Hello everyone and welcome to Con 42 Large Language Models 2025. I am Andre bso. I serve as both CFO and CLO, and my background includes engineering, web development, and of course finance. Today I am excited to be here and I'm excited to talk about how large language models are. Changing the way we do finances for our companies. Okay. Let's start with a small introduction. I have been working at the intersection of finance and legal operations for a while, and I've seen firsthand how emerging AI technologies are transforming the way. Companies manage data, optimize cash flow, and ensure compliance. And this presentation will explore how LLMs are not only automating routine processes, but also how they can unlock new strategic possibilities like. Freeing up capital and, empowering businesses to make faster and smarter decisions. So let's dive in by looking at what the financial landscape looked like before LLMs. To understand the impact of LLMs, it helps. To look back at how we did things in finance not too long ago. We lived in a world that revolved around spreadsheets and reports. Our data usually sat in silos, whether in separate departments or across outdated databases that. Didn't play nicely with each other. This made it nearly impossible to get a real time snapshot of, our financial health. So we would wait for our month and or quarter end close. We would see what happened, and only then we could decide how to respond. It wasn't just the reporting cycles that slowed us down. We also faced a challenge in dealing with unstructured information, regulatory documents, contracts, emails, invoices, and even social media. Comments about our company often stayed outta reach for analytics. Because traditional finance and finance tools just, just were not built to handle natural text. That was the environment I worked in for many years. When data was stale, by the time we got it, human bias was always in the mix. And compliance frameworks were so rigid that we always worried that we might miss something. Then came the rapid development of large language models and what sets large language models apart is their ability to process. And interpret massive amounts of text, everything from invoices to emails, and even complex legal documents. More importantly, LLMs can do this in a way that captures the context and the nuance of human language. Instead of handling only structured data like numbers and columns, LLMs can, can let us leverage the huge amount of text-based data that was previously off limits or too time consuming to pars for finance. That's a remarkable shift. Now we can look beyond a single spreadsheet or a single dashboard. We can gather insights from the natural language we produce every day. When you feed enough relevant data to your LLM plus the right instructions about how your business operates, you get nearly. Real time guidance on potential risks or bottlenecks or opportunities. And, you also get a system that keeps learning constantly. This is not just about generating reports faster, it's about noticing patterns and issues that their human eye could easily miss. Which can lead to more proactive finance operations. Okay. One of the first areas where I saw LLM Shine was in fraud detect detection, running online platforms that process millions. Our transactions a day. I have seen many fraud attempts. These fraud attempts ranged from, from the usage of stolen credit cards to really elaborate identity theft schemes. And, later we used to rely on rule-based systems. Those systems were basically sets of if then conditions and, the systems could flag transactions that seemed suspicious. if they fall under any of those conditions, for example, it can be unusual purchase amounts. Foreign IP address or a mismatch in location, et cetera, et cetera. Such systems were easy for bad guys to outsmart because fraudsters adapt quickly, and our rules had to be manually updated all the times. LLMs by contracts, by contrast, can digest both structured and unstructured data. This means they can analyze user interactions, user transactions, user behavior, everything, even the text of messages a user generates. this can be analyzed to find subtle warning sites, signs of fraud. LLMs don't just rely on a single rule. They can spot differences in writing style or detect discrepancies in someone's transactional history, all of which might suggest fraud. Or some other system abuse. What I love most is how these models learn from each new data block. they take, constantly refining what suspicious looks like. Since we started using an LLM based system to detect fraud, we've seen a significant drop in fraud in fraudulent transactions while also cutting down on false alarms for legitimate customers. That was a win on both fronts. Another transformation involves cashflow management. If you are into finance and if you've ever tried to figure out exactly where your money is at any given moment of time, or if you try to predict where your money will be in a few weeks, you know how critical the cashflow management is. In many businesses, we used to rely on monthly or quarterly forecasting, which meant we didn't always react quickly if something, if something went wrong. For instance, we couldn't react quickly if, key suppliers started shipping late or if our sales suddenly spiked beyond expectations. But with an LLM driven approach, we can integrate data from our resource planning system, from supplier websites, from the markets, and even, from the world news that might affect our money flows because the LLM understands language and it can parse. Any emails about delayed shipment, LLM can see, the changes in contracts or agreements, and this information then can be fed into the bigger financial picture. It's like having a continuous conversation with your cash flow. It updates you the moment something changes instead of weeks later. For my companies, this has allowed us to keep just the right amount of cash on hand and we could invest the rest of our cash where it can, generate some returns or we could, for instance, support. New ventures with the rest of the cash, which we don't need. Okay. given that I also oversee legal compliance in our company, I'm especially enthusiastic about using LLMs in this domain. because managing compliance is hard, especially if you operate across borders with different laws and regulations. in other words, if you operate in different