More than a Feeling

How AI is Creating Real Treasury Opportunities

Explorations of AI in treasury management yield many questions for the cautious and adventurous treasurer alike. Barclays’ Divya Kesavan, Head of Corporate Data Science & Insights, Global Transaction Banking, and Ope Olomo, Innovation Director, European Transaction Banking, present a balanced view of AI’s role in treasury and beyond.

Many corporate treasurers know they have certain opportunities to use AI, and RPA. A sizeable number also understand that if their preparation and deployment lack proper care, then, at best, they will have an expensive toy on their hands.

The obvious solution to avoid the latter is to set out with well-defined goals. But in a technological field that is evolving rapidly, that may be harder than it sounds. This is the point at which the wise will consult the voices of experience, and preferably stay in touch throughout the journey.

Both Kesavan and Olomo, as Data Science & Insights Head and Innovation Director respectively, fulfil the criteria required of ‘voice of experience’ and as such, harbour views that carry significant weight. So, what, from their perspectives, does AI have in store for treasurers, and how can its promise be safely delivered?

From the outset, Olomo notes that many more potential users now understand that AI has a huge capacity to offer productivity gains, “especially in terms of how they can carry out their duties in a smarter and more efficient manner”. Indeed, citing a report published by Barclays Research and IBM, he says this opportunity was a key benefit referenced by Christian Keller, Head of Economics Research at Barclays.

“For treasury teams, it’s about leveraging AI to become more efficient by being able to use their data more effectively to enhance decision-making,” he explains. For example, treasurers can use data from their ERP systems (payables, receivables, historical collections and payment transactions) with AI application to forecast currency hedging needs, enhance strategic approach to cash management and decision making for working capital planning. “But the article also shines a spotlight on how teams can make better use of their employees’ strengths to enable speed and accuracy in areas such as financial reporting.”

Cautious excitement

With this in mind, a popular AI application for treasurers is in cashflow forecasting. “AI and machine learning models can help make forecasting more accurate and less time consuming,” notes Olomo. “The arrival of ever-faster processors means we can leverage higher volumes of data than we could ever use before. An understanding of this is helping treasurers begin identifying other opportunities for AI in their space.”

With motivation for treasury to at least consider how they might deploy AI, Kesavan sees firsthand how enthusiasm is mounting. “The team that I lead includes data scientists and analysts. We’re looking at a range of AI and ML use cases and have certainly seen interest from Treasury and Finance functions because they want to know how they can use some of these new technologies to enhance their productivity,” she reveals.

With a significant focus on forecasting within these teams - liquidity and cashflow forecasting, for example – automation projects are high on the agenda, but Kesavan observes that using more advanced forms of machine learning and AI can detect seasonality, and other kinds of noise in the data, much better than any other technology. This is why so many professionals want to tap into it. But, she warns, the enthusiasm must be tempered with a degree of caution.

“AI, and especially enterprise-wide adoption of AI, comes with its own challenges, especially around the management of the data being used. It really is a two-edged sword. AI is an excellent tool to harness in business, but it is vital to ensure that every data source has the right provenance, is free from biases as far as possible, and that the right to use that data is established.”

With “cautious excitement” defining the mood around AI in a treasury context today, the feelings towards its stablemate in this space, RPA, could be described as slightly muted, notes Kesavan. “RPA has found a place within treasury, but I don’t believe the technology has been able to make great strides here.”

There are fundamental differences in the methodologies between RPA and AI - RPA uses rules-based logic whereas AI learns from historical data. From a business use case perspective, a relative difficulty in scaling RPA-based solutions thus emerges. As Olomo adds: “The effort required to keep changing the rules’ logic not only limits RPA’s scalability, but also notably adds to the cost of its management. AI, on the other hand, is self-learning and so requires less effort to derive usable intelligence.”

Another disadvantage to RPA, notes Olomo, is the more demanding nature of its integration with legacy systems. “Many financial services firms and corporate treasuries retain some legacy systems. Integrating RPA with these, especially niche systems, can be quite difficult. RPA still has a role to play but, looking forward, I think more effort will be exerted in reaching AI’s potential.” That said, Kesavan notes a drive in some quarters to explore how AI and RPA can work together in Treasury, extending the utility of both by creating AI-powered RPA. It’s one to watch.

I spy…

“One of the appealing aspects of AI is its capacity for pattern recognition,” observes Olomo. “An AI model could easily identify complex patterns and detect anomalies or exceptions within them. In addition to cashflow forecasting, an increasing area of interest for treasury is its application as a risk and control tool, identifying errors, duplications, or even potential fraud.”

Another emerging treasury use case Olomo observes is in predictive analytics and the creation of self-learning hedging models. The capacity to consume and process vast data pools, and identify within macro patterns, enables treasurers to “make a call on the fly” around currency protection, for example.

The evolving use cases of AI, and to a degree RPA, should have a positive impact on treasury now and in the future, comments Kesavan. She believes GenAI in particular will have a major role to play in the treasury productivity and knowledge management space. By speaking to treasury professionals, she says it has become apparent that, throughout the entire treasury process lifecycle, obvious candidate areas for automation and the deployment of AI are developing. Resistance to these technologies is slowly beginning to dissolve.

