Moderated by Emily, Digital Transformation Consultant at Hyperbots
Emily: Hi everyone, good morning, good evening, or good afternoon depending on where you are. I am very pleased to have you back. The topic that we will be discussing today is large language models and their applications in accounting. But before we dive into it, Ayo, would you mind telling us a little more about yourself?
Ayo Fashina: Hi Emily, it’s good to see you again. My name is Ayo Fashina, I’m the CFO of Kobo360. It’s an e-logistics company startup, about five going on six years old now. We match goods owners to transporters and operate in seven African countries.
Emily: Got it. Thank you so much, Ayo, for the introduction. Today we’ll be talking about large language models (LLMs). It’s a vast topic in itself, but let’s start with the basics. Ayo, can you explain what large language models are and how they are relevant to the field of accounting?
Ayo Fashina: Thank you, Emily. In short, a large language model (LLM) is a type of artificial intelligence program that can recognize and generate text, among other tasks. LLMs are trained on huge datasets, hence the name large. They are built on machine learning, specifically a type of neural network called a transformer model. They analyze large datasets to learn what to look for when queried and use that knowledge to improve their performance. This makes them highly relevant to accounting, which involves data analysis and data collection.
Emily: Got it. Traditional methods of data analysis in accounting are human-based. Accounting staff or finance analysts do all the data entry and analysis, often using tools like Excel. How do LLMs compare to these traditional methods?
Ayo Fashina: One of the most significant benefits of LLMs is that they substantially increase the efficiency of accounting tasks. These models can process large volumes of data at an unprecedented speed, drastically reducing the time accountants spend on routine tasks such as data entry, transaction categorization, and report generation. Essentially, LLMs make the life of an accountant easier.
Emily: Understood. Can you dive into some specific applications of LLMs in accounting that you’ve come across or implemented?
Ayo Fashina: One application I’ve come across involves processing accounts payable. LLMs are trained on data from invoices, purchase orders, and delivery notes. They can match these documents to process accounts payable on behalf of accountants and even handle general ledger entries. Impressively, these models can read both typed and handwritten documents, which is a significant advancement.
Emily: How do LLMs contribute to improving efficiency, accuracy, and decision-making in accounting tasks?
Ayo Fashina: LLMs can process a lot of data very quickly and with high accuracy, thereby minimizing human errors. They ensure consistency in handling transactions and applying rules uniformly, which is crucial in accounting. By automating routine tasks, LLMs allow firms to allocate human resources to more strategic tasks, enhancing overall productivity. They also provide valuable insights for financial forecasting and decision-making by identifying trends and patterns that might be overlooked by human analysts.
Emily: What challenges do finance professionals face when implementing LLMs in accounting processes, and how can these challenges be addressed?
Ayo Fashina: Despite their advanced capabilities, LLMs lack human judgment and the ability to interpret complex and ambiguous financial situations. They operate based on the data provided to them without understanding the nuances of certain accounting decisions. Therefore, LLMs should be applied to tasks where human judgment is not a high priority. It’s also crucial to have robust data validation processes to ensure the quality of data used to train LLMs, as inaccurate data can lead to erroneous outputs.
Emily: What opportunities do LLMs present for innovation and advancement in accounting?
Ayo Fashina: LLMs hold great promise in revolutionizing accounting and finance by automating routine tasks, enhancing efficiency, and providing valuable insights. They can be used for risk assessment, fraud detection, and even forensic accounting. For example, LLMs can analyze data to uncover hidden patterns and trends, help identify risks, and flag anomalies in transactions, which aids in fraud prevention.
Emily: With the increasing use of LLMs in accounting, how do you ensure the security and privacy of sensitive financial data?
Ayo Fashina: Ensuring data confidentiality and security is paramount. Data hygiene is essential, meaning the data used to train LLMs should be sanitized of any personal identifiable information. Organizations should implement stringent data cleaning and sanitation procedures to remove sensitive information and identify potential biases and errors in the data.
