Question Answered: AI in treasury

Published: Jan 2024

“How should treasury integrate AI?”

AI robot selecting dollar currency on map
Onkar Liddar, Assistant Treasurer & Managing Director, Accenture

Onkar Liddar

Assistant Treasurer & Managing Director

Accenture’s treasury is based on SAP which runs alongside the ERP of the organisation. This has been an important baseline in our digital journey because it has created a single instance data lake or source of truth that we trust and has enabled us to leapfrog into using advanced technologies.

We have begun by developing what we call intelligent cash. This is made of three components, the first of which is a comprehensive dashboard, easily available to everyone in the finance function, which lays out where our cash is and in which currency; with which counterparty or if we have a cash holding that doesn’t align with our guardrails.

The second element comprises forecasting. In the past, forecasting would involve taking historical data and streamlining it. Now we use sophisticated algorithms developed by our own data scientists and consultants applied to each individual cash flow around, say, accounts payable, receivables or tax. The machine learns and evolves its algorithms, and our latest GBP forecast was spot on for the short and medium term – it forecast what actually happened. The machine is so sophisticated it can forecast better than what our people have historically been able to do.

Forecasts lead to the third component where the technology is helping us take business decisions and decide the best course of action. For example, it tells us whether to reduce our holding in a particular currency, taking into consideration its own forecasts and external data sources like the FX rate. The machine does the grunt work and people can spend less time on operations and more on thinking about risks that could happen. It also speeds up the velocity of our cash, and treasurers always need money to move quickly and without friction.

There are already many companies that profess to sell the perfect intelligent cash tool but I haven’t seen a fintech out there yet that can really do this. Tools like SAP can facilitate it, but the risk with using fintech is that treasury ends up tailoring its process to fit their model, rather than the other way around. Treasurers are also mindful of how much data they send to a third party to process on their behalf because fintech and technology like ChatGPT requires putting data in the public domain and this is not the right approach for a corporate treasurer. The technology is also developing so fast, tools are unlikely to keep pace with changes in the market. Still, treasurers can’t integrate AI internally and in isolation because it is complex and they have such a busy day job, so the best approach is to use trusted advisors with a proven track record.

Treasury can’t operate on systems alone; it is a context-based discipline and experts will always be required to make final decisions. What AI gives treasury is an opportunity to upskill people in an environment that is becoming increasingly challenging. Take our growth for example, as the business grows, our exposure to currencies, bank accounts, partners and integration work (Accenture is very acquisitive) makes having the right money in the right place at the right time increasingly challenging.

I envisage a point of time where we can engage with the technology using prose. The machine will suggest courses of action and identify issues and we will also be able to instruct it. Once we have identified risk and interacted with it, we will be able to instruct it to exchange this currency; issue a bank guarantee, downsize debt or increase a bond offering. The tool will carry out the transaction and manage the accounting in the background in a utopia for busy corporate treasurers.

Treasurers need to determine where they stand in terms of technology, conscious that it is fluid and changing so fast that they don’t tie themselves in. My advice is to set out a clear roadmap for how you want to operate in the future and start building towards that goal. AI is a fast-moving space; it will be as impactful to how we operate today as was the loom in the industrial revolution. It is transforming our industry.

Nils A. Bothe, Partner, Finance & Treasury Management, KPMG

Nils A. Bothe

Partner, Finance & Treasury Management

The ability of treasury teams to apply AI depends on access to vast data sets from which algorithms then support treasury processes like cash forecasting, confirmation matching, transaction tagging and much more. But this data needs to be accessible and in the right format, and I believe treasury is sometimes struggling to get a seat at the table as corporates set their data strategy.

IT departments typically lead on the data strategy, setting up data lakes and warehouses. If treasury is to make use of at least part of this data – predictive cash forecasting is a good place to start – treasury needs to be part of the initiative from the beginning to articulate how the data is stored and in which format. Treasury departments also need to ensure they have sufficient access to the data. Not all companies have consistent data strategies, and we notice some companies are doing it with their own resources and IT departments, while others look for external help.

Integrating AI into existing treasury technology is the next step and depends on the kind of data and formats. Treasury will need to consider how to feed AI into TMS processes and API interfaces and decide whether a hybrid landscape that uses a specific AI tech is preferable. TMS are mostly standard solutions so integrating AI development initiatives will follow prescribed processes. But it will be interesting to see how the vendor market responds to developments. Vendors won’t engage with a new piece of software unless there is demand in the market and although some vendors are already investing in AI solutions, it remains to be seen how they will be adopted by corporates.

And a lot of our clients are still hesitant about AI. Many are watching what others are doing in a wait and see approach. This reticence is linked to the availability of data, particularly around predictive forecasting, and the expected benefits. Still, AI is not something that treasury can simply switch on and reap the benefits right away. It is a question of beginning and allowing the algorithms to improve over time. Our recommendation is to hone on a particular business case; start deploying it and learn as you go.

We advise clients to develop a roadmap. If they have a treasury solution coming to the end of its life cycle, they could use that point to upgrade, re-evaluating if it is the right TMS or if another solution fits better with innovations in AI. A roadmap means learning about what’s out there, developing a strategy to make use of certain AI components backed by a solid business case. Like digitisation, AI is not something of itself. It needs to follow an economic case.

The deployment of AI will eventually lead to the management by exception principle. It will free up human resources to deal with exceptions, where AI can’t help. Humans will be able to concentrate on more high value tasks, but it might be difficult for treasury to retain the knowhow and talent to deal with exceptions if the technology is doing an increasing portion of the daily tasks.

