Governing AI Agents Without Losing Cost Control With Basware

[00:00:05] [SPEAKER_01] Who gets blamed when an AI agent makes a costly financial mistake? Spoiler alert, the regulator is unlikely to accept, hey the bot did it or AI did it. Well in today's episode of AI at Work, I'm joined by Jason Kurtz. He's the CEO of Basware. And together we're going to look at governed autonomy and all the controls that allow AI agents to act while keeping people accountable.

[00:00:35] [SPEAKER_01] And Jason will explain today why CFOs remain responsible as agents handle greater portions of finance and how companies can train their new digital teammates like employees and what metrics reveal whether automation is actually producing a positive return on investment. And we'll also dig into an obstacle receiving far less attention, although it has been getting more and more recently,

[00:01:02] [SPEAKER_01] that is token-based pricing, which is making enterprises' AI costs incredibly difficult to predict and maybe that could slow down adoption. So today Jason is going to be joining us to offer a view of guardrails, audit trails, workforce training and why companies might need people whose job is managing it and why companies will need people whose new job is managing agents.

[00:01:30] [SPEAKER_01] Plenty for every single leader listening to take upstairs. But enough from me. Let me introduce you to Jason right now. So thank you for joining me on the show today. Can you tell everyone listening a little about who you are and what you do?

[00:01:46] [SPEAKER_00] Hello everyone out there in podcast world. This is Jason Kurtz. I'm Basware CEO. I've been with the company now three and a half years. And Basware is the world leader in invoice lifecycle management. And we're super proud and excited about that. And I am an Atlanta, Georgia-based CEO of a Finnish company and thrilled to be here. I've been in this industry for nearly 25 years. For many years, I worked at a competitor.

[00:02:15] [SPEAKER_00] And I had a lot of respect and admiration for Bazware and the products and the company. And as we say in the States, if you can't beat them, you join them. And so that's how I've ended up here.

[00:02:24] [SPEAKER_01] Wow. What a great story. Well, thank you for sitting there with me today. And before you join me on the show, I was doing a little research on you. And I read that you argued. Yeah. Don't worry. It was all good. Okay. Okay. What I actually encountered was you just arguing that the future of enterprise AI depends on governed autonomy. So for people who say, I don't want to lose anyone here. What does that mean in practice?

[00:02:50] [SPEAKER_01] And where can organizations draw the line between what an AI agent should decide independently? And also, what still requires human approval? Because I go to so many tech conferences, they all talk about human in the loop. But what that looks like and where it is, that's where things often get a little murky.

[00:03:09] [SPEAKER_00] Yeah, that's a great question. Look, I think from a governed autonomy perspective, and when you think about AI, it is AI you can put your name and reputation to, right? You, as the CFO or as a user or as a company, you set the guardrails that AI works inside. And then every decision is logged, audible, auditable, explainable. Those are the important things, right?

[00:03:36] [SPEAKER_00] And then this operating model, if you think about it in that governed autonomy approach, lets every organization decide and expand AI's authority over time within their own defined controls, policies, audit trails, so that autonomy, accountability, and integrity advance all together in kind of our harmonious approach. And this really, I think, allows companies to ensure that they don't take on compliance risks,

[00:04:04] [SPEAKER_00] but they get all the benefits of automation.

[00:04:08] [SPEAKER_01] And as I said, I'm into so many tech conferences, and everyone's talking about agentic AI. And one of the things that stood out to me is the world of finance almost, it feels like the ultimate test case for agentic AI, because mistakes like financial, regulatory, and sometimes legal consequences enough to keep anybody awake at night. But as agents begin handling more finance workflows, how does the role of the CFO change? How do you see that evolving?

[00:04:36] [SPEAKER_00] Yeah, I think the role for the CFO does change a lot. But maybe let me start with what doesn't change. And that's probably the most important thing. And that is still, ultimately, the CFO has responsibility for signing off on the books or the financials of the company and ensuring 100% compliance with all the laws and governing requirements around the world, wherever that company operates. And that's still where are the most risks. So that doesn't change.

[00:05:05] [SPEAKER_00] At the end of the day, no matter how many AI agents you have in place, the CFO still ultimately has that responsibility, which is an important thing to keep in mind. Given that, though, I think the CFO role does change in a couple ways. One, I think you need to lead the organization in adopting AI, but within the comfort of those guidelines, within that comfort of government autonomy and showing folks how to allow AI to build autonomy over time within your guardrails, leading on that front.

[00:05:34] [SPEAKER_00] But you also have to learn some of the more interesting things around how do you work with a team of agents, right? Finance and AP, as an example, accounts payable, is a great place to use AI agents or agentic AI. And I think that changes the role to being one of an orchestrator, of being one that is even more analytical probably than in the future. And really, you need to learn how to work with agents as, and you'll hear me use this phrase probably, as digital teammates quite a bit.