Emily: How important is it for LLM-driven accounting solutions to integrate seamlessly with existing financial systems?
Ayo Fashina: It’s very important. If an LLM solution cannot integrate with existing systems, it defeats the purpose of having the LLM in the first place. Full integration ensures that the efficiency gains from LLMs are realized. Without it, the benefits are eroded by manual data transfer, which reintroduces human error.
Emily: What strategies do you recommend for ensuring smooth integration and compatibility with other accounting and enterprise systems?
Ayo Fashina: Testing compatibility ahead of full implementation is key. The provider of the LLM-driven solution should ensure seamless connection to existing systems and workflows. A sandbox test run before full implementation can help identify and resolve any integration issues.
Emily: What steps should organizations take to ensure compliance with relevant regulations while leveraging LLMs in accounting operations?
Ayo Fashina: Regulatory compliance is dynamic and always changing, which poses a challenge to LLMs. Ongoing monitoring and adjustments are required to keep LLMs compliant, which can be resource-intensive but necessary. Developers should ensure that LLMs are regularly updated to reflect changes in regulations.
Emily: Are there any emerging trends or advancements in LLMs that you believe will shape the future of accounting and finance?
Ayo Fashina: LLMs have the potential to revolutionize the finance sector in numerous ways. They can be used for risk assessment, fraud detection, and forensic accounting. LLMs can also simplify audit processes by organizing data for easier transaction tracing. As these technologies evolve, they will continue to provide valuable insights and efficiencies.
Emily: Based on your experience and insights, do you have any additional advice or recommendations for financial professionals looking to harness the power of LLMs?
Ayo Fashina: Financial professionals should consider learning Python, a versatile programming language that many LLMs are based on. Understanding Python fundamentals can help professionals leverage LLMs for code generation and task automation. As LLMs become more prevalent, the demand for Python skills will grow, making it a valuable skill for finance and accounting professionals.
Emily: Thank you so much, Ayo, for sharing your expertise on the applications of large language models in accounting. Is there anything else you would like to add before we conclude our discussion?
Ayo Fashina: Just to summarize, LLMs hold great promise for revolutionizing accounting and finance by automating routine tasks, enhancing efficiency, and providing valuable insights. However, it’s crucial to be aware of their limitations and the need for human judgment. Data security and regulatory compliance are also important considerations. Thank you, Emily, for having me. It’s been a pleasure.
Emily: Thank you, Ayo. It was great having you, and this discussion was truly insightful.
Moderated by Emily, Digital Transformation Consultant at Hyperbots
Emily: Hi everyone, good morning, good evening, or good afternoon depending on where you are. I am very pleased to have you back. The topic that we will be discussing today is large language models and their applications in accounting. But before we dive into it, Ayo, would you mind telling us a little more about yourself?
Ayo Fashina: Hi Emily, it’s good to see you again. My name is Ayo Fashina, I’m the CFO of Kobo360. It’s an e-logistics company startup, about five going on six years old now. We match goods owners to transporters and operate in seven African countries.
Emily: Got it. Thank you so much, Ayo, for the introduction. Today we’ll be talking about large language models (LLMs). It’s a vast topic in itself, but let’s start with the basics. Ayo, can you explain what large language models are and how they are relevant to the field of accounting?
Ayo Fashina: Thank you, Emily. In short, a large language model (LLM) is a type of artificial intelligence program that can recognize and generate text, among other tasks. LLMs are trained on huge datasets, hence the name large. They are built on machine learning, specifically a type of neural network called a transformer model. They analyze large datasets to learn what to look for when queried and use that knowledge to improve their performance. This makes them highly relevant to accounting, which involves data analysis and data collection.
Emily: Got it. Traditional methods of data analysis in accounting are human-based. Accounting staff or finance analysts do all the data entry and analysis, often using tools like Excel. How do LLMs compare to these traditional methods?
