibilities as a Board. Consumer Understanding and the Consumer Duty The FCA emphasizes that financial promotions on social media must support consumer understanding by providing clear, fair, and balanced information about the benefits and risks of our products. This aligns with the new Consumer Duty principle, which requires us to deliver good outcomes for retail customers. As a Board, we must ensure that our social media marketing strategies prioritize customer understanding and informed decision-making. For example, if we are promoting a new savings account on Facebook, we must clearly communicate the interest rate, any limitations or restrictions, and potential risks (such as variable rates) in a balanced manner. We should challenge our marketing team to demonstrate how their approach supports the Consumer Duty and customer understanding, rather than just focusing on business success metrics. Prominence of Disclosures and Risk Warnings The FCA guidance stresses that specific disclosures, such as risk warnings, must be prominent and not obscured by the design features of social media platforms. This has significant implications for our oversight responsibilities. For instance, if we are promoting a mortgage product on Instagram, the key information about fees, terms, and potential risks should be clearly visible and not hidden behind "see more" links or truncated by character limits. As a Board, we must scrutinize how our marketing presents these important disclosures and insist on full compliance with the prominence requirements. Suitability of Social Media for Complex Products The guidance raises concerns about the suitability of social media for promoting complex financial products. As a Board, we must critically assess whether social media is an appropriate channel for each of our products, considering the target audience and the level of complexity. For example, promoting a structured investment product with multiple features and risks on Twitter may not provide sufficient space for balanced and comprehensive information. In such cases, we should direct our marketing team to focus on more suitable channels or simplify the social media content to signpost customers to detailed information elsewhere. Oversight of Affiliates and Influencers If our bank/building society engages affiliate marketers or influencers to promote our products on social media, we have a crucial oversight responsibility. The FCA emphasizes that firms remain responsible for the compliance of promotions made by affiliates. As a Board, we must ensure robust due diligence, monitoring, and control processes are in place for any affiliates or influencers we work with. This could involve: Setting clear guidelines and approval processes for affiliate/influencer content Regularly reviewing affiliate/influencer posts for compliance and balance Having the ability to quickly rectify or remove non-compliant content Terminating relationships with affiliates/influencers who consistently breach guidelines We should require regular reporting from our marketing team on affiliate/influencer management and be prepared to challenge any deficiencies robustly. Non-UK Entities and Group Governance For banks/building societies that are part of international groups, the FCA guidance has important implications. Financial promotions from non-UK entities in the group that are accessible to UK consumers must still comply with the UK financial promotion rules. As a Board, we must ensure effective governance is in place to mitigate the risk of UK customers being exposed to non-compliant promotions from overseas entities in the group. This could involve measures such as geoblocking, using UK-specific social media profiles, or having UK-based teams review and approve global social media content. Conclusion The FCA's guidance on financial promotions on social media has wide-ranging implications for our responsibilities as a Board. We must provide robust challenge and oversight of our social media marketing strategies, ensuring they prioritize clear communication, customer understanding, and good outcomes in line with regulatory expectations. Compliance with this guidance should be a standing item on our Board agenda, and we should demand regular reporting and evidence of adherence from our marketing and compliance teams. By fully engaging with this guidance and embedding its principles into our governance, we can ensure our bank/building society harnesses the power of social media marketing in a responsible and compliant manner, protecting both our customers and our reputation. Yours Sincerely, [Your Name] My comment on this While initially, the output may not perfectly meet your requirements, (international groups etc), a bit of trial and error can refine it to closely match the business's specific context. Moreover, generative AI can ask questions designed to promote board-level debate on the topic. Claude again: Dear Board Members, Following on from my previous paper, I have compiled a list of 10 key questions for the Board to consider and discuss. These questions are designed to stimulate debate and ensure effective oversight of our financial promotions on social media, specifically in relation to our regular savings and mortgage business, our marketing channels, and our use of intermediaries. Please find the questions below, prioritized based on their importance: How are we ensuring that our social media promotions for savings and mortgage products are clear, fair, and not misleading, and that they support customer understanding in line with the Consumer Duty? What processes do we have in place to review and approve social media content for our savings and mortgage products, including content created by intermediaries, to ensure compliance with FCA guidance? How are we monitoring and assessing the ongoing compliance and effectiveness of our social media promotions, and what management information do we receive to facilitate our oversight? What due diligence, monitoring, and control processes do we have in place for the intermediaries and affiliates we work with to promote our products on social media? How are we ensuring that the specific disclosures and risk warnings in our social media promotions for savings and mortgages are prominently