· 5 min read

Unlocking the Potential of Large Language Models in Financial Use Cases

Escape the endless maze of financial documents with large language models.

Escape the endless maze of financial documents with large language models.

In the intricate landscape of the modern financial system, an expansive web of interconnected contracts weaves together a labyrinth of risks and opportunities. With each passing year, this web has grown larger and more complex, posing challenges for organizations seeking to navigate its intricate pathways and effectively identify and manage assets and liabilities. However, with the emergence of large language models, a new era of possibilities has dawned for risk management and financial analysis.

Like torchbearers who know the way, large language models can illuminate the paths of this maze, allowing organizations to better understand their contractual obligations and opportunities. In this article, we will explore how large language models can be used to directly unlock financial value in financial services through use cases like embedded leases, revenue recognition, and stress testing and resolution planning.

A Lease by Any Other Name

Financial reporting standards require organizations to identify their leases so that they can be properly reflected on financial statements. This is relatively straightforward for something like the lease of an office building, but can become much more difficult once you consider the fact that embedded leases can live in numerous other documents, including service agreements, data center contracts, management agreements, retail contracts, advertising agreements, and more.

Although it’s possible to identify embedded leases manually - ideally, as part of the overall contracting process - this particular area is ideal for natural language processing. Using systems like Kelvin that are powered by large language models, organizations can automate this first time-consuming and error-prone step. Once potential contracts are identified, we can follow the instructions specified by ASC 842 to determine whether an embedded lease exists within a contract. First, Kelvin identifies whether a contract contains or is itself a lease. This is a two-part assessment, as it involves first determining whether there is an identified asset within the contract, and next, whether the right to control the use of the asset (for a period of time) is conveyed in exchange for consideration. Next, Kelvin classifies the identified leases as in-scope or out-of-scope for the purposes of the lease standard. While this discussion simplifies what’s happening behind the scenes (i.e., that ASC 842 lease rules that are built into Kelvin workflows), it highlights the value of automating the document review process.

Taking the Stress out of Stress Testing

In financial services, stress testing and risk management simulations are critical exercises carried out by institutions and their regulators. In these what-if scenarios, companies evaluate their capacity to withstand extreme but plausible scenarios, such as significant interest rate increases or decreases. In some cases, such as for firms covered by the Dodd-Frank Act, these tests are mandated to be performed every other year, and for many other firms in financial services, risk management frameworks like Basel create similar ongoing requirements.

The process is often complex and involves a rigorous review of vast amounts of contractual data, terms, and condition documents among others. This is precisely where AI can play a transformative role. Through the use of machine learning and natural language processing, AI-enabled solutions like Kelvin offer superior capability in identifying, reviewing, and parsing relevant information from voluminous and complicated financial documents. Massive historic losses, such as Credit Suisse’s 2021 Archegos collapse or JPM’s London Whale, demonstrate that manual processes that rely on human review to identify problematic clauses are not sufficient for an effective top-down risk management process.

Kelvin’s ability to connect with and process numerous systems and file formats means that it can identify key elements such as parties involved, contractual obligations, renewal terms, interest rates, penalties, and more, regardless of where the information is located. Moreover, Kelvin can identify subtle nuances in language, identify potential ambiguities, and other “non-standard” or unusual terms that may create risk in a stress scenario.

For example, when looking to identify all financial instruments that have a stated or variable interest rate, Kelvin can find these in emails, spreadsheets, contracts, and other related (or even unrelated) addendum documents and exhibits. The relevant interest rates and related information is then extracted and analyzed automatically, excluding non-applicable instances of interest rates.

As financial firms handle increasingly larger amounts of data with stringent compliance requirements, the power to streamline complex documentation tasks can be a game-changer that enables resource shifting from document review to higher value activities.

Recognizing the Potential for Revenue Recognition

While we all loved a good game of hide and seek as children, it becomes less fun as an adult when you’re looking for something hidden in thousands of documents and your prize for finding it is accurately presented financial statements. Revenue recognition entails combing through documents to ensure that sufficient information is available to determine an organization’s performance obligations and the related transaction price. Frequently, the documents that include contractual elements that impact this recognition exercise are scattered throughout various systems and formats, making it extremely difficult to confirm that all of the information for a particular transaction has been contemplated.

The potential for streamlining and enhancing the process of revenue recognition with AI-enabled tools is quite significant. Kelvin can parse through a multitude of contracts and other related documents quickly, accurately identifying elements such as performance obligations, transaction price, and the timing of payment.

By embracing approaches like those powered by Kelvin, businesses can get to where they’re going - an accurate, consistent characterization of their financial performance - with a simpler, faster route.

To learn more about revenue recognition, see our post dedicated to the topic of ASC 606.

The Way Out

The maze of financial documents can be a daunting place, but with the right tools, it can be navigated with ease. Large language models and systems that embed them within domain-specific workflows like these can dramatically improve the efficiency and accuracy of financial reporting and risk management processes. In the process, they can help organizations escape the endless maze of financial documents and find their way back to the outside world.

Back to Blog