Fuel for the Treasury:

While the first major revolutions in the world of work took years and decades to fully gain a foothold, the artificial intelligence (AI) revolution is penetrating the world of work and other areas of life with remarkable speed. Put simply, AI describes the ability of computers to solve certain tasks with "human-like" intelligence. Tools and computers based on AI models are already being used in various business areas. Companies and system providers have also already discovered the first areas of application for this technology in treasury.

Nevertheless, the dynamics and speed with which AI solutions are transforming a wide range of business processes are only partially perceptible in treasury departments. The areas of application are currently unclear, and the question of what the technologies can be used for remains largely unanswered. A recent study conducted by HSBC together with DerTreasurer shows that a third of all treasury departments are not even looking at the new technologies and opportunities.1 It is therefore high time to take a closer look at the potential, but also the challenges and risks of using AI solutions for treasury departments.

Hardly any other topic currently occupies companies of all sizes as much as the concept of artificial intelligence - or AI for short. The explanation of its underlying logic and the differentiation from related terms alone could probably fill several books these days. A detailed technical definition is therefore deliberately omitted at this point. The elasticity of the word AI in relation to its abbreviations and terms is very high, especially in everyday language. In order to be able to assess the potential and risks of this solution for treasury, it is nevertheless important to at least provide a rough definition of key terms in the AI universe.

From a technical perspective, artificial intelligence is a branch of computer science and describes the idea of transferring the problem-solving and decision-making abilities of humans (i.e. human intelligence) to computers and machines. Unlike familiar automation solutions (such as those in the field of robotic process automation (RPA)), where a machine is given individual steps and processes them in a specific order, AI-based tools are based on algorithms. Such tools can, for example, independently adapt the parameters of a decision to a specific problem. In essence, the AI attempts to find general patterns in the data to be analysed, structure them and thus independently derive rules. The resulting decision trees are used by the AI to continuously learn and optimise, ultimately solving even complex problems in a "human-like" way.

Contrary to popular belief, the terms AI and AI (artificial intelligence) are not clearly distinguishable concepts, but merely a marginal semantic distinction. Essentially, both terms have the same meaning and can be used as synonyms for each other. The term "machine learning" is also frequently used in connection with AI. Machine learning can basically be defined as a subset of the AI universe.

While AI defines the overarching concept of how a system or machine "works", machine learning describes a set of methods that are used in the development of those systems or machines - in other words, a way in which AI can be implemented. As the term suggests, these methods focus on the learning process of the systems. When designing such methods, no clear programming commands are issued, but rather the system is enabled to "learn" with the help of algorithms, huge data and insights.

Compared to AI programmes, automation solutions have already been established in treasury departments for some time. The areas of payment transactions and trade finance should be mentioned here as examples, but in practice, corresponding solutions are used in many more areas of treasury.

Payment transactions

In payment transactions, payment instructions are already frequently automated in order to make the payment process faster and more efficient. Automation allows payments to be processed more quickly, which leads to a reduction in manual errors and greater accuracy. In addition, automated email notifications can be sent to relevant employees to communicate treasury actions or upcoming deadlines. This allows employees to react quickly to important events such as payment deadlines and ensure that all necessary actions are taken in a timely manner.

Trade Finance

System solutions have also been developed in the trade finance area to automate and standardise processes. For example, many companies use automated workflows for the lifecycle management of their guarantees. In addition to reducing manual errors and increasing accuracy, automation allows compliance requirements to be better met as all processes are standardised and documented.

Automation potential of the TMS

It is not always absolutely necessary to use separate system solutions to utilise automation. The treasury management systems already available in many companies today also enable the automation of selected processes. Treasurers are advised to analyse this potential in detail in order to make the best possible use of the potential offered by the system solutions already in place. Overall, automation in the treasury area can help to make processes more efficient and effective and minimise risks.