jurisdictions, companies usually have entire teams devoted to reading the latest updates. Check in. Checking if they apply to them and then modifying policies or procedures. This is expensive, time consuming and prone to errors. But now imagine an LLM that constantly monitors official governmental websites news, some re some regulatory bulletins. And the moment your LLM sees something relevant, like a change in data privacy law or a new payment regulation or something similar, your LLM can match that against your existing policies and highlight areas that might need updating or adjustments. it can, your LLM can even draft a first version of those changes, saving your legal team hours and hours of manual work. Of course, we still need human oversight to confirm final decisions and maintain accountability, but. The reduction in manual search and repetitive tasks is significant, and that translates into cost savings and the more secure compliance posture. I don't want to paint you a picture. That's too bright. Integrating LLMs into financial operations has its, share of hurdles. One of the biggest is data preparation. If you are financial information is scattered, unprepared, and distribute, distributed across different systems or poorly document, poorly documented. You will need to clean it and standardize it before any AI model can deliver any reliable results. In my experience, this foundational work can take much longer than you would think, but anyway, it's absolutely necessary. Another challenge is, model transparency. For example, company stakeholders may want to know why the AI system is suggesting a particular action or decision, and we can just shrug and say, it's ai, it's smart, and we have to trust it. No, but to solve this problem we've incorporated explainability tools that show which data points and which patterns infl influenced the final LLMs decision. It's not always perfect, but it works and it's a good start Then. There is human factor. Of course. We had to train our staff to work with LLMs in a collaborative way. We have to make them learn how to put questions so the, so that the system understands them well and training my own team. It took some time, but once people saw the benefits, they became really enthusiastic and they accepted the new approach. Putting these elements together, you get a new model for running a finance department. routine processes are mostly automated. Data analysis is continuous rather than sporadic, and staff members focus on higher value tasks like strategic planning and collaboration with the other parts of the business. It feels less like we are just keeping the books, it feels more like we're steering the organization towards its goals. Being informed by real time insights and guided by advanced ai. I can't overstate how freeing this can be for A CFO. Instead of drowning in a sea of everyday tasks and trailing behind the business, you find yourself anticipating obstacles. Allocating resources more efficiently and spotting profitable opportunities sooner than before. And, this change from reactive to proactive approach extends across the entire finance team and elevates the role of finance within the organization. As LLMs and related technologies continue to advance, I think we'll see even more interesting possibilities in the near future. LLMs will likely to be able to handle more than just natural text. LLMs might interpret visual data, audio streams, and other media all at once, giving us an even better understanding on of what's happening both inside and outside our companies. LLMs might act as collaborative agents that interact with different departments in real time. Offering budget adjustments or compliance checks right on the fly. We could also see more autonomous finance features. Autonomous finance means that for some simple tasks, like approving a common type of invoice or transferring funds between. Internal accounts and AI could, act entirely on its own without supervision, but under well-defined rules and, human intervention would come in only for more complex or unusual situations. While that might sound futuristic now. Many large companies are already testing the waters here and, the benefits in terms of speed and accuracy are really compelling. all of this means that the role of the CFO is changing. It's changing in fundamental way. We are no longer just the keeper. Of the historical records. Instead, we are the designers of financial intelligence systems. The overseers of complex data flows and the architects of risk management strategies that adapt in real time. We have the chance to build our companies more effectively, blending financial. Aspect with the technological aspect, but that opportunity also brings responsibility. As CFOs. We need to ensure that AI is used ethically and transparently, that our data is secure and that our decisions remain fair and human centered. We can't just. Pass the book to an AI system if something goes wrong, but it's up to us to put governance and all necessary structures in place and maintain oversight at all times. All right, in closing. I believe that large language models represent one of the most significant shifts in financial operations in the last few decades. they turn unstructured data into actionable insights. They give us near real time oversight of cash flow. They supercharge fraud detection and prevention, as well as reduce the burden of compliance. While there are hurdles like data preparation, and sometimes model explainability and sometimes team training, the payoff is fantastic and the insights are really useful. as someone who's had the privilege of building the AI systems in my own project, I can't imagine going back to the old ways of separate spreadsheets and endless manual checks with lms. our finance department is more proactive, more collaborative, and better informed. And that opens the door to new strategic possibilities. Okay. Thank you very much for spending this time with me exploring how LMS can revolutionize financial operations. I look forward to hearing your questions, experiences. And perhaps even your skepticism. After all, it's through open conversation that will continue to define these AI tools and, ensure that they service responsibly and effectively. I really appreciate your attention and I'm excited about the journey ahead for all of us in the finance world being assisted. By LLMs, thank you very much. Goodbye.