Generative AI: A Game-Changer for Expense Management Automation?



Part 1: Expense Management in the Age of Gen AI: Addressing Shortcomings and Maximising Potential 

Generative artificial intelligence (Gen AI) has set the world alight since the commercial launch of OpenAI’s Chat-GPT in late 2022. Since then, Gen AI has found a wide range of applications in fields ranging from generating software code and creating marketing content to powering smarter customer service chatbots.  

There is no doubt that Gen AI and large language models (LLMs) will play a key role in digital transformation in the future. With that in mind, we will be writing a few blog posts to outline our perspective on Gen AI and how it can enrich cutting-edge FinOps tools and expense management automation solutions and user experiences.  

In this first post, we’ll outline the challenges that LLMs and Gen AI face in creating AI-driven insights and elevating expense management automation. However, we preface the discussion by stressing that we believe LLMs will enable us to deliver a more personalised and more automated expense management user experience in the near future.  

Gen AI—A New Kind of Intelligence 

AI and machine learning are by no means new—traditional AI that focuses on analysing historical data and making future numeric predictions is already used for applications such as fraud and anomaly detection in financial software. Gen AI, however, is a new form of intelligence that can create new content, including text, images, videos, and more. 

Gen AI tools such as Google’s Bard and ChatGPT are built on LLMs, which are a specific type of AI model designed to understand and generate human-like language. This makes Gen AI tools and models an exciting prospect for applications such as interrogating a FinOps or expense management solution for smarter, more personalised AI-driven insights.  

However, Gen AI has a range of shortcomings related to the fact that it wasn’t specifically designed for mathematical applications. While LLMs may perform adequately in basic arithmetic, their performance can deteriorate when faced with more complex mathematical calculations involved in financial analysis and expense management. LLMs are prone to making calculation errors because they are built to recognise language patterns rather than do math. This is a significant challenge in expense management automation, where financial data integrity is of paramount importance.  

Seeing Things That Aren’t There 

A related challenge in using Gen AI for expense management automation is that AI systems can have ‘hallucinations.’ This refers to instances when the machine generates false, misleading, or nonsensical information. These errors arise because AI doesn’t understand the context of its source data or was fed noisy or incorrect data.  

This can lead to inaccurate data interpretation, misrepresentation of expenses, or the creation of entirely fabricated transactions. The system might misclassify expenses, duplicate content, and in extreme cases, generate fictitious expenses that do not correspond to any actual records. Such mistakes pose a serious risk to financial data integrity.  

Another crucial issue lies in the transparency, quality, and veracity of the data used to train LLMs. The performance of LLMs depends on the quality and diversity of the training data. If the training data lacks sufficient coverage of financial or numerical contexts, LLMs may exhibit suboptimal performance or biased behaviour in outputs.  

Feeding the Machine 

If an expense management automation solution leverages an open-source AI model or a commercial Generative Pre-trained Transformer (GPT) like ChatGPT as the foundation, it has less control over the data used to generate the outputs. This data may be too noisy and not specific enough to train a solution designed for a focused use case like expense management. 

While it’s essential for an LLM to be trained on sufficient and large volumes of data, it’s also important to be cognisant of the dangers of ‘drift.’ This is a phenomenon where the statistical properties of the data used to train a machine learning model change over time, leading to degradation in performance. There are concerns that drift is making ChatGPT dumber.  

Expense management automation software also draws on a company’s internal data for real-time expense tracking. There are many pitfalls to consider here as well. Data recording errors may occur if the AI systems incorrectly record transactions or misclassify expenses due to algorithmic biases or contextual misunderstandings.  

Compliance and Governance Issues 

Data retrieval issues may arise when pulling data from diverse sources, which can ultimately lead to incomplete or erroneous analyses. These issues are caused by factors such as data fragmentation, inconsistency, incompatibility, and accessibility constraints.  

Finally, from a compliance and governance perspective, many organisations will want to tread carefully in sharing real-time expense tracking data with a solution that shares information with a public LLM. They will want to know that any cloud-based software they use has measures to ensure confidential data is not shared with the outside world.  

The Future of Gen AI and Expense Management Automation 

Enterprises in most sectors are relatively cautious about operationalising AI platforms with good reason. Their outputs are far from perfect, and any mistakes could be expensive and harmful to a company’s reputation.  

Yet we believe that responsible use of Gen AI and traditional AI for AI-driven insights and customisable workflows will be game-changers for expense management automation. Focusing on data quality and ensuring that AI has human oversight are among the keys to success. Additionally, building your own data model and controlling how Gen AI models consume and compute financial data is crucial to addressing these issues. This approach allows for better management of data integrity, reduces the risk of errors, and ensures that the AI outputs are reliable and accurate.  

In part 2 of this series, we will explore the challenges but also the massive potential of Gen AI in expense management automation. We will delve into how understanding these challenges can help in harnessing the full capabilities of Gen AI to create robust, efficient, and innovative solutions. 

Looking for a partner in FinOps and expense management automation to support your digital transformation journey? Get in touch to learn how we can help you simplify operations and accelerate your cloud journey. 

Similar Blog Posts