Finance is full of repetitive work, often done manually. Reconciliation, reporting, and various calculations to evaluate assets are part of daily life. But in recent years, the AI storm has arrived.
We are constantly bombarded by newspapers and social media with promises of automating work with AI. It would seem obvious, then, to apply AI to these routine tasks and automate most financial work. As someone whose job is to find ways to automate these processes, I think about this every day. And the truth is, AI is just another tool in our toolbox—and should only be used when it fits.
Rule-based systems
Sometimes the task really is so repetitive and predictable that we can describe it in clear rules that won’t change too often. Great, we can just write a rule-based function. It will be consistent, fast, and easy to understand. You see this pop up everywhere, either as macros in Excel or more complex software apps—we can get a long way with just these solutions.
This is great when you need to retrieve and process data, move documents around, and run standard Net Asset Value (NAV) calculations. Once you have the correct information, the actual calculations of NAVs are always the same, so let’s leave this task to some predictable script.
Machine learning
But what about the tasks where writing down rules just doesn’t cut it? Either because the rules are too numerous or complex, or maybe they are prone to changing with time. These sorts of problems are usually tackled by human experts. If you tried to write a macro or code in such a setting, then someone (also a human expert) would constantly have to update this function—if it’s even feasible to write it in the first place. This is where Machine Learning (ML) can help us. As powerful as ML can be, it comes with a catch. We need high-quality data for it to work. Luckily, in the financial industry we often leave a clean trail of data for auditing purposes. This makes our industry a great fit for ML insights.
A practical example is labeling cash transactions. Each transaction comes with signals, such as descriptions, counterparties, and historical patterns, that tell you how it should be booked. For decades this has been done manually by accountants, and because of that, we now have large datasets of correctly labeled transactions. So, we trained models to do the labeling for us, with accountants reviewing the results and improving the system over time.
Machine learning itself is not new, but the combination of high-quality financial data and modern computing power now allows us to apply it effectively in everyday financial workflows.
Modern AI
So, what about the AI that everyone is actually talking about? The large language models and the surrounding ecosystem of technologies, such as vision-language models, agents, and vector databases. For example, we could have technically solved cash transaction labeling with LLMs. But it would be way more expensive and way less reliable. It would be impossible to troubleshoot. So clearly not the right fit.
But where do LLMs shine? Although slower, AI is considerably better at extracting information from documents than anything we had before with traditional Optical Character Recognition (OCR) techniques, especially with complex layouts or handwriting. Beyond document processing, AI can also augment our daily work, for example, by transcribing and summarizing meetings or helping my tech team quickly access knowledge through frontier language models. However, we shouldn’t rely on AI too heavily for tasks where precision and consistency are of utmost importance (e.g. most accounting work). That said, it’s also worth addressing a tempting but misleading opportunity.
Augmenting professionals, not replacing them
This brings us back to our initial ambitions of automating financial operations—that is my day job at AssetCare, where we take care of the legal and administrative work behind investment funds. For every task or problem our investor administrators and fund accountants face, we consider the tools at our disposal. Most of the time this actually means augmenting their work so that they can skip the tedious chores and focus on things that require human expertise and are often more gratifying.
It is not always instantly obvious which approach is right. We need a deep understanding of the limitations of cutting-edge technologies combined with the guidance from domain experts. But this is what keeps my work intriguing and fun.
Keep a level head
I invite you not to get caught up in the hype, but to follow the fascinating advancements of AI with a level head—and evaluate the possibilities it brings to your work. Let’s use the right tools for the right job.
Archie Bumbieris is an AI Engineer at AssetCare. AssetCare supports fund managers with the administration of their funds and investors through its proprietary fundtech platform. AssetCare is part of the expert panel of Investment Officer.