Ayo Fashina: One of the most significant benefits of LLMs is that they substantially increase the efficiency of accounting tasks. These models can process large volumes of data at an unprecedented speed, drastically reducing the time accountants spend on routine tasks such as data entry, transaction categorization, and report generation. Essentially, LLMs make the life of an accountant easier.
Emily: Understood. Can you dive into some specific applications of LLMs in accounting that you’ve come across or implemented?
Ayo Fashina: One application I’ve come across involves processing accounts payable. LLMs are trained on data from invoices, purchase orders, and delivery notes. They can match these documents to process accounts payable on behalf of accountants and even handle general ledger entries. Impressively, these models can read both typed and handwritten documents, which is a significant advancement.
Emily: How do LLMs contribute to improving efficiency, accuracy, and decision-making in accounting tasks?
Ayo Fashina: LLMs can process a lot of data very quickly and with high accuracy, thereby minimizing human errors. They ensure consistency in handling transactions and applying rules uniformly, which is crucial in accounting. By automating routine tasks, LLMs allow firms to allocate human resources to more strategic tasks, enhancing overall productivity. They also provide valuable insights for financial forecasting and decision-making by identifying trends and patterns that might be overlooked by human analysts.
Emily: What challenges do finance professionals face when implementing LLMs in accounting processes, and how can these challenges be addressed?
Ayo Fashina: Despite their advanced capabilities, LLMs lack human judgment and the ability to interpret complex and ambiguous financial situations. They operate based on the data provided to them without understanding the nuances of certain accounting decisions. Therefore, LLMs should be applied to tasks where human judgment is not a high priority. It’s also crucial to have robust data validation processes to ensure the quality of data used to train LLMs, as inaccurate data can lead to erroneous outputs.
Emily: What opportunities do LLMs present for innovation and advancement in accounting?
Ayo Fashina: LLMs hold great promise in revolutionizing accounting and finance by automating routine tasks, enhancing efficiency, and providing valuable insights. They can be used for risk assessment, fraud detection, and even forensic accounting. For example, LLMs can analyze data to uncover hidden patterns and trends, help identify risks, and flag anomalies in transactions, which aids in fraud prevention.
Emily: With the increasing use of LLMs in accounting, how do you ensure the security and privacy of sensitive financial data?
Ayo Fashina: Ensuring data confidentiality and security is paramount. Data hygiene is essential, meaning the data used to train LLMs should be sanitized of any personal identifiable information. Organizations should implement stringent data cleaning and sanitation procedures to remove sensitive information and identify potential biases and errors in the data.
Emily: How important is it for LLM-driven accounting solutions to integrate seamlessly with existing financial systems?
Ayo Fashina: It’s very important. If an LLM solution cannot integrate with existing systems, it defeats the purpose of having the LLM in the first place. Full integration ensures that the efficiency gains from LLMs are realized. Without it, the benefits are eroded by manual data transfer, which reintroduces human error.
Emily: What strategies do you recommend for ensuring smooth integration and compatibility with other accounting and enterprise systems?
Ayo Fashina: Testing compatibility ahead of full implementation is key. The provider of the LLM-driven solution should ensure seamless connection to existing systems and workflows. A sandbox test run before full implementation can help identify and resolve any integration issues.
Emily: What steps should organizations take to ensure compliance with relevant regulations while leveraging LLMs in accounting operations?
Ayo Fashina: Regulatory compliance is dynamic and always changing, which poses a challenge to LLMs. Ongoing monitoring and adjustments are required to keep LLMs compliant, which can be resource-intensive but necessary. Developers should ensure that LLMs are regularly updated to reflect changes in regulations.
Emily: Are there any emerging trends or advancements in LLMs that you believe will shape the future of accounting and finance?
Ayo Fashina: LLMs have the potential to revolutionize the finance sector in numerous ways. They can be used for risk assessment, fraud detection, and forensic accounting. LLMs can also simplify audit processes by organizing data for easier transaction tracing. As these technologies evolve, they will continue to provide valuable insights and efficiencies.