In addition to the already established treasury automation solutions, more and more AI solutions are also being used. AI solutions can not only help to make processes more efficient, but also actively assist in decision-making. By analysing large amounts of data, AI models can provide valuable insights and can make independent decisions based on predefined rules and algorithms. By introducing AI solutions, companies can further optimise their processes and become more competitive. An excerpt from the range of solutions available on the market is intended to demonstrate the potential of AI-based systems.

Cash management

In view of high market volatility and uncertainty, more and more treasurers are concerned with their company's liquidity in times of crisis. Predicting cash flows is crucial for effective financial management, well-founded business decisions and long-term success. Machine learning and predictive analytics help to create automated, accurate forecasts, minimise risks and enable optimised decisions. Financing decisions are supported by simulating scenarios and stress testing virtually at the touch of a button. The major providers of treasury management systems have already integrated AI into their cash management and liquidity planning modules. They make use of various modelling options, such as ARIMA(X), additive regression and neural networks. ML tools can recognise complex correlations that may be overlooked by humans. Nevertheless, human expertise in combination with modern tools is important to optimise predictions and be prepared for unforeseen events. Successful application requires clean data, correct driver selection, suitable IT architecture, professional and technical expertise and clear communication. In addition, analytical expertise is crucial in order to interpret the results correctly and translate them into well-founded business decisions. In our article "Cash forecasting with the support of artificial intelligence (AI)" (issue 131), we have already dealt with this topic in detail.

Payments fraud detection

The use of AI in payment transactions is another example of the use of machine learning to recognise anomalies and prevent fraud. More and more system providers are also integrating these solutions into their treasury management solutions. Payment fraud detection software uses AI to check payments against historical payment data from a data source. A machine then analyses this data, enabling it to identify anomalies in future activity. Typically, the solutions provide users with insight into calculation variables and show how the reconciliation of payments is calculated. Users can define their own tolerance values and thus determine the sensitivity for fraud detection. This is important because one of the biggest challenges is to ensure that the system does not produce too many false positives, disrupting the payment process. Through an integrated workflow in the treasury management system, conspicuous payments can first be checked and then either rejected or authorised for further approvals. The use of AI in payment transactions therefore offers many advantages, particularly in terms of fraud detection and improving the efficiency of the payment process.

AI algorithmic trading

AI solutions are also increasingly being used outside the treasury management system. In high-volume derivatives trading (for commodities, FX transactions or interest rate hedges, for example), they can be used to suggest trading strategies and control automated trading systems. This involves adding AI components to conventional algorithmic trading systems. Conventional algorithmic trading refers to the method of trading in which solutions are used to make automated trading decisions. These programs analyse market data and make decisions based on predefined rules and algorithms with the aim of making trading faster and more efficient.

AI-based systems can also independently identify and execute trades, perform risk management and manage the flow of orders. In this way, they can improve liquidity management and the execution of large orders by dynamically optimising size, duration and order size based on current market conditions. What differentiates AI-driven systems from conventional trading is that the AI model learns and adapts to changing market conditions. The time delay caused by human intervention is drastically reduced.

Not only AI in general is currently on everyone's lips, but also the chatbot development ChatGPT in particular, which uses AI to enable text-based dialogue with users. The abbreviation "GPT" stands for "Generative Pretrained Transformer". This is a type of artificial intelligence that is based on the Transformer architecture and has been trained to understand and generate natural language ("Natural Language Processing", NLP for short). With the help of GPT models, texts can be generated, questions answered, texts translated and much more. Anyone who has already used ChatGPT will quickly think of several use cases for treasury: creating and commenting on financial reports, analysing data or processing and summarising contracts. The possible applications seem almost unlimited. Nevertheless, ChatGPT and similar, publicly accessible programmes cannot initially be used in treasury without hesitation due to data protection concerns. A growing number of companies are therefore developing their own secure solutions to meet these challenges. But even with in-house and secure solutions, the question remains: will the treasury department be prepared to share sensitive financial data within the organisation?

One possible answer to this question could lie in the integration of GPT models in treasury management systems. The solution could help to build a bridge between strict data protection and the treasury's need for information. The idea is that the integration could help to process sensitive data without exceeding the limits of internal security measures. The willingness of the providers of treasury management systems to invest in this will show the direction in which the development of this technology is heading. In any case, it remains exciting to observe these developments.