[00:06:04] [SPEAKER_00] So while there's some things that don't change, there's a lot that does change as well.

[00:06:09] [SPEAKER_01] And if we go back before the days of AI, I suspect if we've both learned anything from our careers in tech, it's if something can go wrong, it will go wrong, probably when you least expect it, which is for some reason it's always on a Friday night, but I don't know why that happened. But of course, when it does now, regulators are not going to accept, hey, the AI did it as an explanation when something goes wrong. So how should companies establish ownership and accountability

[00:06:36] [SPEAKER_01] for some of the decisions made or even recommended by autonomous agents?

[00:06:40] [SPEAKER_00] I think that goes back to what we were just talking about. What doesn't change for CFOs and companies is that they still are ultimately responsible. As you said, regulators don't accept that AI did it. It's still the company's responsibility. It's still the CFO and CEO's responsibility. And then I think what we need to, it really goes back to that mentality of agents are digital teammates, right? We need to treat them just as you would a new employee. You don't give a new employee who starts on day one,

[00:07:07] [SPEAKER_00] the ability to pay out millions of dollars or euros or pounds in supplier invoices on their first hour of the job, right? You have to train them on your processes, your policies, and you need to validate their behavior and they're doing it the right way. It's the same mentality with AI, right? AI should start as an advisor, then a collaborator. And then only then once you've gotten that, those policies in place that govern autonomy, then it can be an operator making decisions on its own.

[00:07:37] [SPEAKER_00] I think that's really the key to success in this new world of AI agents.

[00:07:43] [SPEAKER_01] And if either of us took a quick scroll down our LinkedIn news feed, it would reveal a lot of understandably anxiety around AI removing jobs, positions, even entry-level jobs for people coming out of education. But I was reading that you see another possibility, and that is entirely new roles emerging around managing AI agents. And really, if we look back, there's not many articles on there,

[00:08:09] [SPEAKER_01] roles that didn't exist 25 years ago that exist now. The world does keep changing and evolving. But I'd love to ask you, what would this new workforce look like, do you think? And what skills could possibly people listening of any age need to supervise digital workers effectively? Maybe we can point a few people in the right direction here.

[00:08:30] [SPEAKER_00] Yeah, absolutely. And I do think it's an important point, right? I think I wasn't around when the manufacturing or the factory came out and there was this fear, but I do know I've read that there was this fear that it was going to destroy jobs and eliminate jobs. It didn't destroy jobs. It changed jobs. It created jobs. And I think that's the same mentality that we need to have with AI, that it will ultimately probably create more jobs, but it will certainly change jobs, right?

[00:08:59] [SPEAKER_00] And one example is, again, I love the analogy of these digital teammates as AI agents, as digital teammates, is we need to learn as people how to manage AI agents, right? That's a different skill set than having you or I come to work and we go have a coffee and we get to know each other, right? In the world of AI, right, we've got to, it's that defining boundaries, educating on rules and the way we work and validating data.

[00:09:29] [SPEAKER_00] That's how our jobs are going to change. I think we'll be more analytical. I think we'll be orchestrators of process. We will be more, you know, we'll take away less of the day-to-day manual, an hour-by-hour manual task, and we'll do more analysis. We'll do more exception handling and we'll use more judgment, right? That's where I think it's going to be really important is that we have experience with people, knowledge with people, and we'll need them to exercise that knowledge and judgment

[00:09:59] [SPEAKER_00] that AI may not have right away. And we'll have to decide, we'll have to review the outcomes of AI, that's the analytical piece, and decide where AI should and should not have authority. And I think that's the really exciting part is, ultimately, you've always heard about automation making jobs better. AI really has the power to take that to another level, eliminating those manual, mundane tasks and giving people more knowledge, analytical, insight-based tasks.

[00:10:29] [SPEAKER_00] And I think that's really exciting. And, you know, at Batchware, we're trying to make it really practical for folks. We introduced a training program around how to work with AI agents, right? Because that is something that nobody is thinking about and no one is doing today. They've never done it. And so how you get over that fear of eliminating jobs and how my job might change, let's show people practically what each AI agent that we create is supposed to do, what its goals and objectives are,

[00:10:57] [SPEAKER_00] and how you as an individual should manage that agent as a digital teammate. I think it's the first of its kind in our industry, but it's something that's going to be needed across every function around the world in every company.

[00:11:09] [SPEAKER_01] And I think most companies out there believe that human oversight will remain important. And again, to try and give people a few actionable takeaways here, when should they begin preparing employees for these new responsibilities and how should they do that? What should AI training and upskilling even look like beyond simply teaching people how to use these latest tools? And one of the things that springs to mind is teaching them to question the output, which I think is important too.

