How Agentic AI is Reshaping Modern Treasury

The Payments Podcast by Bottomline

Episode Transcript

Owen McDonald (host): Welcome to The Payments Podcast. I'm your host, Bottomline Managing Editor, Owen McDonald. Of all the business functions now benefiting from AI, treasury is making above average strides. With the rapid rise of agentic AI, we're witnessing a step change in B2B treasury marked by more visibility into cash positions, better liquidity forecasting, superior anomaly detection, the list goes on. And since AI works best with humans at the controls, we're fortunate to be joined by Bottomline's Leo Gil, our go to guy for treasury tech and related trends.

Leo Gil, welcome back to The Payments Podcast.

Leo Gil: Hey, Owen. Thanks for having me.

Owen McDonald: To start right off, Leo, let me ask you this. In practical terms, how does AI free up treasurers to be more strategic, moving them away from back-office chores and toward higher value decision making? That's the dream, but how does it work in real life?

Leo Gil: That's a great question. I think the first step is to think about AI and use AI as another member of the treasury team.

Instead of thinking about or leveraging AI as a replacement for what day-to-day responsibilities of the treasurer are. By leveraging as another member of the team, or multiple people in the team, then, of course, we'll always help free up the treasurer to lift their heads up and be more strategic, help their company solve real problems they have with the changes in market conditions and all of the other dynamics they face today.

Owen McDonald: Okay. Fair enough. And it does seem to be doing that in the early going. That's what we're seeing. 

We hear a lot about the superior efficiency of AI. Leo, what kinds of efficiency gains can treasury teams realistically expect to achieve in 2026 using AI's ability to parse large datasets and surface patterns?

Leo Gil: It's always a journey as companies start their journey into AI and leveraging AI capabilities. The first step is always getting data quickly. Think about how many companies have to navigate multiple systems or even if they do have centralized systems like a treasury management system, they still have to click around the user interface to get at information.

So, by leveraging AI, large language models, conversational, user interfaces, then it's easier to get at the information you need. If I would like to find out, do we have any accounts that are overdrawn? Instead of clicking through menus, running reports, filtering, and getting that data, I can simply just ask, and type into a conversational interface and ask that question, which will return the result much more quickly. So that, I would say, is the first step, realistically, that companies can gain, but, of course, as they continue their journey there's a lot of other factors that can help them, things like being able to provide them insights and actionable insights about the data and about what's happening to their business.

An area that we have been working a lot on is forecasting. So being able to leverage AI machine learning to build better prediction models, to give more insights to treasurers so they can maybe see things that they were not thinking about. That is going back to that additional member of the team is able to have AI look through your data and provide you insights that maybe it's something you weren't really thinking about. So, surprising and delighting users with information is where AI can definitely help.

Owen McDonald: That makes a lot of sense, and it kind of dovetails with this next question. You've told me in conversation that, quote, data is challenge number one in AI usage. So, what does having the right data with proper controls and labeling look like in a mature treasury organization that's adopting AI?

Leo Gil: That's a great question. Data is paramount. Without data, you can't really build any models, or really leverage, AI in a lot of ways.

So, as you know, lots of companies have challenges because they have fragmented systems. They may have multiple ERPs. They may have treasury management systems and accounting systems and so on. And it's really difficult. And, of course, lots of spreadsheets that we can't forget about.

So, it becomes quite challenging when the data is in separate systems. Data is labeled differently across different systems. In one system you may call something an account number. In another system that account number may mean something else or it becomes challenging to normalize all of that data together.

That's why one of the areas that we focus on as step number one is building an integrated model where all of the data from all of these disparate systems come to one place. From that point on, we can then leverage AI machine learning. 

But you're right. That's a big challenge to a lot of companies, and being able to do it properly, it has shown to be a challenge, and that's an area that we work really closely with our clients. Day one is to help them understand their data, normalize the data, and integrate their data into platforms so all of that work is done on day one.

Owen McDonald: That explains very, I think, succinctly why partners are so important in this transition. Wouldn't you say, Leo? Payment service providers like Bottomline and others, I mean, it's important to have people who understand this and can help you adopt it.

Leo Gil: Yeah. Exactly.

Owen McDonald: I want to move on to fraud. The anti-fraud uses of this technology are fascinating. Explain, Leo, how AI agents made for payment anomaly detection are reducing errors and helping uncover potential insider fraud, especially in treasury?

