Today, every forward-thinking enterprise and financial organization is exploring how AI and large language models (LLMs) can be integrated into their operations. Sage's annual CFO research report, The Secrets of Successful CFOs, found that eight in 10 CFOs are embracing AI and automation to save time and increase their own strategic value.
But is the excitement warranted? Properly applied, AI can transform and accelerate everything from financial reporting and investor communication to fraud detection, investment analyses, asset allocation, and more.
The key, however, is to use AI appropriately and mitigate improper responses. There's an understandable temptation to quickly deploy general-purpose LLMs in chatbot-driven applications. Yet, as some early adopters have experienced, this approach can open the door to grossly inaccurate responses.
To date, many of today's early implementations of AI have relied on broad foundational models that can do everything from write Shakespearean sonnets to analyze econometric data to generate photorealistic artwork.
But the fact is, the broader the mandate, the greater the chance for error. And in finance, where precision is paramount, unpredictable results and errors are unacceptable.
AI must present uncompromising levels of safety and trust to succeed in the heavily regulated industry of finance. A domain-specific LLM - continually and carefully trained on rich sets of high-quality financial data and a narrowly defined set of use cases - is designed to establish helpful guardrails that ensure greater accuracy and increased trustworthiness. Another way of thinking about those guardrails is to understand that they are the designed outgrowth of a system that surrounds the LLM. The LLM isn't simply pushed out to users who are left to fend for themselves and worry about accuracy.
How do we solve these issues? At Sage, we're collaborating with Amazon Web Services to launch the first domain-specific accounting LLM. This collaboration aims to greatly improve how SMBs optimize their operations and navigate the complexities of local compliance challenges.
The results and behaviors of broad, off-the-shelf consumer-facing LLMs can change over time, which can become problematic. A finance-specific LLM eliminates that disqualifying variability.
To avoid errors, the finance-specific LLM must be part of a larger application that strictly governs and curates what can and can't be done, what data can be used, what questions can be asked, and what data to which each user is entitled. This is the key distinction between enterprise-grade LLMs and the more unpredictable and unregulated nature of consumer GPT usage.
AI has spurred such excitement and enthusiasm in almost every corner of business (and society in general) because it's easy to quickly see the rapid value it can provide by automating tactical, time-intensive tasks.
A study from the Alan Turing Institute found that financial organizations that use LLMs employ them primarily for lower-risk tasks, accompanied by significant human involvement, such as summarizing text, creating literature overviews, increasing the speed of analyses, and reinforcing decision-making processes.
When AI frees finance professionals from endless hours of repetitive, predictable, tactical, and/or routine tasks, it presents extraordinary opportunities in the form of reclaimed productivity. This 'AI dividend' can help companies - from small or medium-sized businesses to large enterprises - unlock new insights and strategies that propel the business forward.
Consider one scenario: the CFO of a mid-market manufacturer reviews the latest financial statements and questions the declining profitability of certain product segments. What's driving those declines? Are there hidden contributors to rising expenses?
Previously, such questions might initiate cycles of emails and custom queries, consuming days of work. With a finance-specific LLM - one that's versed in accounting, local regulations, GAAP principles, international nuances, and other relevant topics - the CFO can get immediate answers and explore new questions without any latency and without needing to be an expert in queries, reports, or data. The AI module fetches the right data, formulates the answer, confirms its accuracy, and delivers it in a digestible format.
What's more, now the controller is free to perform higher value work instead of running reports. They might start by modeling on the fly, asking questions about various scenarios and combinations. For instance, "What happens to this product line's profitability if we add 10 new customers this month and bring aboard a different supplier?"
This 'elevation effect' cascades all the way down the organization, allowing even entry-level accountants to accomplish more in less time. Finance professionals can engage in strategic work, such as modeling scenarios, instead of performing routine tasks like bank reconciliations or general ledger adjustments.
Additional (and perhaps less recognized) value of a finance-specific LLM lies in its ability to level the playing field for large and small organizations alike. For instance, a successful regional bakery executive can ask broad, open-ended (but nonetheless highly strategic) questions like: "How do I grow my business?" or "What do I need to do to increase sales?" The LLM examines your data - such as inventory figures and sales histories - and delivers specific, custom, actionable recommendations: "Reduce prices by 5% on four key products." "Expand production in the Springfield factory." Previously, such data-driven insights were beyond the reach of all but the very largest companies, because they required teams of consultants and weeks or months of analysis. AI and an accounting LLM can entirely transform our expectations.
Domain-specific LLMs lay the groundwork to a more strategic approach to finance by automating the layered, somewhat complicated low-value tasks that are nonetheless required of the finance team. In this manner, AI can create a culture of continuous accounting, continuous assurance, and continuous insights that frees the office of the CFO to step beyond traditional boundaries, ascend to more strategic roles, and rely on deeper insights to guide the business to higher and faster outcomes.