AI is already being used in treasury and will change and characterise numerous areas of the function in the future. Progressive automation is already ensuring that repetitive tasks are carried out efficiently and quickly. AI will further strengthen this trend and set new standards in data processing and analysis. By processing large volumes of data and, in the best case scenario, in real time, the ability to make forecasts will be increased on many levels. This means that more informed decisions can be made for all control elements, such as liquidity or risk. AI-based models have the potential to play an increasingly important role in forecasting market developments, interest rates and currency fluctuations as well as in risk management. By using machine learning algorithms, these models can analyse large amounts of data and identify patterns that are difficult for human experts to recognise. This can help to identify risks more effectively and make informed decisions.

However, the use of AI not only brings opportunities, but also risks and challenges that need to be considered with every new use of AI functionalities. Due to their function as a central point of contact for financial information and key figures, treasury departments regularly collect and process highly sensitive data. Therefore, as with all new implementations in treasury, it is important to keep an eye on data protection requirements. The use of cloud solutions, which are frequently used in AI technologies due to the large volumes of data, should therefore be treated with caution: it should always be carefully checked where the sensitive financial and personal data is stored (i.e. where the provider's servers are geographically located) and whether the data storage fulfils the company's specific data protection requirements. As with all data processing, the garbage-in/garbage-out principle also comes into play. AI systems are also dependent on valid or high-quality input data, otherwise incorrect analysis results may be produced.

Furthermore, the use of AI places new ethical demands on the treasury. The extent to which unintentional bias or discrimination could occur should be scrutinised for each application. One example of this is the use of AI-based models to predict credit risks, where the models can make predictions based on historical data and other factors such as demographic characteristics and credit behaviour patterns. This also raises new legal issues in terms of liability for AI-based decisions in treasury.

The above makes it clear that the increased use of artificial intelligence will undoubtedly not only change systems and processes, but also their users. The requirements profile for a treasury manager will have a stronger technological character in the future. This is because the benefits and impact of AI on treasury must be assessed to ensure that technologies are used in a targeted manner and support the company's objectives in the best possible way. The treasurer will also need to ensure that the data generated and the data processing are valid and reliable. The security and protection of AI systems against cyberattacks must also be taken into account. Critically scrutinising the results of AI solutions will become an essential task for treasurers. Targeted training on how to use AI solutions, for example, could be an effective way of equipping employees for the new challenges.

Nevertheless, the increasing level of automation will allow treasurers to spend more time on strategic considerations and decisions. Instead of data preparation and key figure calculations, the focus will shift more towards the needs of various stakeholders. The requirements of customers, suppliers, banks and other business units need to be recorded and taken into account - ultimately with the aim of ensuring that the company's financial alignment is in line with its business objectives. However, it is essential to emphasise that despite the use of AI, human validation and data verification remain indispensable. The treasurer will need to be able to assess data accuracy and ensure that AI analyses are in line with business objectives.

In a world that is changing faster and faster, artificial intelligence will be the next big revolution. Opportunities such as increased efficiency, flexibility and improved decision-making must be consistently utilised both in general and in the treasury area. However, AI also brings challenges and risks. Artificial intelligence needs to be used and handled correctly, especially when it comes to sensitive data, which is largely located in treasury, among other areas. We will also be addressing this topic in the panel discussion at this year's Digital Treasury Summit 2023 and discussing in-depth insights into the opportunities and limitations of AI applications in treasury.

Source: KPMG Corporate Treasury News, issue 136, September 2023

Authors:

Nils Bothe, Partner, Finance and Treasury Management, Corporate Treasury Advisory, KPMG AG

Karin Schmidt, Senior Manager, Finance and Treasury Management, Corporate Treasury Advisory, KPMG AG

¹Der Treasurer & HSBC, 2023: KI, Robotics @ Co.: So beurteilen Treasurer die Chancen neuer Technologien