How does treasury realize the benefit of AI? Ali Curi: Treasury ConversatION is an ION podcast where we discuss topics of importance with CFOs, group treasurers, and treasurers. Join us as we explore critical topics with industry leaders, product owners, and subject matter experts, providing insights and strategies tailored to the dynamic world of treasury management. Hi everyone, and welcome to Treasury ConversatION. I'm Ali Curi. On today's episode, Michael Kolman from ION Treasury will discuss how corporate treasuries can harness the power of AI to drive efficiency, improve decision making and create strategic value for their organizations. In the fast paced world of corporate finance, artificial intelligence is emerging as a game changer for treasury operations. From revolutionizing cash flow forecasting, to enhancing risk management and automating routine tasks, AI is transforming how treasuries operate. We'll dive into key applications, discuss potential challenges, and offer insights on how companies can successfully integrate AI into your treasury operations. It's going to be an exciting conversation. So, let's get started. Michael Kolman, welcome to the podcast. Michael Kolman: Thanks, Ali, for having me. Ali Curi: Michael, before we get to our conversation, let's learn a little bit more about you. Tell us about your background and what is your current role and responsibilities at ION. Michael Kolman: At ION, I am currently the Chief Product Officer for ION Treasury. So my responsibility is really to ensure that we have a clear strategy for the business, we are understanding customer needs and we are exploring new technologies such as AI and incorporating that into what we do. Prior to joining ION, I spent the majority of my career in corporate finance roles. Most recently before joining ION, was leading a treasury transformation for a large multinational organization, which I think has provided me a lot of perspective that I carry with me and think about within my current role. I think that's been extremely valuable and the learning that I got from that experience actually is what drove me to ION in the first place. Because realizing the value of technology and what technology can enable in terms of executing more efficiently just creates a universe of possibilities that gets me pretty excited. Ali Curi: Great, Michael, thank you. Let's jump right in. Your background and expertise are perfect for our topic today, especially at the intersection of finance and technology. So to start things off, can you explain how technology needs and corporate treasuries differ from other areas of a business? Michael Kolman: That's a good question. A good way to sort of frame some of these topics up. If we just think about why is treasury in place in the first place. Treasury is really there as a risk management function. I think the approach would be, "Let's try to avoid risk wherever possible, right?" Treasurers we're typically quite risk averse people, and then where we can't entirely avoid risk, ensure that we have proper mitigations in place. And so when it comes to technology and treasury, you can kind of cluster into three different areas. So one is controls. Do I have control over my operations? I'm managing cash. I want to make sure that cash is being moved from one place to a place where I intend it to be moved. I want to make sure that I have the controls that maintain the assets that I have in my organization. The second area is visibility. So do I have visibility to all of my exposures? So all of my positions, whether that be my cash position or the position I have in any specific currency or my exposure to interest rates. And I want to make sure that I am mitigating those risks, but also at the same time, the way to mitigate those risks is to ensure that I have real time visibility as markets change, as money moves. And then the third piece of technology is really around decisions. If I have control over my operations, if I then have visibility, now I have the ability to use data to make decisions, use the technology to be able to visualize, am I hedged appropriately according to my policy? Do I have sufficient cash to be able to meet my ongoing needs? And if any of those answers require action, now I'm actually using technology to make those decisions. So I think we think about this is how is that different from other parts of the business, which I think is another part of the question. I think it's also, you know, we think about treasury, maybe even the finance department more broadly as a cost center. And so for cost centers, this is a bit of a difference, right? If you are a profit center, your focus is on driving profits. If you're a cost center, you're looking more at cost containment. And so that leads to needs to drive greater automation, obviously lower costs and increase efficiency. The other unique thing I think about treasury and also more broadly, finance department, is what makes it unique is that we are a consolidator of data from so many different departments across an organization. One business unit would typically focus on that business unit's operations and execution and strategy, but in the finance department, we're really consolidating the results from a number of different areas. And in treasury specifically, consolidating cash, consolidating risk into one central area. We sit on a lot of data in treasury, and in the broader finance department, which really makes it sort of ripe to benefit from AI. Ali Curi: Well, I think that's the perfect setup because that background you shared with us, we're going to use that to move forward and talk about the big buzzword AI. So everybody's talking about AI these days, everybody wants it. But when it comes to actually implementing AI in corporate treasuries, where do organizations begin? And before we get that, share with us just a little bit of background on AI. And then let's talk about what are the first steps that an organization would want to take to implement AI to assist, especially