Emily: Based on your experience and insights, do you have any additional advice or recommendations for financial professionals looking to harness the power of LLMs?
Ayo Fashina: Financial professionals should consider learning Python, a versatile programming language that many LLMs are based on. Understanding Python fundamentals can help professionals leverage LLMs for code generation and task automation. As LLMs become more prevalent, the demand for Python skills will grow, making it a valuable skill for finance and accounting professionals.
Emily: Thank you so much, Ayo, for sharing your expertise on the applications of large language models in accounting. Is there anything else you would like to add before we conclude our discussion?
Ayo Fashina: Just to summarize, LLMs hold great promise for revolutionizing accounting and finance by automating routine tasks, enhancing efficiency, and providing valuable insights. However, it’s crucial to be aware of their limitations and the need for human judgment. Data security and regulatory compliance are also important considerations. Thank you, Emily, for having me. It’s been a pleasure.
Emily: Thank you, Ayo. It was great having you, and this discussion was truly insightful.
Conversational UX is gaining traction in tandem with rapid advancement in AI tech. It seems intuitive that humans would want to communicate with AI agents or bots as naturally as possible. Nothing about conversational UX is new, of course. We just happen to be at a tipping point where various AI technology trends are pushing it into prominence. Substantial research and successful application of that research for real-world scenarios over the past decade have made conversational UX ubiquitous and ready for primetime where the best is yet to come.
The new age of sophisticated applications using generative AI demands that designers dig deep into the art of conversation. This is an exciting time to explore possibilities to make a dialogue between humans and bots, natural, meaningful, fun, and engaging. On the B2B SaaS front, we are just scratching the surface.
With AI infused in all kinds of finance process automation, there are a ton of possibilities to make conversational UX a key part of such applications. It begs a radical question. What if there is no traditional UI layer in finance applications? Can business outcomes be achieved through good old easy-going conversations between humans and AI solely through a chat window with no regular app interface to speak of? How can designers design and orchestrate the creation of these environments? Designers must investigate, for instance, whats the equivalent of a casual business meeting in a cafe versus a mission-critical exchange in a conference room about budgeting between a CFO and their AI assistant.
At Hyperbots, designers are exploring ways to create a humane, relatable avatar for the powerful AI capabilities of our platform addressing the automation needs of finance processes like Accounts Payable and Expense Processing for the CFOs office processes that are still woefully manual.
The core work that the Hyperbots AI Assistants do are:
Accountant queries can be as practical as asking the AI assistant about invoices that can be safely bulk-approved or the ones that need their manual review.
The AI assistant reponds with real-time actionable data about pending invoices that need manual review.
Analytics critical to business decision making can be easily pulled up with a simple query.
These are sneak peeks into the early work that’s emerging as part of the conversational UX design charter for the design team at Hyperbots. Within broad chatbot categories that exist today, here’s where we might fit in.
Designers at Hyperbots know that if they want to create distinct AI Assistant identities, they need to focus beyond the visual elements of an avatar or the UI layer of a dialogue box. They must ask the question what makes a dialogue meaningful? Especially between a machine and a human. They must dive deep into the science of Human-Computer Interaction and the art of conversation. So far our secondary design research has pointed to some seminal work already in the public domain like the recent ethnographic study by NNGroup into usage patterns of ChatGPT, Bing, and Bard users suggesting there could be 6 different types of conversations with generative AI.
These provide a great basis for brushing up on fundamentals and taking the right first step. What should follow is arriving at a solid hypothesis of what specific approaches might work for ur CFOs and their teams and then testing these hypotheses with rigor.
We are early in our exploration of conversational UX at Hyperbots. We are more than convinced this space cannot remain untapped if we are to create a groundbreaking experience for our customers grappling with legacy applications to conduct their finance operations central to the customer experience we want to build for our CFOs and their teams. As they say, watch this space!