Indeed, Kesavan continues, transformative technologies follow a pattern in that they are first embraced by consumers for personal use and later adopted in the corporate world such as the usage of smartphones and corporate mobile banking by treasurers,. Similarly the use of tools such as ChatGPT in their personal lives is beginning to bring that acceptance into the office. Then, seeing firsthand how these tools can substantially reduce the time taken to perform certain tasks, such as report generation, compelling business cases naturally develop.

So, while the understanding of potential gains in the knowledge management space is gathering pace, so too is the view that AI can significantly power-up data as an asset. “Financial services providers, including banks, and many corporates too, are realising the value of their data, and the extent to which this can be commercialised,” comments Kesavan. “AI is the tool that will help accelerate data monetisation.”

Believing that quality data fuels AI, Kesavan suggests that the first step towards realising data’s value is to ensure that it is arranged in a way that best enables AI models to be built across that structure. This means systematic consideration of the most effective means of collecting, cleaning, organising, protecting, and storing that data.

Doing so, she notes, will not only ensure data accessibility is optimised for all, including treasury, but also meet the needs of, for instance, the extremely useful ‘next-best-action’ AI advice modules now offered by many system providers to service each user’s own needs, preferences, and context.

AI probe

Most treasurers looking at AI today should have a number of questions to ask of their own organisations, their banking partners, and their technology vendors. To this end, Kesavan has formulated a “three-part initial framework” encouraging the probing of “process, people and platform”.

‘Process’, she explains, is simply about asking how AI is going to enhance what the treasury organisation is currently doing. “It’s essential to consider how existing processes will be managed without disintermediating anyone. Failure to address process issues will see increased resistance, and that makes it harder to drive adoption.”

‘People’ is about exploring upskilling, because while AI may generate a lot of enthusiasm, it can also bring a perceived threat to employment (see Positive outlook P4). “There needs to be the understanding that this technology should not be about displacing people’s jobs but instead be about automating manual and mundane tasks so their time can be spent on more value-add tasks,” she clarifies.

The ‘platform’ element calls for reflection on how AI will be technically and commercially integrated, especially where legacy or siloed platforms are still live. Many vendors today have an AI offering, but it’s important to understand how it can be integrated into existing platforms, says Kesavan. “I recommend treasury brings data, technology and architecture teams into the discussion. They need to assess how easy it will be to deliver data into the hands of the people actually using it.” She adds that the best way to start a project is to pick a low-risk use case and carry out a small proof of concept  that will help assess and quantify business value. “If successful, only then should the project progress to full implementation.”

With data management being front and centre for all AI deployments, Olomo cautions on the need not only to identify the required internal and external data elements, but also how to obtain these responsibly as an associated issue. He advises treasurers to “stay close to their compliance teams, to ensure they fully understand the regulatory provisions that impact their acquisition and use of data within AI”.

Naturally there will be an appraisal of the cost of any AI project. The hardware and software required to run its models can attract a premium but even if not, as a relatively new technology, they are often subject to frequent updates, cautions Olomo. “Even if you procure a system today, if you’re not careful, in a year or two’s time, it might become obsolete.”

To mitigate that risk, he urges treasurers to identify which aspects can be safely outsourced to cloud service providers, and what the right conditions are for outsourcing these functions. This can enable companies to future-proof their technology stack against the rapid pace of advancement in this space in a cost efficient manner in the long run.

Building business cases

The challenge for many treasurers once they have found an appropriate path to AI and RPA adoption will be securing budget. Promises of productivity gains, for example, need to be quantified, says Olomo. “That could be expressed in terms of time-savings for certain tasks, because the capacity to do more with less is compelling for most business leaders.”

AI may also provide treasury with the ability to innovate faster, he continues. By analysing individual processes using AI, it may be possible to reveal quick and simple optimisations that could otherwise not have been discovered.

Yet another evolving use case for AI, maintains Olomo, is in the field of counterparty assessment in treasury and beyond. Effective evaluation of the credit worthiness, or even the suitability of a counterparty from a KYC or AML perspective, will involve a significant volume of data. This must be examined to ensure that compliance and legal requirements are being met. However, he notes, this data is often paper-based, unstructured – not computer-readable – and it takes time and effort for humans to extract the necessary information. Of course, the more voluminous the document, the greater the risk is of missing key points.

“Alternatively, GenAI can rapidly read large tracts of unstructured data, and collate what’s needed into an analysable format. This will accelerate decision-making while overcoming human limitations,” he comments. AI can be used to read documents like company profiles, financial statements, and credit memos in a faster and more efficient manner than humans, addressing human factors like fatigue and error. “And by applying AI to credit-risk analysis and third-party risk analysis of counterparties, it can facilitate far greater confidence in the decisions made by treasury or those of its procurement or sales partners, for example.”

Drawing parallels between treasury’s AI approach, and the organisation’s wider business strategy, can be an effective means of securing buy-in, suggests Kesavan. “It’s constructive to explain that treasury is not deploying AI for the sake of AI, but in fact to better enable current business strategy.”

Indeed as an example, a corporate with a focus on sustainability that is positioned to leverage AI and automation could dramatically enhance its reporting of internal and external ESG data. The wider implications of facilitating such opportunities should not be missed by treasury.