treasuries, in their operations. Michael Kolman: With AI, it's really quite remarkable. I mean, we hear about artificial intelligence almost every day, it feels like these days. Whether not just in thinking about treasury, but even outside of treasury, a number of us are now avid users of ChatGPT, right? So generative AI is really becoming part of our own personal lives. And it really dates back to the 1950s where I think Alan Turing's really the one who's credited with conceptualizing AI, not necessarily building the technology, or he didn't have the technology to build, but he came up with the concept that the machines would have the capability to do more than just what it was programmed to do. And I think that concept is really taking shape and materializing now. And we're seeing quite a massive evolution in terms of automation. AI is such a broad term and you can apply AI in a number of different ways. And I think that when we think about machine driven automation, historically, we've always thought about it in sort of a rules-based way. So this would mean you tell the machine, if a certain set of criteria exists, then take this action, right? So you essentially created a rule, you've programmed the machine to execute on that rule, and it will always execute as you tell it to. Now we're moving into a world where machine learning is actually, is becoming quite prevalent and adding quite a unique value proposition and what machine learning is doing is, which and just maybe just definition wise, just take a step back because I think we tend, and I'm probably gonna do the same thing in this podcast, to use AI and machine learning somewhat interchangeably. So AI, just taking a step back is probably a bit of a broader term for how machines can mimic cognitive functions. Now, machine learning, I would say is a subset of AI. Where in machine learning, the machine has been created to automatically learn. And we have sophisticated models like neural networks that are essentially algorithms that are trained to start to think, recognize patterns and think like the human brain does. So I think that gives us probably a perspective of the history of AI. Really our focus to date has really been around machine learning at ION, and we've looked at leveraging the machine learning capability to be able to introduce the concepts that we have in place. Ali Curi: So thank you for sharing that background. So now once AI is in place. This is where the transformation can happen, right? Can you walk us through some examples of the benefits that corporate treasuries experience once they get the implementation right? What kind of impact can treasuries expect with implementing AI? Michael Kolman: The first expectation is really around what is the quality of your data? First of all, do you have data? And then the second is, what is the quality of that data? Because if you want to set the expectation, if you don't have data or the quality of your data is pretty poor, then the expectation should be set that the value you get from AI right off the bat will be pretty insignificant. And so I think that data is the primary building block to being able to realize benefits from machine learning. You need that data to be able to train the machine to generate the output that will be beneficial to the organization. Does that make sense? Ali Curi: Yes, Michael, I see what you're saying. Michael Kolman: Yeah, so if we then try and apply the aspects of machine learning to how corporate treasuries can benefit, there are certainly some clear use cases where the technology can be applied. At ION the first place that we started was really around cash forecasting. And I say that that's the most obvious because you think about machine learning and you think about the ability for the machine to make predictions. The thing we've been trying to predict in corporate treasury with quite a bit of struggle and investment of a lot of people's time is, "Will I have sufficient cash to be able to meet the needs of the business going forward?" In any organization, there is a consolidation or an aggregation of data from a number of different people. So you can imagine each person is spending time to create a forecast based on their role and responsibility, that's then submitted to treasury. Treasury then consolidates that into a global forecast so we can see all the cash movements that are predicted over time. When we think about the benefit of AI had on cash forecasting, we could see it in really two areas. So one is we wouldn't necessarily have to rely on the input from individuals as frequently as we might have to today. So there is an enormous time savings. In our initial use case we saw the machine really saves 3000 times faster or can execute 3000 times faster than the individual submissions. And then if you take that time from the large number of people that are submitting forecasts to treasury, well then we just are, we're saving an enormous amount of time. So I think that's one area, is time savings. The other savings is around accuracy. And so with the treasury system, having visibility to all of the actual data, it has the capability to bring in all of the bank statements, be able to tag and you know, categorize each of those cash transactions. So we have actual data within the treasury systems that positions as well, apply machine learning, train the machine, and actually in some cases be even more accurate than what the submissions are saying. And you can apply the machine learning models very specifically. So perhaps in some business units that are quite steady, the machine will actually be a great predictor and we can save people time and maybe even improve accuracy. But then also where specific line items might be a bit more predictable, the machine can also do a very good job there. And then we can actually focus the human on areas where the machines forecast is perhaps not as accurate because there's a bit more volatility, right? Capital expenditures might change. There might be some M&A activity