Articles
December 11, 2025
Listening to Treasury Teams Hones Kyriba’s Vision for Agentic AI
# AI
# TMS
Input from corporate treasury co-innovators on use cases and improvements plays a big role in shaping Kyriba’s strategy for agentic AI.
Navigating the road ahead in the brave new world of artificial intelligence is not only a source of stress for corporate treasury teams tasked by senior leaders to drive transformation, achieve efficiency gains and improve decision-making by finding use cases for machine learning, generative AI and AI agents. The pressure is also intense for cutting-edge solution providers committed to remaining relevant, raising their games and enabling clients to meet this digital moment as corporate finance functions grow laser-focused on leveraging advanced automation and AI.
Among the vendors leading the AI charge is Kyriba, whose liquidity performance platform is used by thousands of customers—including many NeuGroup members—for its treasury management system (TMS), risk management, payments and working capital solutions. Today, the company is fast progressing in integrating AI functionality within its ecosystem with the help of customers taking part in its Co-Innovation Lab.
Treasury team members from Koch and six other companies participated in the first phase of co-innovation leading up to KyribaLive in May, when the vendor announced the prototype of its agentic AI solution, TAI, which consists of an embedded large language model (LLM) and agentic AI models. The co-innovation group now stands at 11 corporates; its second phase ended with the October commercial launch of TAI. The current phase is focused on the 2026 TAI road map prototype.
“They are a part of the innovation,” said Dory Malouf, Senior Director of Kyriba’s Global Value Engineering team. “They have a seat at the table, sharing what they need, what they don’t need; what will work, what will not work; what is priority one versus priority two versus priority three. I don’t think this would have been as successful if we didn’t do it that way. We don’t want to be dictating their needs. We want them to tell us their needs so we can create a solution that truly works for them.”
Dory Malouf, Senior Director of Global Value Engineering, Kyriba
The trust factor. One of the most important needs of finance teams is gaining trust that allowing AI tools to access company data is safe, the answers provided by LLMs and agents are accurate, and that AI is actually adding value. What Kyriba dubs an “AI trust gap” informs the names of TAI and the complete “Trusted AI” portfolio—and reflects the hesitation about AI voiced by treasury practitioners. Many are highly motivated to make use of AI’s analytic and reasoning powers—if they can ensure it doesn’t bring added risk.
Listening closely to customer concerns about the technology has given Kyriba’s team a deeper, nuanced understanding of the complicated dynamic surrounding AI for treasurers. Tom Callway, Kyriba’s VP of Product Marketing, put it this way: “There’s a sort of tension between treasurers who are super excited about the potential of an AI tool but also fairly worried that this is some of the most important proprietary information they have in their business. And do they really want an AI tool looking at all their data?”
Tom Callway, VP of Product Marketing, Kyriba
Kyriba has addressed that question in part by embedding AI functionality within the confines of its platform—distinguishing the company from competitors, Mr. Callway said. “We didn’t do what everybody else has done. Most vendors have just signed an agreement with a third-party, hosted LLM provider like OpenAI and will farm out customer data to that LLM in order to do processing. That was never acceptable to us and was never going to be acceptable to our customers.”
Instead, he said, Kyriba “has taken the models themselves—whether it’s Llama, whether it’s Claude or others—and brought them into the platform so the customer data doesn’t leave their environment. So they essentially have their own LLM that will provide all of the advice that they need without their data going outside the platform. So that’s the basis for the ‘trusted’ moniker. It’s not just marketing.”
Kyriba recognizes, however, that not all corporate treasury teams will proceed at the same pace in adopting all that advanced AI has to offer. Some will go through a lengthy internal process answering questions and requesting approvals about data governance and other privacy issues. “Many of our customers have an IT policy that requires them to ensure that only their data trains the AI and the AI outputs are restricted to only their treasury team,” said Bob Stark, Global Head of Market Strategy at Kyriba. “And we ensure that those questions, which should be rightfully asked, are easily answered.”
But “some organizations are not at the stage where they’re going to use AI within sensitive areas such as treasury,” Mr. Stark explained at a recent virtual session with NeuGroup members. “While that may be their reality in 2025, the 2026 or 2027 answer is probably different, making it all the more important to collaborate internally with IT and information security teams to understand their AI boundaries. In many cases, treasury teams are surprised that they are permitted to use AI.”
Bob Stark, Global Head of Market Strategy, Kyriba
Priorities and co-innovators. Listening to clients at all stages of the AI journey is the governing principle behind the Co-Innovation Lab. It gives treasury teams a major role in Kyriba’s decisions about what functionality and use cases to prioritize and what changes, improvements and additional features are necessary after clients try out the tools. In the first phase, the co-innovators also provided “a reality check,” Mr. Callway said, essentially saying, “’It’s really cool you’re bringing the LLM inside of the application. That’s exactly what we want. But we want to focus on the biggest return on investment. If we’re going to invest in this, what’s it going to do for me?’”
Mr. Callway noted the significant contributions during phase one of a senior treasury leader and NeuGroup member at a tech company also investing heavily in AI. “She’s provided an enormous amount of hugely useful feedback and suggestions in terms of helping us prioritize the use cases that we’ve tackled first,” he said. “The work that we’ve been doing together has been so productive that the concept of co-innovation is something we want to do more of because it’s been so influential in the way that we’ve developed TAI.”
