How AI/ML adds value to the role of corporate treasury, Michael Kolman, CPO, ION Treasury

By Robin Amlot

Michael Kolman, Chief Product Officer at ION Treasury discusses how innovations in AI/ML are changing the corporate treasury function

For a number of years, there has been a desire that tasks within the treasury department add value to the organisation and are not just administrative tasks, such as manual reconciliation of cash transactions—someone spending time on that is not as valuable as potentially analysing data and driving insights. Thus, there has been a notable increase in investment in artificial intelligence and machine learning within corporate treasuries in the last 2-3 years.

Michael Kolman, Chief Product Officer at ION Treasury, said: “With the use of AI, now you can look at historical patterns, and AI can recognise those patterns for specific cashflow categories, such as accounts payable or capital expenditures. It can, without the need for input from around the organisation, generate a forecast based on that historical shape of the curve. That opens a world of opportunity, but it also opens up a world of questions.

“When you get submissions from somebody, you can go back to that person and ask them, why did the forecast change from the last forecast you made? It’s a little bit more difficult to ask the machine. So, a bit of this is cultural change and being able to adapt to a new way of working. There’s a transition period that happens to get more comfortable with AI.

“There are also other use cases. If you’re relying on AI to generate a cash forecast for you, and then you’re making an investment decision based on that information, if the forecast was wrong, and hence your investment decision was wrong, the costs of that decision can be much greater than, for example, another use case, which could simply be automating the reconciliation of two cash flows on a bank statement.

“Corporations need to get comfortable with their use of AI, how it can be leveraged in their organisation, how processes can change. Sometimes technology can front run readiness to adopt. In the case of AI, in many ways, I think that is one of the challenges that we’re facing right now.

“There is so much that can be done with using machine learning tools, using artificial intelligence. The corporate treasury is going to become an even more powerful part of the business because they will be able to make or offer advice on decisions that they would perhaps previously not have been involved in, in the business planning.

“If we think back, treasury was put on the map in organisations following the 2008 financial crisis when capital markets largely closed. The financial crisis hit and all of a sudden counterparty exposure is important, and liquidity management becomes much more important. The 13-week, 16-week forecast now sits at the heart of the treasury team. That makes treasury a significant part of the organisation where it’s gathering data from across the group, and then can measure and identify trends, which can lead to actions, making treasury not just a cost centre but with the opportunity to generate real value.”

Do corporate treasuries really need real-time data?

“Not necessarily all data needs to be or should be real-time. In accounting, it’s important that you have a closed period. So, when does the day close? If everything was real-time and constantly changing, closing the books becomes a bit challenging. So certain batch processes that run at a point in time, I think it’s important that they continue to run.

“However, when it comes to driving decision-making, especially around risk, it is important that you do have real-time visibility to the data. Changes in market pricing could influence the value of your hedge position, for example.

“The desirability of real-time data, or the answer to the desirability of real-time data, is it depends. It depends on what you’re looking at. It depends on what you need to use it for.”

Identifying trends or spotting patterns?

“Machine learning is pattern recognition. It looks at the pattern of actual cash flows over time. So, the machine needs data to be trained on, that data needs to be good data so that it can then be relied on to predict the go-forward forecast. Trends are derived from that pattern recognition. So, if you start to see that your cash balances over time continue to increase, well, great. What does that mean?

“Maybe you don’t need to keep as much cash on hand and can put that cash back toward repaying some debt. Or you might have a forecast to raise additional capital. Maybe it needs to be adjusted. Maybe you should be looking at longer-term investments. For a number of years, we’ve been operating in an interest rate environment close to zero. The opportunity cost of leaving cash in the bank account or investing it was pretty small. Well, now the opportunity is to optimise the level of liquidity in the organisation. Identifying those liquidity trends means you can start to make smarter decisions on how best to utilise cash.

What use cases for AI/ML has ION developed?

“Machine learning is a very good fit for cash forecasting. The most accurate forecasting method has actually been to forecast based on historical data. (You can always poke holes and say, what happens if the business changes, and history isn’t necessarily reflective of the future.)

“We have evolved our capabilities so that you can apply different forecasting techniques to different line items in your forecast. We’ve also added little things that may not be considered, such as holiday calendars, so that the machine does not forecast cash flows on holidays.

“ION provides the capability to leverage machine learning, but at the same time, in parallel, run your existing process so that you can then compare a forecast generated by machine learning to a forecast generated by the traditional method, and also compare forecasts to actual as well.

“A second use case of machine learning that we’ve developed is what I call rules-based automation. We’ve always had the ability to write a rule that for reconciliation says, if this criterion is met, then do this. That would then automate the matching process. I think companies who leverage our technology, basic rules-based automation will reconcile maybe 80% of their cash flows. ML can take that 80% and make it closer to 100%.

“Machine learning can identify things that rules can’t be written for. Say a bank reports a cash payment for $20,000 that was actually made up of six different payments. So, you’re trying to match up one item reported by the bank to a number of items expected to be met. Identifying that one-to-many match or many-to-one match, is something that rules will struggle with. However, ML is able to identify those matches automatically.

“ML can also be applied to the categorisation of cash flows. A number of times rules can’t identify what the right category is for a cash flow. However, with ML learning and new message formats that contain richer data, you now have the ability to leverage that data even further to automate more.

“On the payment side, it’s really important that companies feel comfortable that they have secure processes to ensure that payments are being made, cash flow is moving out of the company to the intended beneficiary of that payment.

“With machine learning, we can use what we call anomaly detection to identify something in the workflow or in the amount of a payment that looks peculiar, that doesn’t look like similar payments that have been made to that counterparty. By leveraging this, we can identify mistakes. But it could be used also to have early identification of potentially fraudulent activity. So, it is providing greater controls around the treasury environment and providing an added level of protection.”