AI & Robotics in Corporate Treasury

This call was organised in the midst of media hype about AI (Artificial Intelligence), with several banks having banned the use of ChatGPT, and highly convincing fake videos being created and going viral. The first question was whether our members share the banks’ concern over the danger of tools such as ChatGPT running out of control and executing highly credible frauds. The answer was “No” – in the sense that it was not on anyone’s radar screen.

The second question was about the adoption of robots to automate treasury processes: these typically execute pre-defined processes, with no decision-making function. There are high levels of interest in this, and increasing levels of adoption.

AI is of interest for areas such as forecasting, where an increasing number of applications are available which look at historical data and use it to predict future patterns. However, only one participant is currently using it for cash forecasting. But:

• Banks are increasingly using AI to screen transactions, and highlight unusual ones. This is good – but one participant saw valid, but unusual, transactions fail to execute on time as a result.
• AI’s ability to handle large data lakes or oceans inside companies is a positive – but it is hard to be sure all the relevant data has been included.
• One participant is concerned that the questions being asked on ChatGPT could potentially be tracked, leading to confidentiality breaches.
• The same participant is also concerned that the inability to track and verify the source of data in this tool could lead to unintentional copyright infringements. Usage rules are being put in place to manage these risks.

On the other hand, most participants are actively using robots for various tasks:

• Allocation of cash receipts against outstanding invoices
• Recording bank transactions in the general ledger
• Sending out automated e-mails with management reports
• Managing approvals processes, including automatically cutting and pasting relevant e-mails
• Ensuring reports are correct – for example, that exchange rates have been updated
• Performing reconciliations automatically and immediately – this brings significant benefits as errors are detected and corrected earlier.
• Collecting and disseminating SOX information, and setting up a tracker
• Automated hedging and FX confirmations with banks (note: most portals, such as FXAll, have automated this process, which used to be a nightmare)
• Many bots start themselves, based on data triggers

Organisation considerations:

• A couple of participants’ companies have set up robotics centres, which create and manage bots on demand for the whole company, after submission of a business case and prioritisation.
• One participant has obtained programming skills and creates bots themselves
• One participant is only allowed to pursue new hires after having demonstrated that the task cannot be automated.
• Staff generally appreciate being relieved of having to perform boring and repetitive tasks
• In turn, this frees up staff for more value-added activities – and helps the company to grow without adding headcount

Robotics apps used by participants:

• UiPath
• Alteryx
• Tableau
• Robot Process Automation (RPA) within SAP
• Anaplan (machine learning)
• Microsoft Power Apps, (freeware within Microsoft Office Suite)

Management discussions:

• Robotics can be an effective short term solution. But, often, there is a real need to update the underlying system. While this danger is recognised, a short term solution is viewed as being better than none.
• There are challenges with change control, documentation and testing before bots go live. There was a feeling – though no hard data – that this may not always be addressed properly (note: this is a major failing of nearly all Excel uses).
• At the same time, creating the bot often requires documenting processes which have been executed without being documented previously
• Many bots fail when there is a change in the underlying data – for example, when a customer changes the number of zeroes in a field. In many cases, a person needs to check and regularly update the bot.
• Of course, increasing the ability to process data only helps if the data itself is accurate – the need to avoid GIGO (garbage in, garbage out) remains a key management responsibility. It can be necessary to clean historical data.

Bottom line: robotics are not without their drawbacks, but they are an attractive and effective way of increasing productivity without having to go through a major IT development. There was an impressive amount of uptake and enthusiasm amongst participants for this, and some good tools out there, which can even be accessed by end users. AI is drawing interest – but our participants were more at the looking and asking stage, than in active implementation.

Watch this space!