that is expected and the machine's not going to pick up those one time events. So cash forecasting was sort of the place we started at ION, but we have a number of other areas, so anomaly detection. So this would be detecting fraud, automating reconciliation are just two other areas where I think there can be some significant benefit that treasury teams can realize. ION Ad: This episode is brought to you by ION. Unlock the potential of your data through machine learning by ION Treasury. ION Treasury Solutions, powered by artificial intelligence and machine learning, unlock the potential of your data with our fast, accurate algorithms, and deliver new insights to help you better plan, prepare, and drive your business strategy. To learn more, visit us at iongroup.com/treasury or email us at treasury@iongroup.com. Ali Curi: So you mentioned a couple of things that I want to circle back to. So this is a powerful tool. And what are some challenges that treasuries might encounter when leaning on AI versus relying on the human insights? Because you talked about the human factor a little bit. Can you tell us more about that? Michael Kolman: This is such an important question, because I think in a number of cases throughout history that sometimes technology can actually front run adoption. And so here is one of those perfect examples where in some cases, we have to ask ourselves, "Are we actually ready to adopt AI in our current workflows?" And I think that some people will get scared, of AI, that AI will replace the humans. I honestly do not believe anytime in the near future that that will happen. I think that the machine learning tools that we have and we're applying are really just a compliment. So if you go back to the cash forecasting example, I had highlighted there might be some specific areas of volatility that the machine just can't predict and being close to the business, the human can. And so I think that just really goes hand-in-hand of figuring out how do we work together between us as individuals in our roles, working and using the tools that we're given to compliment what we're doing and to help us, not necessarily replace us. We have to think about it like that. And from that, then we can start to really embrace the benefit of AI. So then when it comes to truly adoption, I think one of the challenges there is when you have automation, that's rules based, you know exactly when the rule is kicking in and what action it's taking. So now I have, let's say, let's go back to the cash forecasting example. I have a forecast and if I am generating the forecast and I'm responsible for that forecast and I send it off to the CFO of the company and the CFO looks at it, understands, great. And then a week or a month later, send another forecast for a similar time period, and the forecast has changed, the immediate question is going to be, "Why? Why did the forecast change so much from the one you submitted two weeks ago?" And I think that we have to be in a position to be able to provide an answer because just simply saying, "Oh, well, that's what the machine said," is not going to be a sufficient answer for a CFO, for a treasurer, for anybody who's ultimately responsible for the decisions that are being made as a result of that forecast. So we have to really think about, "Well, how do I get to a point where I can actually understand my forecast that the machine is giving to me?" I can get confident in it, and I think that there's a bit of a ramp up period for sure. And there's also some things that perhaps technology companies like ION can also do, which is providing visibility. You know, how confident is the machine learning model in the forecast it has provided? Maybe there's some visibility that can be provided in terms of providing some sort of explanation as to what has changed because we're always comparing actuals to forecast or one forecast version to another forecast version. So if we can provide some insights into why that has changed, just like we would try to explain in the traditional forecasting method, I think those are things that will help actually ease adoption and build that trust. Ali Curi: Well you made a very good point about AI and human decision making complementing each other instead of AI replacing human decision altogether. What is an example of this decision making complementing each other in a corporate treasury operation scenario? Michael Kolman: That's a great question. And this is hopefully an example to demonstrate that. So I had mentioned that we are pursuing a number of different use cases of machine learning. And one of those is to automate bank reconciliation. So this is the process, it's quite tedious for those that don't have automation tools. Basically, you're looking at the bank statement and the transactions on the bank statement, and you're trying to compare that or match that to what you expected to happen. And so when you do that, most of the treasury tools, our ION Treasury Management Systems have the ability to automatically match the two transactions. One, the expected versus the reported. And so with that matching today is done by defining a certain set of rules and a certain set of criteria that allows us to, to match like terms. So if there's a match, it looks at the rules and it makes that match. But the challenge with the rules is that the rules have to constantly be updated. They have to be maintained. And that requires some additional time. If we try to apply machine learning to reconciliation, we now actually have a tool that actually requires less maintenance because rather than, remember we go back to this different way of thinking, right, the machine isn't being given some rules, it's given some data. And when you provide that with historical reconciliation data, it might be able to look at today's transactions that are coming in and say, I've actually seen this transaction before, and I am 99% confident it is a transaction related to a certain reason. That reason could be, it is a capital expenditure