AI has undoubtedly been making waves for some time now. However, it's interesting to observe that within organizations, finance teams often appear somewhat perplexed, and/or uncertain about what the emergence of tools like ChatGPT truly means for Finance. Can these advanced AI models genuinely assist finance teams in streamlining their operations? Can they automate routine tasks, expedite data analysis, and foster more informed decision-making?
The response, it seems, is affirmative – but not without reservation. In this article, we will dive into the nuances and practicalities of this AI tool, provide some handy prompts and also shed light on potential pitfalls you must watch out for when using AI.
ChatGPT, or Chat Generative Pre-trained Transformer, is an AI chatbot built on a large language model (LLM.) Launched by OpenAI in November 2022, the bot helps users execute tasks involving complex datasets. Built on an LLM, ChatGPT offers better output the more data it consumes.
Some examples of tasks you can use ChatGPT for are:
ChatGPT can be reliably used to automate simple financial workflows if a human professional reviews its output regularly. Mark D. McDonald, Senior Director Analyst at Gartner says: "ChatGPT is an excellent example of the value that AI can bring, but as it is now, it has some limitations for practical use in finance."
"For example, using it to help a finance leader generate software code or SQL statements is a great use; however, asking it questions about their financials won't work. That’s because algorithms such as ChatGPT can only answer questions that they have been trained to answer. Without the detailed data of a CFO’s organization, they aren’t able to provide accurate feedback.”
Furthermore, it doubles as an efficient search engine, not only facilitating « blue-sky thinking » and brainstorming sessions but also offering the ability to sift through extensive data sets and compile extensive research, all of which can bolster productivity and expedite the decision-making process. Once again, human critical thinking remains essential.
Bottom line: ChatGPT holds the potential to speed up decision-making and raise productivity levels by taking over repetitive manual tasks. Yet, it's important to remember that human judgment is essential for both training and supervising the tool.
Despite its limitations, ChatGPT is helping finance teams automate low-impact routine tasks. This gives them more time to focus on higher-value work like managing cash flow roadblocks or boosting budget efficiency.
But does ChatGPT pose a threat to financial jobs? McDonald thinks it doesn't. "Rather than seeing automation in the finance function as reducing the number of staff needed, look at it as a way to increase the scope of what it can achieve with the same headcount.”
He cautions that finance professionals must embrace advanced technology, nonetheless. "It will replace people who don’t use AI as that becomes the most productive way to do business."
Wondering how you can leverage ChatGPT in your day-to-day work? Here are seven ways finance teams at small to mid-sized businesses can use ChatGPT.
You can use ChatGPT to quickly summarize large structured datasets instead of entering formulas in a spreadsheet. Note: ChatGPT will struggle to summarize unstructured data accurately.
For instance, you can rely on the bot's summary of financial data organized under headers. It will struggle to summarize data if you give it a list of numbers and ask it to find patterns in them.
Prompt: My dataset has expense data for all employees across four departments. Based on this data, please tell me: Which department recorded the most expense? What is the average value of each expense item across all departments?
Financial reports are critical to an organization's success, but writing them is time-consuming. ChatGPT can help you summarize data in easily-read text to deliver maximum impact.
Prompt: Create a detailed report of the data in these tables. This report must include an analysis of trends in these datasets and observations about them.
ChatGPT can retrieve specific data from large datasets with a simple prompt. For example, you can upload a competitor's financial statements and ask ChatGPT to search for Q4 revenue.
Label your datasets since the bot cannot process unstructured data, as we explained previously.
You can even upload multiple datasets and request visualization suggestions from ChatGPT.
Note that ChatGPT cannot create visualizations. However, you can ask the bot for visualization ideas, code snippets, and troubleshooting help.
Prompt: Here is the financial data for Acme Corp and Widget Inc over the past year. What was Acme Corp's total revenue in Q4? What was Widget Inc's total revenue in Q4? Suggest a chart that best highlights trends in both companies' revenues over the past year.
If you keep tabs on a list of macroeconomic indicators or news releases, ChatGPT can summarize the latest developments and trends in a few seconds. You can include them in daily briefings or reports for senior management.
Note that ChatGPT cannot access external links. The best way to get the bot to summarize news for you is to paste the text from your sources and ask it to create a short briefing.
Prompt: Create a 200 word summary of the following text in easily understood language: <<paste text from source>>
Excel is a great tool for organizing and analyzing large datasets. However, writing macros is time-consuming, and troubleshooting errors takes weeks. ChatGPT can quickly resolve these roadblocks and even solve code errors.
Prompt: Write an Excel macro that changes the value of cell B2 to 0. Then, set the value of cell A2 to equal whatever value the user enters in a dialog box. <<continue describing steps.>>
Prompt: Troubleshoot this piece of Excel macro code that is generating this error <<paste error message.>>
Creating SOPs is time-consuming and takes experienced employees away from critical tasks. Some workflows are so complex that creating an SOP becomes a project, producing a time sink.