[00:11:37] [SPEAKER_00] Yeah, I would say if they haven't started, they should have already done so, right? And I think the reason why it's so important to get started now, if you haven't already, is because you have this interesting gap. You have boards and CEOs and CFOs saying, hey, thou shalt go do AI, right? Everybody's doing it. We've got to do it. We've got edicts. And you've got people on the front line who are doing work who are worried about, is this going to take my job?

[00:12:07] [SPEAKER_00] Is it going to change my job? Am I ready for that? And unless you can bridge that gap with training and education, like that AI agent program that I talked about from Bazware, then you've got this gap of people at the top saying, thou shalt, and people at the bottom saying, I don't want to adopt that. It's going to change the way I do things. It might eliminate my job. We've got to educate and show people very practically how working with AI agents, how it works, what it looks like, and how it makes their jobs better and more valuable

[00:12:36] [SPEAKER_00] and makes their jobs future-proof. That's the gap that exists today. That's the gap that Bazware is trying to eliminate from an accounts payable function perspective, but it needs to be done across all functions.

[00:12:48] [SPEAKER_01] And another issue that we've been seeing grab more and more attention this year is the economics of enterprise AI, especially with token-based pricing. Costs can become unpredictable as usage scales. We've seen a lot of people run out of budget by Q2 and wondering how they're going to get through the rest of the year. But I mean, how big a problem is this when companies just try to move from a successful pilot to thousands or even millions of AI-driven transactions out there?

[00:13:16] [SPEAKER_00] Yeah, this is a really important topic, and I'm glad you asked about it. And it really, I had this epiphany moment around this a couple months ago, I was at a dinner, a CFO dinner sponsored by Gartner back with about 40 or 50 CFOs. And I was standing at the front of the room, and I said, how many of you can point to and measure the ROI you've gotten from your AI investments? No one could raise their hand. No one raised their hand. I said, how many of you can tell me

[00:13:45] [SPEAKER_00] exactly how much you spent on AI investments? And every single hand went up. So I think the point of that is, it's really easy to measure cost. It's really hard to measure the ROI in many cases. Now, fortunately in AP and with Basware, I think that's different. But here's what I think is really the issue. Companies, particularly enterprise size, so the ones that we serve at Basware, have spent decades, literally decades, putting procurement practices in place

[00:14:14] [SPEAKER_00] to avoid or eliminate variable pricing, right? They want to move to fixed, predictable pricing because that's the way companies work. They work on budgets. They work on visibility, managing to budgets. And tokens are the exact opposite of that. It is a variable cost that is unpredictable. And frankly, it's slowing adoption, right? The token model reduces the adoption that could be happening from an AI perspective. And companies are spending hours upon hours

[00:14:43] [SPEAKER_00] and full-time people roles just managing the adoption of AI only on the cost piece of it because that's what they can measure. And they're seeing these costs, as you said, running out of budget in Q2. It makes no sense, right? Why would you spend full-time people to manage that cost? It just is a barrier to adoption. So I think companies with finite budgets and finite skill sets shouldn't have to choose

[00:15:11] [SPEAKER_00] where to build and buy their AI, right? They should just be able to adopt and have predictable ways of doing it. What is forcing people who have finite AI skill sets and finite AI budgets is to really pick and choose. And I think what you're seeing as a result of this is you have companies who are trying to build their own AI capabilities where it is growth-oriented, revenue growth-oriented, and where they have specific data and domain knowledge, for example, customer-facing functions

[00:15:41] [SPEAKER_00] and processes and experiences. And because they have limited budgets and want predictability and limited skill sets, they tend to buy that capability, that AI capability, and embed it in the software that they already own. It's already embedded in their systems and workflows because that is more predictable and is faster time to value and has better economics around it in terms of ROI. So you're really creating this massive adoption issue around tokens

[00:16:10] [SPEAKER_00] when I think companies want more predictability and it's forcing this kind of build versus buy mentality.

[00:16:15] [SPEAKER_01] Yeah, 100% with, yeah, I think unpredictable consumption-based pricing will ultimately hold back agentic AI adoption if it's not put in its place there. So I'm curious, what kind of commercial models do you think would give enterprises enough cost certainty to automate more workflows without discovering that successful adoption actually means an unexpected large bill landing on the doormat at some point?