or it could be a payroll payment. The machine actually knows because of its history, that it's 99% confident in this example that it is a match. If the machine is 99% confident, then the process can be automated. There's no rules involved. Nothing has to be maintained. The transaction can be accounted for properly. If we say though, that the reality is that doesn't happen a 100% of the time. And so the machine then will get other transactions where maybe it sees a match, but it's not entirely sure. And its confidence level is actually closer to 50%. So now if the confidence level is 50% now actually is where the human intervention comes in. A person who is responsible for this task would look at that one where the machine says it's 50% confident and say, "You know what, I know what that transaction is. I can identify it, I can tag it to the right category, I can match it to the expected transaction and I can complete my reconciliation process." So there's some manual intervention and where the human wants to step in and that makes sense, but imagine if you are sitting on your desk and every single day, you have tens of thousands of transactions that have to be matched. And if the machine can match those transactions, largely 95% or more for you, that just takes away that time that you're actually spending and allows you to focus that time and energy, [on] things that you would rather be working on than just simply matching transactions together. Ali Curi: Michael, we've talked about the benefits of AI, some of the benefits, some of the challenges. And let's step back and look at a bigger picture. How would corporate treasuries know when they're ready to embrace AI? Like, what should they look for to know that it's the right time to invest in these kinds of technologies? Michael Kolman: There are a number of factors at play. So first there's probably a greater organizational wide initiative to adopt AI in the organization. And there you're then starting to see support from the rest of the company to start to embrace AI and seek out options where AI can provide some value for your team. So there you have the company support to do that and to move forward. And I think that creates an environment that says, "Yes, I wanna do that. I have the support to do that." And in many ways, for treasury, you guys have the great opportunity to really be a leader in your organization and a champion for AI, because you are a consolidator of data and likely you have the good data. That's, I think what, you know, "Are you ready?" Well, one is if your company overall is saying, "Yes, we want to adopt this." Then I think you're ready. You can't get to step one before you can get past step zero and step zero is data. So it really all comes back to data. Do you have good data? Do you have access to that data? Do you have data governance in place to be able to sustain the quality of that data? Is that data secure? All of those questions need to be answered before you can even get to step one. So if your organization is saying, "Yes, we want to embrace this, we want to adopt it." Well, that's great. Do you have the fundamental building blocks in order to do that is really question number one. I think in your organizations, you will know whether you have that support or not. Now, if you feel like you don't have that support, my advice would be go on the offensive, right? Say there's a great opportunity for us to be able to embrace AI in our treasury team. Because the reality is that if you don't actually get started now, everybody else who is starting now or has already started, they're going to be years ahead of you. And every single person who has started to experiment with AI will tell you, as you start to use it, you start to realize that there are some nuances that need to be tweaked. There are some things that have to be adjusted. There's certain issues with the data that have to be cleaned up. So if you don't start to recognize those challenges that are ahead of you by just getting started, then you're going to be left behind. Ali Curi: Michael, you talked about the importance of data and how you shared some points on how treasuries can prepare their data infrastructure. Is there anything else that you can share on how they should prepare their data infrastructure to ensure that their setup is successful when they're implementing AI solutions? Michael Kolman: I think that for the data to be successful, you have to really have an enterprise data strategy. There are a number of different applications that use the same type of data. So is your data sort of centralized in a way and consistent so that it's being used with consistent naming across systems, a good example of that, that we see in treasury all the time is with banks, right? So if a company is banking with let's call it, "Bank ABC," and they manage accounts and get statements for Bank ABC and make payments for Bank ABC out of their treasury system, that's one area. Perhaps they trade with Bank ABC also out of the treasury system. In the ERP system perhaps they also manage bank accounts with Bank ABC. But in each of those two systems, Bank ABC might be called "Bank ABC USA" in one system and just "Bank ABC" in another, or maybe there's a "Bank ABC Holding." And all of these things are exactly the same bank, the same entity, and there's just not a common naming convention. And so we think about centralizing counterparties in a certain place and relying on that central data repository to be able to then feed other downstream systems. Well, now all of a sudden our data has become a lot better. It's become a lot more consistent and we're able to really benefit from a lot of the learnings that can happen. I think when we talk about data quality, a lot of that has to do with really that governance, right? Where is the data coming from? How is it maintained? How is it cared for? How is it set up? And the great thing is, is that once that's centralized and you have control over that, there are a number of benefits that will