Leo Gil: Fraud is really one of the first, if not the first, use case for AI machinery was fraud prevention.

Owen McDonald: Right.

Leo Gil: Large volumes of data, large volumes of payments and transactions, very challenging for anyone to really feel sure that the payments that are being made are correct. And as you said, they're not fraudulent. A big area in a lot of companies, a big focus, is, of course, on fraud prevention, internal fraud, or even someone internally being lured by a fraudster outside of the organization. But errors are also really important and really costly. Making errors with payments and having to resubmit the payments. The bank fees aren't reversible in a lot of ways, so companies can spend quite a lot of money, with errors and, of course, fraud and AI machine learning is a perfect example for it. 

One practical use case is anomaly detection. For example, we have the ability to build models based on the how payments are made in an organization and how payments come into an organization. So, we can see beneficiaries, we can see amounts, we can see things like days of the month, which payments are made. And then, as you build those models, then you can alert and build an anomaly detection engine that helps the company identify potential errors or fraud.

Knowing that, typically, you never make a payment of a particular account in a particular currency, on a particular day of the month, it may be an error. Maybe someone is making a payment by mistake from account that they shouldn't be making, or it could be a fraudulent event. So, building those models and they are very specific to each company.

Each company is going to have their own patterns. Each company is going to have their own processes, and procedures depending on their customers and suppliers. So that's where we see AI as a really strong capability because for a human to be able to identify all of those potential issues, it's impossible. Right? The machine learning models and AI can help a lot in that area for sure.

Owen McDonald: Yeah. I hadn't planned on asking you this question, but, Bottomline has its own agentic AI known as Bea specializing in treasury, and it can help with many of these things. Isn't that right?

Leo Gil: That's correct. We've just launched Bea, our AI agent, which works both across cash management, liquidity, and payments. So, the agent covers a lot of these topics we're covering from cash forecasting to conversational user interfaces to payment anomaly detection.

All of those are tasks that our AI agent can perform today. And, of course, as we work closely with clients, we build more and more capabilities. We have a long road map for things we are going to be releasing like automatically generating reports through our conversational user interface and so on. We see real usage of clients leveraging these capabilities, asking for it.

Owen McDonald: Last question. Leo, I'd like to know how AI is changing core liquidity management tasks today. Things like cash forecasting, predicting payment behavior, even deciding when to offer early payment discounts. What are examples of AI taking on those types of tasks?

Leo Gil: You covered some of them already, like the ability to predict liquidity. Companies would have different patterns. Seasonality is a big factor for many companies. So, leveraging all of that data and help them predict where their cash will be then opens up all sorts of interesting use cases.

Knowing that couple months down the road, the company will have a liquidity shortfall, then what are the actions that they could potentially make? Should they borrow from a loan at the particular rate that they have, and what's the impact of that? So, doing scenario planning and what is the analysis on those cases. Or, as you said, could we maybe look at a population of customers we have and offer them an early pay discount? 

It's knowing which customers and when do they pay as part of the process, and as part of forecasting, also to be able to predict when customers pay you. It's quite interesting when you talk to treasurers and you ask how they forecast inflows into the organization. A lot of them will say they're looking at invoice due dates. And we all know customers rarely pay, or not many customers pay, in the due date of the invoice. There's always a delay. 

So, being able to understand when customers actually pay and forecast based on that can be a big, big factor on change in liquidity, especially if you're talking about large invoices. It could be a difference between borrowing from a loan at some interest rate, versus offering a discount for a large invoice that will cover liquidity shortfall that you know that a customer will be able to do, can be really game changing for a lot of organizations.

Owen McDonald: And now we know. The coming months will see treasury going in new directions powered by AI and directed by human experts. From this, many expect a wave of optimization, transparency, and new dimensions being added to what treasury can do. Stay tuned on that front. 

Our thanks again to Bottomline's Leo Gil for his sharp insights. To our audience, the smartest people in B2B payments, thanks for listening. Hit subscribe. Catch us again on your favorite podcast platforms, including Apple, Spotify, Blubrry, iHeartRadio, and YouTube. Bye for now.

Hit subscribe. Catch us again on your favorite podcast platforms, including Apple, Spotify, iHeartRadio, and YouTube. Bye for now.

The Payments Podcast, from Bottomline.