Recap & Recording | The Treasury revolution with OpenAI

Introduction

The live session began with an introduction GenAI. The technology was introduced by Marc Verdonk as a type of AI, steered with natural language inputs. It can answer complex questions, summarize data, and analyze information, thanks to large language models trained on extensive data sets. He highlighted that this technology has the potential to increase and streamline the workflow in finance departments, especially for a CFO where data aggregation and analysis is crucial. GenAI can help interpreting balance sheets, highlight interesting points in charts, and facilitate interaction with data in far more intuitive ways than traditional methods.

Discussion

The conversation moved on to discuss how the treasury community is facing challenges and opportunities with these type of technologies. Joseph began revealing that corporations often hesitate to adopt them rapidly. Specially, when there is no guarantee of security and data privacy when dealing with sensitive information (i.e. banking and/or customers data). Therefore, he emphasized that an organization must ensure there has been a risk assessment and security approvals prior to engaging the usage of these technologies. Nevertheless, Joseph gave an example of how some on his clients started using the beta version of Trovata (Cash Management system that uses GPT-4 API for queries). Initially customers had naïve and simple questions, and once the trust was built, he saw that more complex questions were asked (i.e. Could you help me build a FBAR report). This highlighted the productivity booster and how some activities will be enhanced by these technologies.
Marc reiterated on the risks that GenAI can bring, however, focusing on the integrity of data. He emphasized the critical importance of data reliability as any inaccuracies could lead to erroneous outputs and conclusions. A major challenge with GenAI lies in its potential to induce a ‘hallucination effect’. This could lead one to incorrectly assume that the output is accurate when it is not.

Use case of similar technology

Daria then discussed the recent success of her treasury team in implementing a solution to address the company’s dollar exposure. She explained that they often need to purchase parts for their indoor machines, which they later sell in Europe. This process creates a dollar-based exposure that they mitigate by hedging with forward deals up to nine months ahead to minimize margin volatility. Daria mentioned that they used a nice forecast running in spot. However, with the aid of their data science team and machine learning-based AI technology, they implemented an AI-based solution that improved their accuracy from 70 to 90%. The effect was immediately visible, and she expressed the satisfaction with this first use case. They plan to explore new ways and use cases in the future.

Conclusion

Mark and Joseph shared their perspective on the topic related to treasury, emphasizing the importance of hands-on experience over theoretical knowledge due to the fast-paced innovation in the field. They pointed out that any read material is likely to be outdated quickly. They recommended using existing office tools or Microsoft’s binging system, which offers chat and GPT capabilities, as a safe and accessible starting point to interact with the technology directly. They also stressed that many of the tools discussed are built on this core technology, with examples simply being extrapolations.
Daria encouraged her colleagues to explore and experiment the possibilities Artificial Intelligence in their day-to-day Treasury activities.

The live session concluded with the panelists recommending the adoption of GenAI and AI technologies now, rather than waiting for it stabilize, stating their potential as substantial productivity boosters.