You can use ChatGPT to create an SOP quickly. Record a video, transcribe it using AI note-taking software, and upload that transcript to ChatGPT.
For example, you can record yourself updating invoice data in your accounts receivable software and walk through different scenarios to illustrate use cases. ChatGPT will transform your transcript into an SOP in a few seconds.
Prompt: Create a step-by-step SOP for invoice data updates based on this transcript <<paste transcript.>>
You can even record a conversation with someone and turn their thoughts into an SOP. For instance, if your CFO wants to view different data in your reports, ask them what they want to see, transcribe your conversation, and let ChatGPT do the rest.
Prompt: Create an SOP for report creation based on this conversation <<paste transcript.>>
Non-financial personnel regularly read financial reports. Explaining complex financial terms in easily understood language is the key to creating engaging reports.
ChatGPT can help you explain complex terms quickly.
Prompt: Explain NPV in language that a person with little financial knowledge would understand.
While ChatGPT does have some obvious benefits for finance practitioners, you must be aware of its limitations before using it.
ChatGPT needs constant human supervision since it can "hallucinate" and make facts up. The bot is a generative AI—it creates output from existing data.
Unfortunately, ChatGPT does not understand what counts as a bad generation versus a good one. It relies on its human operators to correct output and learn further.
ChatGPT relies on data that was publicly available till 2021 to build assumptions. This means you cannot use it to access the latest data.
For instance, you cannot use it to stay abreast with the latest currency market developments.
Like every AI bot, ChatGPT relies on existing data to arrive at conclusions. This makes it prone to biases within that data, potentially leading to incorrect results. For instance, using it to evaluate the insurance applications of two applicants might lead ChatGPT to overvalue their ethnicity or race when calculating premiums.
Anyone can connect their systems to OpenAI's API and integrate ChatGPT into their software. However, ChatGPT doesn't work with most financial platforms right now.
This lack of integration reduces your ability to automate tasks, and you may find that using ChatGPT doesn't reduce workflow times significantly.
OpenAI, the company behind ChatGPT, is under scrutiny for its data practices. The company faces allegations of deceptive data security practices, leaking user conversation data, and stealing data from other companies to build its LLM.
Finance professionals must think twice before using the bot to run an analysis using confidential data. Masking data and removing identifiers is a good step, but do not guarantee anonymity.
Discuss ChatGPT use with your company's executives before incorporating it into your workflows.
AI will grow more powerful as its ability to learn accelerates. Here's what finance teams must understand about generative AI's future.
Predictive analytics has long been a Holy Grail in the AI world. While the idea of a machine crunching unstructured data and forecasting insights is appealing, this picture is far from reality.
AI’s extreme dependence on data hampers its ability to get creative. Simply put, AI knows nothing outside the data it is fed and this makes it unreliable when accounting for edge or improbable cases-a critical part of predictive analytics.
The rapid scale of AI development is a cause for optimism, though. For instance, firms are increasingly turning to synthetic data sources extrapolated from real-world data to train AI and account for edge cases.
ChatGPT is a pioneer. However, it isn't alone. The financial world woke up on March 30, 2023, to Bloomberg GPT, a bot trained for finance.
Bloomberg claims its bot outperforms existing LLMs, including ChatGPT, and highlighted its abilities in performing sentiment analysis and news classification.
While Bloomberg GPT is restricted to institutional finance, ChatGPT will likely face similar competition within different sectors. As a generalist tool, only time will tell if ChatGPT will retain its advantage.
Advances in AI will introduce more automation in finance. However, human input will remain vital as AI augments workflows. For instance, AI assistants will speed up existing tasks like contract drafting and GL reviews.
Risk management will also become more efficient thanks to AI’s rapid analysis capabilities, something that will only get better as technology advances. More sophisticated automated analysis use cases will arise, helping finance teams offer deep insights into their business.
From generating reports to translating financial jargon, Generative AI models are already changing the face of finance in many ways.
Rather than fearing this change, financial leaders have an opportunity to improve productivity and enable faster decision-making. By carefully assessing how these new tools can best integrate with their department's responsibilities, gaining hands-on experience, and adjusting workflows with a realistic view of the technology's limitations, they can harness its potential.
Ultimately, with finance teams under mounting pressure to evolve from mere number crunchers into proactive business partners, AI can be the game-changer that eases the load and paves the way for a more efficient, data-driven, and strategic future.
ChatGPT can be used in finance but only with significant human oversight of its outputs.
You can use ChatGPT for the following:
ChatGPT simplifies rote task execution and can speed up decision-making. It gives finance teams more time to focus on higher-value work, like strategy building and business partnering.