come. Ali Curi: Perfect. Now, once this is all set up and they've gone through the process of making sure that their data is qualified, how can corporate treasuries measure the success of the AI upgrade? Because often there's a gap between implementing technology and realizing its full potential, right? So you want to be able to measure that success. How can they do that? Michael Kolman: Right. So I think that a couple of the things we've said kind of, let's bring those together. So what's the benefit they realized? Well, it goes back to, how well did they adopt it? And I would imagine that at the beginning, adoption is somewhat experimentation. You're experimenting. Does it work? Has it actually replaced anything that you're doing today? I think it takes time to actually get to that point. You have to go through this experimentation phase. Once you get through that experimentation phase and you start to really incorporate it into your workflow, that it is part of your process, the document that is part of your procedure. And this is the machine learning capability, whatever process you're applying it to, it really becomes part of that process. Now you're going to start to see more adoption. And so in the case of forecasts, you can measure that success in terms of, "How much more accurate is my forecast? How many hours less is my team spending on generating that forecast?" With forecasting, there are so many possibilities, but also there's also some larger consequences which makes AI harder to adopt for something like forecasting. If we think about there's improved accuracy, there is time spent going down. Well, this is all great. And we want to work in compliment with traditional processes. So perhaps if you are updating your forecast every two weeks today, perhaps you can incorporate machine learning and then run a new forecasting run should you see any significant change from the machine learning generated forecast? So the two are working in compliment with each other. But when you measure the success, you're saving a bunch of time. Same thing with reconciliation. You might be automating a lot of your reconciliation today and that matching, but perhaps what you're seeing, and we're starting to see this from a number of other customers that are experimenting is that the value is actually in categorizing those cash transactions so that you can clearly identify what the transaction is for and you can account for that properly. So today it might find a match, but it might categorize it into some sort of miscellaneous category. So even though the matching is automatic, you still have to then go through and categorize that transaction manually. With machine learning, you now have the capability to save even that time. So I think a lot of this is really about there is some significant time savings and that time savings doesn't mean that jobs are replaced. I know I said it before, but I'll say it again. It doesn't mean that jobs are replaced. It means that it's freeing up time for you to actually spend working on something else. A perfect example is cash forecasting again. We have the capability to compare forecast to actuals, forecast to forecast, and analyze how good we are at forecasting, but nobody has the time to do it today. Because they're spending all the time just generating the forecast. So if that time is saved and given back to people, now we can actually spend the time to analyze the forecast and think about how can we get better. And that's where treasury can start to really add and be a true contributor of even greater value to the organization than just an operational contributor. Ali Curi: Well, I really appreciate you sharing your insight with our listeners and reiterating that the machines are not going to be taking over, at least not anytime soon. Michael, what is the one big thing you hope listeners take away from this episode? Michael Kolman: I hope that we have evolved our thinking. That AI is not just simply a buzzword, but it's actually something that can be applied today and it can be real and it can provide in some cases, immediate benefit. I think the second thing is also to just make sure that you are working together with your technology partners, with your internal stakeholders to really drive that adoption and that learning. Get started now. That's the other thing I hope people take away from this. Get started now, because if you don't, you'll get left behind, and you'll constantly be trying to catch up. If you get started now, then you'll go through that learning process. You will focus on the data. That's another piece. Focus on the data, and your data will get better the more and more you use it. Ali Curi: Well, I think that is some great advice. And speaking of great advice, let's have a quick sidebar and talk about career advice. What is some advice you wish you had heard earlier in your career? Michael Kolman: Certainly my career going from corporate finance to ION, we're seeing that intersection of finance and technology. And realizing the potential that technology can provide to the finance department is just so exciting. If you appreciate and adopt the capability, there's so much more that we can do. And I think that hopefully will resonate with those in the finance department, because I think you're constantly being asked to do more, but you're constantly being asked to do more with less. So how do you accomplish that? And I think technology has to be the enabler for that. And for me, that's what led me to the role that I'm in today. And that's what drives me forward. So hopefully that's some, some career advice thinking about ultimately what is your passion? For me, it was to make things better. I could see that being enabled through technology. Ali Curi: Well, that sounds like great advice. Michael Kolman, thank you for joining us today. I hope you visit us again. Michael Kolman: Thanks Ali for having me. Ali Curi: And that's our episode for today. You can follow ION Treasury on X and on LinkedIn. Thank you for joining us.