by Viola Hechl-Schmied, Product Owner for Machine Learning at ION Treasury.
Since the launch of ChatGPT in November 2022, conversations have centered on how these kinds of artificial intelligence (AI) tools may change our private and professional lives. The trend is no different for the treasury function, which has seen the development and adoption of machine learning (ML) technology increase rapidly in recent years. AI and ML tools have been used in treasury departments for some time in payment transactions and trade finance. However, their ability to transform the treasury function has been tempered by the complex nature of risk forecasting. The next step needs a human-AI hybrid approach – placing augmentation at the center of technology strategies.
In today’s complex geopolitical and macroeconomic climate, treasurers are turning to digital transformation to combat market volatility and the time taken by typical data- and labor-intensive tasks. So, what are the benefits in employing AI and ML in treasury, and how can treasurers deploy these tools for optimal results?
Faster, better, stronger
The most obvious application of these technologies in treasury lies in reducing the time taken for repetitive tasks, such as cash forecasting, and cash tagging. Creating accurate cash flow forecasts in today’s uncertain market conditions can be challenging, particularly for the long term. In response to this market volatility, treasurers make more frequent forecasts and with a shortened forecast window. For organizations facing skills or staff shortages, undertaking more cash forecasts within tighter timeframes, with access to fewer resources, is challenging.
By deploying AI- and ML-based automation tools, companies can automate financial processes across the business, moving away from manual- and labor-intensive spreadsheets and bank portals. ML algorithms can rapidly analyze and predict patterns from large volumes of cash transaction data originating from different sources – such as accounts payable, accounts receivable, or bank account balances. Repetitive tasks – such as data processing and analysis – are carried out faster and more effectively, streamlining the cash forecasting process.
Importantly, ION Treasury data shows that cash forecasts can be done 3,000 times faster using ML techniques. Likewise, automating repetitive tasks – such as cash forecasting and cash tagging – allow treasury teams to dedicate more time to higher-level, more strategic work.
Quality data in, quality data out
ML is central to unlocking the power of an organization’s data. However, the efficacy of any ML model depends entirely on the quality of the data it is trained with and operates on.
Therefore, although the treasury system can facilitate data storage, treasurers must also ensure that elements are appropriately configured to support the necessary decision-making processes. This is the case with cash flow types, where ML can also help by automatically tagging transactions without the need for pre-defined rules.
For ML and AI to work effectively, it is crucial for a treasury function to possess well-structured and relevant data that is codified to the enterprise’s needs.
Augmentation vs automation
AI and ML may be able to approximate human output with a high degree of accuracy in manual data entry and processing tasks. However, a more collaborative approach is necessary for generating market forecasts that require judgment and context.
AI and ML can provide general analysis and identify patterns, but these tools cannot intuit and apply contextual understanding, or comprehend broader economic trends, regulatory shifts, and market dynamics. When events deviate from historical patterns – such as the unprecedented COVID-19 pandemic or 2023’s “mini-budget” – models are much less able to adapt and adjust. Yet, these anomalous market events are crucial for treasurers to understand the risk landscape, or anticipating liquidity or cash shortages.
To sense-check the use of AI and ML, generated forecasts can be compared or back-tested against actual cash flows to see where the patterns inferred by AI and ML are relevant, and where more manual intervention may be needed. For example, AI-powered transaction tagging models are able to classify transactions correctly and can detect payment outliers through deviations from learned behaviors and patterns. Treasurers can interpret and apply the outcomes to enhance their decision-making.
A balanced approach
As the uptake of AI and ML tools continues to grow, treasurers need to find the right combination of efforts from machines and humans to yield the best results for their organizations. It’s all about augmentation.
Human judgment will always be necessary to execute a strategy successfully, but AI and ML can greatly reduce the time taken and likelihood of errors. By finding the right balance of machine and human efforts, treasurers will be able to unlock the full power of these tools and achieve the best results for their department.
