How has AI helped you navigate cash forecasting challenges in the last couple of years?
With a forecast, you’re looking to validate two things: that your understanding of the world is correct, and that your expectations around timing are correct.
The thing I really focus on is the ability to run different models, and identify which model tracks closest to reality. While it doesn’t help us predict the future, what it can do is make it easier to hypothesise different iterations of the future, and then identify which one looks like it’s the most appropriate. This has practical uses in terms of working capital.
During the pandemic, for example, our test centres were closed. Our customers typically pay in advance for tests, and so people asked for their money back. These are high-stakes tests, such as accountancy exams, so we knew they still wanted to do the exams at some point.
We had various hypotheses about how that would play out – for example, people might rebook their tests straight away, or cancel and rebook at a later date. In the event, a lot of people asked for refunds straight away. From a treasury planning perspective, it was useful to be able to inform the business about what was going on. You would have to use a lot of resources to get the level of insights we’re able to get relatively easily.
Which other areas of cash management could benefit from AI?
AI is quite a broad term, and many companies use AI in cash management – even a zero-balance account (ZBA), at the simplest level, is basically AI.
What’s happening now is that AI is becoming democratised. The ZBA exists as a function because a bank can set up a process to target balance accounts and set up a structure to facilitate that. If you’ve got your legal and regulatory set up, then the mechanic of physically making the transfers can easily be automated.
ZBA tends to be a bit of a blunt instrument. It’s quite difficult to say your target balance is £0.5m for 20 days of the month, but on days when payroll and other expenses go out it needs to be a bit higher. As new technologies come through, it’s becoming easier to do that, and some treasury management systems (TMS) have functionalities that allow you to be quite nuanced.
Of course, the cash forecast is closely linked with your cash management activities. One of the things we’re working on is generating a rules engine that says, ‘If X or Y is happening, we recommend investing or transferring cash.’ You can then use tools like Python to generate the entries and import them into the TMS.
How could treasurers use technologies such as Microsoft Copilot?
Something like Microsoft Copilot will put ChatGPT in people’s hands. The way Copilot works is that if your company is on Microsoft Office, Copilot has access to all your emails, Word documents, PowerPoint documents, etc. So, if you ask it to create a summary of the treasury policies relating to deposits, for example, it should be able to go through your documents and files, find the relevant items, and send back to you a Word document that gives you that summary. So that’s an example of AI becoming much more accessible – you don’t need to be an expert in AI to be able to use that.
Similarly, GPT and Bard are pretty good at coding, although you need someone who knows what they’re doing to polish it. In my team I brought in someone who’s a data scientist and a full stack Python engineer. We’re seeing big benefits from this. Some of the things he’s able to generate are really helpful and allow us to use stronger and more powerful tools. That said, it’s quite a big investment if you’re just looking to make incremental changes.
