ChatGPT for Treasury: The Good, the Bad, and the Scary
ChatGPT, the experimental chatbot dominating the headlines, has some interesting—and, in some cases, dangerous—implications for treasury management. This blog discusses the good, the bad and the scary about ChatGPT for treasury.
What is ChatGPT?
ChatGPT is a generative artificial intelligence (AI) natural language processing tool created by San Francisco-based research lab OpenAI. In January 2023—just two months after its launch—the chatbot reached 100 million monthly active users. OpenAI and ChatGPT gained further notoriety by recently partnering with Microsoft for a multi-year alliance valued at $10 billion.
At its heart, ChatGPT is very simple; it writes natural language (e.g. paragraphs and articles) using data it was trained on, including various books, articles and websites. ChatGPT is capable of crafting rather impressive, conversational prose—everything from blog posts to essays to poetry that often appears as if it was written by a real person. It’s so efficient that it recently passed a Wharton MBA exam.
Google has also taken notice; even though ChatGPT is not a search engine, it is already being viewed as competition for the online search giant. Google’s management has developed a chatbot of its own, though in a recent demo, it made a mistake that has already cost the company $100 billion in market value.
From a treasury perspective, here are some interesting applications that Kyriba is monitoring to understand more about ChatGPT for treasury management:
Payments Fraud Prevention
Cybersecurity firm Darktrace shared with Fortune’s Eye on AI that cybercriminals are using natural language AI algorithms to help increase the linguistic complexity of phishing emails. Interestingly, Darktrace reports that while the overall volume of phishing schemes it has observed has decreased by 50%, more complex schemes have increased by close to 20%. Unfortunately, this makes sense.
Check Point, another cybersecurity firm, noted that using ChatGPT helped it to create an end-to-end social engineering campaign from phishing emails all the way to embedded malware within an email attachment. The result was disturbingly convincing and is assured to break down traditional barriers of defense that treasury and finance leaders have erected to detect fraud.
In response, cybersecurity experts all say the same thing: use automated systems to detect and prevent cyberattacks at machine speed. For treasury terms, this means using AI within our payment processes to ensure payments are compliant with all payment policies and employing AI-driven adversarial networks to detect suspicious payments.
The good news is that these tools exist today and can be embedded within your payment software, as well as treasury and ERP-to-bank connectivity.
Treasury Management Systems (TMS)
ChatGPT can also be used within a Treasury Management System (TMS) where the user gives instructions to the system using keywords or questions. With a user experience (UX) that has been optimized for natural language processing, the TMS can respond to basic queries such as “Show me global bank balances converted to USD,” or “What is my exposure to the Yen?” to more complex requests, including “What caused the variance in my forecast last week?” or “How many days of liquidity do we have left?”.
One example Kyriba is testing is for bank reconciliation, where ChatGPT and similar generative AI tools would identify forecast to actual variances and automatically reconcile those transactions that would typically have been manually processed in the cash management module of any TMS.
For the daily user, ChatGPT or similar technology could replace mouse-clicking to reach a menu item or open a screen, as well as offer a next level of treasury automation such as robotic process automation (RPA) offers some treasury teams today.
For the executive user, the benefits could be even more interesting, as many CFOs and treasurers do not login to TMS or ERP platforms. Yet, they would be more than willing to ask questions of their treasury software.
When it comes to ‘workflow’ systems like TMS, ERP and other multi-screen applications that treasury relies on, ChatGPT offers the next level of extreme automation.
ChatGPT for Treasury Documentation
A challenge for many treasury teams is fully documenting treasury processes and procedures, especially when implementing a treasury transformation inclusive of a TMS. Documentation, how-to manuals, and even “when I’m on vacation” instructions take time and effort to compile. This is typically done manually and is a key reason why treasury system implementations can drag on for months.
Formal treasury policies are also particularly challenging. For these policies to be effective, they need to be detailed, yet broad enough that they don’t allow for employees to find loopholes they can exploit. Regular updates may also be needed, as key risks (FX, interest rate, fraud, etc.) continue to evolve. This generally requires a lot of manual work that treasury practitioners would rather avoid.
Fortunately, ChatGPT and similar generative AI models can do all the writing for you. Fed by a minimum amount of information about your policies and the technology platforms that you are using to automate your treasury and payments, the AI model will write your treasury documentation for you.
An additional treasury documentation example would be RFPs. “I love writing RFPs” is a phrase that no treasury professional has ever said. Fortunately, ChatGPT can write RFPs for you, from sharing company background and communicating requirements to suggesting questions that vendors can respond to.
Use ChatGPT for Treasury Education Services
ChatGPT may also prove useful for treasury professionals seeking to expand their knowledge of the profession. Treasury is an ever-evolving field as technology and economic conditions change. And it can often be challenging for treasury folks to gain more education if they are too busy to attend a professional conference or sit in for an online course or webinar.
But a tool like ChatGPT can provide a quick and convenient source of advanced treasury knowledge. By simply typing in questions such as, “When should I hedge?”, “How do I reduce bank fees?” or “Which banks support SWIFT’s UETR tracker?”, treasury professionals can use generative AI to educate themselves online.
ChatGPT for Treasury Management: Today and Tomorrow
While AI is not new technology for treasury, many treasury teams are still exploring how much they need AI to assist with core functions. However, apps like ChatGPT provide a convenient—and currently, free—solution for assisting with important tasks like documentation that typically require hours of manual work. At the same time, they also add another threat element that treasury professionals need to be aware of.
Technologies like ChatGPT are still new and should be handled with care, but they cannot be ignored. While AI is not a replacement for actual employees, treasury teams should take the time to learn about AI and determine how it might be applicable to their organizations.
Figure: A list of example prompts treasurers can use ChatGPT for treasury management tasks – generated by ChatGPT.