Tanya Kuznetzova, Director, Treasury and Cash Cycle Transformation, Baptist Health Care

Tanya Kuznetsova

Director, Treasury and Cash Cycle Transformation
Baptist Health Care

When I think about AI, I always try to look at it as the combination of two distinct work streams, but which also work in parallel. One part of AI involves data processing and modelling, working with structured and unstructured data to apply algorithms that deliver forecasting calculations. The other stream involves Natural Language Processing that models our language and speech.

The data processing models have been around for a while in treasury – banks have been using them for years. The most important element of data modelling involves understanding its limitations. These limitations are only just starting to emerge and can be viewed in different ways.

Take Facebook as an example, everyone knows it monitors our feeds, and feeds back to us more of what we have already seen. The technology doesn’t let us see different things unless we proactively search for something new, and this limits our perspective. The same is true of the data that treasury will be and is using – it won’t flag new risks and can limit perspectives.

Take the fact insurance companies use advanced technology to assess risk by installing devices into cars to monitor a driver’s behaviour. This tech then feeds into data models. Does this data give insurers information that is helpful, or does it take into consideration things they don’t need? Treasury can’t just apply the data and forget about it. Treasury must also apply due diligence.

Due diligence is also important to avoid AI hallucinations. In language models the technology predicts the next word, but it doesn’t understand what it is saying. It’s possible that the AI comes up with something that doesn’t exist, and we can’t tell if its real or fake. Fake news is intentional, but AI is just doing its job. It’s possible to embed preferences, like asking the AI to always quote a source to verify what it says. But this is also difficult because we don’t know if the source is true or fake. Again, treasury needs to be careful that the information it takes from AI is real.

The reality is that AI is here, and everyone must get on board. The key is to learn about AI and learn how to collaborate with it so that treasury can perform better. Very soon, simplified language models will eliminate the need for any translation between financial and IT language; the technology will understand what we tell it through speech recognition and write code to program it.

Open AI, the company behind ChatGPT, is currently building the enterprise business applications on top of the language processing models. AI Assistant is an example of this today, offering an agency that allows people to collaborate with AI; have it understand what they ask, and perform various functions based on API connectivity with systems. In my opinion, strategic involvement will be democratised, and all treasury staff will be included. The person who does the work knows best how to improve it – when it comes to AI companies should work with treasury and give treasury time, capacity and a voice to contribute to organisation’s wellbeing.

Nicholas Soo

Head of Payment Products, Global Payments Solutions, Asia Pacific

In my 20 years as a transaction banker, I have witnessed countless examples of treasuries adopting the latest technologies, from the time when payment instructions were sent using fax machines, to payments themselves being sent and received almost instantaneously, to now more treasury interactions being moved onto blockchain.

Treasury’s adoption of the latest technologies is nothing new. For many companies, the treasury department has become more strategic precisely because it has moved at pace to seek out efficiency using technology. The digital transformation of the economy has dramatically increased both the volume and the velocity of payments –at HSBC, we process 142 payments per second, or 4.5 billion per year – which means spreadsheets and manually consolidated reports have become outmoded and are no longer enough to help treasuries meet business expectations.

At a recent HSBC event, our clients from wide-ranging industries shared views on the impact new technologies have had on businesses and treasuries. Clients showed eagerness to understand how emerging technologies can help them grow and differentiate.

One new-economy client has deployed predictive analytics in combination with a vast amount of internal and external data to improve the timeliness and accuracy of FX margins predictions. Another FI client has developed an AI-based engine to increase analytics capability to speed up and strengthen customer onboarding due diligence.

We have seen treasuries adopt basic versions of robotic process automation, machine learning and artificial intelligence. AI can aggregate data from various sources and format, process and analyse a massive amount of granular data to provide real-time consolidated views. In conjunction with some of the industry-wide developments – real-time payments and SWIFT ISO standards – treasurers can obtain greater visibility and insights to improve efficiency, making faster and more informed decisions.

With the right datasets in place, AI algorithms can help treasurers generate more accurate and timely cash flow forecasts. Machine learning can monitor patterns, identify anomalies and flag threats that may not be visible to human analysts.

At HSBC, one way we are implementing AI in treasury is by using machine learning to analyse a client’s historical payments data. When a client initiates a cross-border payment to a beneficiary account, our service FX Prompt automatically checks if the beneficiary account is in a different currency to the client’s – and if so, we invite the client to pay in the beneficiary account’s currency. This simple but powerful piece of intelligence provides the client with greater certainty and transparency of the exchange rate.

Whilst generative AI is relatively new, we are already very excited by the emergence of use-cases. One healthcare client has used sophisticated chatbots to produce comparative studies and recommendations on the selection process of banking partners.

Some of our new economy clients have acquired generative AI start-ups; others are using applications to improve scenario planning and treasury reporting. Generative AI has the power to bring treasury management to another level of efficiency and unlock the potential to be a more dynamic and strategic function of an organisation.

As with any new technology, generative AI will bring challenges as well as opportunities. Treasurers should carefully consider the risks of adopting generative AI, such as model bias, data quality and privacy control, auditability and explainability of decisions, and compliance with local and international standards.

Next question:

“What key issues will impact the corporate bond market in 2024?”

Please send your comments and responses to

All our content is free, just register below

As we move to a new and improved digital platform all users need to create a new account. This is very simple and should only take a moment.

Already have an account? Sign In

Already a member? Sign In

This website uses cookies and asks for your personal data to enhance your browsing experience. We are committed to protecting your privacy and ensuring your data is handled in compliance with the General Data Protection Regulation (GDPR).