displayed and not obscured by platform design features? What is our approach to selecting appropriate social media channels for promoting our savings and mortgage products, considering the complexity of the products and the target audience? How are we managing the risk of non-compliant or misleading promotions being made by intermediaries or affiliates, and what action would we take in the event of a breach? What training and guidance do we provide to our marketing team and intermediaries to ensure they understand and adhere to the FCA's social media promotion guidance? For any non-UK entities in our group, what governance arrangements do we have in place to ensure their social media promotions accessible to UK customers are compliant with UK rules? How are we staying abreast of evolving social media marketing practices and FCA expectations, and how do we plan to adapt our approach in light of new developments? I recommend that we dedicate time in our upcoming Board meetings to discuss these questions in detail. We should also ask our marketing, compliance, and risk teams to provide their insights and evidence to support our discussions. By actively engaging with these questions and challenging ourselves to maintain high standards, we can ensure that our social media promotions for savings and mortgages are effective, compliant, and always prioritize the best interests of our customers. Yours Sincerely, [Your Name] -------- What have we seen here? The flexibility of generative AI to adapt to your business's context highlights its potential as a tool not just for summarisation but as a facilitator of meaningful governance discussions. This doesn't entail disclosing sensitive information; even basic prompts can yield valuable insights. Indeed, the integration of generative AI into governance processes represents a significant leap forward. It streamlines communication between regulators and boards, contextualises information, and ensures that critical issues are both read and debated, thereby strengthening the oversight process. Additionally, this approach represents a time-efficient solution for all parties involved. This exploration of AI in governance might not be immediately obvious, yet as we go deeper into AI's capabilities, the opportunities for enhancing operational efficiencies become apparent. This isn't knowledge that can be gleaned from a book; it requires hands-on practice to fully realise the benefits that are unfolding as technology advances. Some may await a bespoke AI solution, believing it will prevent the need for experimentation and learning. However, such wishful thinking overlooks the competitive edge gained through early adoption and familiarity with AI. Delaying engagement with AI technology only widens the gap with competitors, potentially leading to costlier and less effective operational models. If you've found this perspective interesting and beneficial, I'd be pleased to hear from you.

I've noticed that GPT-4 is getting better all the time; you can do more, and the results are better. I decided to try a little test. Could GPT-4 take a red, amber, green risk report and explain what it was seeing? Could it then write the narrative so that anyone reading it got a clear understanding of the risks present? Here is the report that I gave GPT-4: Note, this is a photograph of a spreadsheet rather than the spreadsheet itself. The reason is that by trial and error, I've learnt GPT-4 is good with pictures, and it's a feature that has improved over the past few months. This is the prompt that I used: Here is a colour coded report for the board of a building society showing some of its financial risks. Please write a narrative and explain anything you think is important to them or any action required This is the response that I got from GPT-4: The provided image appears to be a financial risk report for a board of a building society, detailing various metrics related to credit exposures, liquidity, interest rate risk, basis risk, and the investment portfolio. Here’s a narrative explaining the key elements: Credit Exposures: Barclays: The actual exposure is £5 million, which is well within the limit of £10 million. Lloyds: The exposure has slightly exceeded the limit by £1 million, which is a point of concern. NatWest: The exposure is £9 million, safely under the limit. Recommendation: The board should discuss why Lloyds' exposure went over the limit and consider strategies to mitigate this risk, possibly adjusting the limit or reducing the exposure. Liquidity: High-Quality Assets: There's a positive variance as the actual high-quality assets are £10 million higher than the limit. Days No Action - Stressed: The society has fewer days than the limit when no action is required in stressed conditions, indicating potential liquidity risk. Stressed with Action: Here, the actual days are above the limit, showing a buffer in the case of stress which requires action. Recommendation: It’s advisable to investigate the reasons for the negative variance in 'Days No Action' and consider reinforcing liquidity management policies. Interest Rate Risk (for a 100 basis points shift): The risk seems to be generally well-contained within limits, but for the 1-3 years category, there's a net loss exposure of £1.1 million. Recommendation: Review the interest rate risk management strategy, especially for the 1-3 years maturity period, to ensure it aligns with the society's risk appetite. Basis Risk (between different interest rates): There is considerable exposure in several categories, with particularly large short positions in the "administered rate" and "fixed rate" categories, leading to a net negative exposure. Recommendation: The board should carefully review basis risk positions, especially those with large short exposures, and consider hedging strategies to mitigate these risks. Investment Portfolio: 0-1 Year: The society has an actual surplus of £10 million compared to the limit. 