Mr. Stark said that during one of the co-innovation sessions, “We asked them ‘what would you like to tell your treasury system to do for you?’ And they came back with their favorite scenarios of what they would like to just type in or voice command to the TMS.” Examples included how to prepare for a meeting with your banks, creation of new bank accounts based on an updated bank service agreement, and optimizing currency hedging processes. In addition, “The level of detail that you expect from AI is also very important. We found that different AI models offered greater reasoning abilities, which translated to more complex queries,’” he said.
“Obviously ‘run my cash position, do my forecast’ would be a task most treasury teams would like to automate, but there’s a balance of how much detail you give it,” he added. “If you save three minutes by automating a process, is that sufficient productivity, or would you rather focus on more time-consuming tasks? Where does that balance exist? We take that feedback into account as we design AI agents. Is there some value to this in terms of a cost structure? So it’s a bit of show-and-tell process followed by confirmation of the impact to them. Does this help you? Would you like it? Would you use it? Would you pay for it?”
Progress report. Mr. Stark described how TAI has progressed. “The first use of AI is often very simple queries within your typical treasury data, using the tools and screens you’re used to using in the platform today,” he said. Such as, “How many bank accounts do you have? How much currency exposure do you have? How many payments are waiting for me to approve? Are there any exceptional payments?” he added. “It looks up a report, finds the data and shows it to you. But those are the first hours of use. After that, it becomes more interesting.”
In response to feedback from customers, Kyriba introduced more reasoning to what Mr. Stark called the “basic treasury agent” within the LLM, teaching it to answer more complex queries. His favorite example is a treasurer telling TAI: “’I have a meeting with my bank tomorrow. Summarize all my business and give me a review of how my business with this bank compares to other banks.’ A few different things are layered into that. You’re still leveraging the treasury agent, but the agent grew in terms of what it’s delivering to you and where it has to go to get that information.”
Kyriba introduced new LLM models to show more transparency in its reasoning. “The reason we did that is because customers came back to us and said, ‘It’s not giving me enough of the ‘show your work’ for me to trust that it is getting the right data from the right places.’ And we recognized that from some of the more complex queries,” Mr. Stark said during the virtual session in November where he and a colleague gave NeuGroup members a demonstration of TAI.
The agentic future of treasury. Mr. Callway said, “We’ve launched TAI with a number of built-in agents, like a basic treasury agent. But the bigger vision for TAI is much more than us building agents for customers. Where we’re going is to get TAI to build out a fully-fledged agentic platform whereby customers will be able to do that themselves or they can work with partners to do it.”
For now, Mr. Stark said Kyriba is introducing AI agents that go beyond “just the LLM, which is effectively one agent at work, into a place where you can actually create agents and then assign them to tasks in your workflow. That’s really what we mean by agentic, where agents are stitched together to deliver automated workflows.”
Take an existing process with four parts within Kyriba today, he said. “Kyriba is well-known for its menu maps where you design what your workflow is going to be. For example, if your morning process is: ‘Get my bank balances, set my cash position, create my forecast and, at the end of the month, how accurate was my forecast?’ then you have four tasks and four icons in your menu map.”
“Agentic AI,” he continued, “allows the user to decide which of those tasks in the menu map AI does and which ones you do. And you may add a review or oversight step after each of those tasks that I, as the human, am going to take to review what AI did; or I’m going to let AI do everything and I’m going to check in at the end to see how we did.”
Ultimately, Mr. Stark added, “You will have multiple agents and then you can apply those to determine how you want to automate your workflow. That’s the power that users want, the ability to personalize the AI experience for workflow in your system.’”
What customers want now. During the virtual session, one member asked about the current ability of TAI to produce reports, construct sophisticated graphs and other needs of treasury teams digesting and reporting data. Mr. Stark’s colleague from Kyriba said he heard comments and questions in the same vein from clients at the AFP meeting in Boston in October. After asking the agent a question or to do a task, he said, clients want to tell it, “Now, go and create the report, create the dashboard, create the worksheet, create the extract,” he said.
He and Mr. Stark told the NeuGroup member—a Kyriba client—that AI will continue to offer more visual reporting and graphics in the coming year. More broadly, agentic AI will allow treasury teams to take “two different paths,” Mr. Stark said. One is telling the agent to perform tasks that involve running processes that already exist within the TMS. On the other path, more advanced technology will allow users to place conditions on how, for example, an agent makes a chart by “basically live coding in the moment, where you can tell it the chart you want it to create. Or ‘make the chart you made yesterday,” he said. “That’s what agentic AI will deliver as it learns what kind of visualization you want to see, and it will continue to learn and improve each time.”
Customer feedback will remain integral to how Kyriba drives its TAI road map and meets the needs of its customers. “What we continue to do is to provide a framework and flexibility so that everyone can stress test the ideas,” Mr. Stark said. “Break the darn thing—find a way to make it not do what you want so that we can leverage the power of AI to fit treasury’s emerging requirements: more models, more flexibility, more APIs, MCP servers. There are all sorts of possibilities that clients will come up with that will raise the bar. But it’s definitely a collaborative road.”
Sponsored by:
Dive in