2-5 Years: Actuals are aligned with the limit. Recommendation: No immediate action is required for the investment portfolio, but ongoing monitoring is essential to ensure it continues to meet the society's strategic objectives and risk tolerances. Overall: The colour coding indicates areas of concern (red and yellow) and areas where performance is satisfactory or better than expected (green). The board should focus on addressing the red areas as a priority, particularly the overexposure to Lloyds and the negative variances in the liquidity and basis risk sections. Monitoring and possibly adjusting exposure limits and risk management strategies will be crucial for maintaining financial stability. GPT-4 provides a straightforward explanation of what it is seeing for all the risks. It differentiates between the colour codes and draws attention to the things that need action. It also raises an early warning for things in amber and doesn't just see things as being right or wrong. This subtle distinction is important because it acts to help preempt future problems rather than waiting for them to happen. Any risk manager would need to apply their oversight to GPT-4's output, but it shows we should be integrating it into workflows rather than spending hours writing things longhand. There is no doubt that by using GPT-4, we will be saving a lot of time. Now, if you can picture this report, or similar, being presented to a board committee complete with the narrative, it's only one more step to give the board the power of GPT-4 to find out more. They could ask about a particular risk and how it was measured, or they could drill down into the bedrock of assumptions to give them comfort concerning their oversight. Let's try this out. Here’s my prompt: You mention the exposure to Lloyds what should I as a board member be concerned about This is GPT-4's response: As a board member, the overexposure to Lloyds should raise several concerns: Risk Concentration: The overexposure indicates a concentration of credit risk with Lloyds. If Lloyds were to face financial difficulties, your building society's financial health could be significantly impacted due to this concentration. Compliance Issues: Credit limits are often set based on regulatory requirements and internal risk management policies. Exceeding these limits can indicate a breach of compliance protocols, which can lead to regulatory sanctions and reputational damage. Risk Appetite Alignment: The limit breach suggests that the society's current dealings with Lloyds may not align with its stated risk appetite. As a board member, it's crucial to ensure that all investments and exposures align with the society's risk tolerance. Risk Management Effectiveness: The overexposure could imply potential weaknesses in the risk management framework or in the operational processes that monitor and control exposures. Liquidity and Capital Implications: If there were to be any issues with Lloyds that affected its ability to meet its obligations, your society could face liquidity constraints or capital losses, affecting its ability to lend or meet its own obligations. Actionable Steps: Review and Assess: Initiate a review of the circumstances that led to the overexposure to understand whether it was a result of a specific investment decision, an oversight, or a lack of adequate controls. Rectify Compliance: If the overexposure is a compliance breach, take immediate steps to bring the exposure back within limits and review policies to prevent future occurrences. Risk Mitigation: Consider diversifying the credit exposure or implementing hedging strategies to mitigate the risk associated with Lloyds. Engage with Management: Discuss with the management team to get a full picture of the relationship with Lloyds, the rationale behind the current exposure, and any measures they are planning to address the issue. Monitor Regularly: Ensure that the situation is monitored closely and regularly, with clear reporting lines to the board. The key is to respond proactively to the overexposure, ensuring it's addressed in a manner consistent with the society's overall strategic goals and risk management framework. All of the points raised by GPT-4 are relevant, although it will be up to individual board members to decide what's important in the context of what they know. GPT-4 isn't doing the board's role but it's helping them think through the issues. If it was felt that some follow-up questions would spark discussion, GPT-4 is there to assist: Me: Please could you suggest five questions that would be high-level and would encourage discussion on this GPT-4: Certainly, here are five high-level questions designed to provoke thoughtful discussion on the board regarding the overexposure to Lloyds: Risk Management and Oversight: "What factors led to the overexposure to Lloyds, and how did our risk monitoring processes allow this to happen?" Strategic Implications: "Considering our current risk appetite and investment strategy, how does this overexposure to Lloyds align or misalign with our society's strategic objectives?" Compliance and Regulatory: "What are the regulatory implications of our overexposure to Lloyds, and what immediate and long-term steps should we take to ensure compliance?" Operational Control: "Can we discuss the adequacy and effectiveness of our operational controls and whether they need strengthening to prevent future occurrences of limit breaches?" Remedial Actions and Future Prevention: "What specific actions will management take to rectify the current overexposure, and what preventive measures are we considering to mitigate such risks in the future?" These questions aim to not only understand the root cause of the issue but also to engage the board in a strategic conversation about aligning investment decisions with the society's risk profile and compliance requirements. What you see is that not only can GPT-4 respond to RAG-based risk reports, it can write the narrative flagging the risks that need to be monitored and then go on to explain some of the concerns that would be relevant to a board member, providing questions that would elicit more discussion. Whether you are a treasurer, risk manager, or board member, you cannot ignore the fact that the way we are doing business is being transformed. Yes, ChatGPT is not perfect and if you run this again you will get slightly different results. But by undertaking these experiments you will learn how to integrate AI into your business. If you have found this helpful would you like to leave a comment on how you use or plan to use AI?