[00:16:40] [SPEAKER_00] When I talk to our customers, what they want is just like what they want from a SaaS software solution, which is a predictable fixed price for some set of value or outcome that they are getting, right? And they don't want completely variable or token pricing. I had the CFO of a prospect that is now signed on with BASWARE. That was the one question she asked me when we met, when we were talking,

[00:17:10] [SPEAKER_00] kind of in final discussion, she said, do you price based on tokens and variable consumptions? Because if you do, we're not doing business with you, right? That's, it is a huge barrier. It is going to be to fix price just like your SaaS software bill is. That's where the market has to get to. I think it's going to be sooner than later, but I think no doubt it's coming in the next couple of years.

[00:17:33] [SPEAKER_01] And at BASWARE, you sit close to finance processes where organizations can measure outcomes rather than simply counting AI usage. So what should leaders be doing to measure and determine whether AI agents are genuinely creating business value? And what would tell you that a process should be maybe not even handled by AI at all and be better handled by a human employee if it's too expensive?

[00:17:59] [SPEAKER_00] Yeah. Yeah. So first of all, I think, and I think your question hits at it, we should only invest in AI if it delivers ROI, right? And so anything other than that doesn't make a lot of sense. And to do that, you have to have great data to train your AI so you have great outcomes. So where there are repeatable tasks where you can collect data over and over and have a data set that can inform, train, and educate AI, that's where it makes sense. So for example,

[00:18:28] [SPEAKER_00] if you took a survey of CFOs and we have done this at Bazware, they would say finance is a great place to start, particularly AP or accounts payable is a great place to start. As an example, Bazware, we've sit on 2.5 billion invoices that we process over our lifetime, which is over 10 trillion euros in spend. And so our AI is trained on that unique, largest data set in the world. And so it can deliver ROI faster than any other solution. So the time to value

[00:18:58] [SPEAKER_00] and the amount of ROI is phenomenal. And so that's a great example of a process that makes sense to use AI versus and then have a human in the loop, but not necessarily be a completely manual process. How would I measure that? And I think with the other part of your question, I would think about what we call kind of life cycle autonomy rate. So if you look at an end-to-end from ingestion to payable invoice processing or AP process, what's your touchless rate or automation rate of that? We should measure the working capital

[00:19:27] [SPEAKER_00] freed as a result of running this autonomous process. We should think about on-time payment rates to our suppliers to make sure that we are providing the working capital and keeping our suppliers engaged and being a customer of choice to our suppliers. We should think about the discounts captured based on payment dates. We should think about AI decision rates. So what percentage of an AI agent's decisions are accurate and were correct the first time? And we should think about, you know,

[00:19:57] [SPEAKER_00] because it's always near and dear to a CFO's heart, what's our compliance coverage, right? Our customers work across the globe and they are subject to a number of compliance and invoice mandates country by country around the world. What does AI do in terms of a coverage rate for that? If you do all of those things well, you will drive extreme amounts of return on your investment in AI. So those are the kind of things that we're recommending our customers measure.

[00:20:23] [SPEAKER_01] And you've had a hugely successful year already. I was reading that Basware was named, I think, Forrester AP invoice automation lead. I think that was in June. But anything else that you're excited about as we race through the rest of 2026? Anything listeners should be aware of your focus for the rest of the year?

[00:20:42] [SPEAKER_00] Yeah, I mean, I think it's a couple of things. So continuing to bring AI innovation and new AI capabilities embedded in our product to our customers. We've had over 400 of our customers already invest in AI ambassadors AI capabilities and over 100 have already bought multiple AI capabilities. So we're super excited about that and we're particularly excited about our roadmap of new capabilities that will continue to drive value for our customers. So that's probably first and foremost. And then we've got

[00:21:12] [SPEAKER_00] some surprises up our sleeves that will be coming so that we'll continue to expand that value that we can deliver from AI and our data which we think is really important from driving value to our customers.

[00:21:26] [SPEAKER_01] Ah, you've left us with a teaser there. Law one of podcasts always leave the audience wanting more. So we will invite you back on there. But for people listening, where's the best place for them to find you or your team online or maybe even meet your team on the road as the tech conference season opens back up soon and find out more about anything we talked about today? Where should they go?

[00:21:48] [SPEAKER_00] Well, we like to say that Basware is everywhere around the world. So you'll find us on the road. I promise you that. But first and foremost, I'd recommend you go to bazware.com B-A-S-W-A-R-E dot com. You should also follow Bazware on LinkedIn. We're huge LinkedIn folks. I am also our chief LinkedIn officer. So follow me, Jason Kurtz, K-U-R-T-Z at link on LinkedIn. Love to have you follow me there. And we'll be at a number of events and conferences, but a couple of big ones coming up. We'll be at the Deloitte Shared Services Conference

[00:22:17] [SPEAKER_00] coming up in a few weeks and also at SAP Connect. So find us there and many more.

[00:22:24] [SPEAKER_01] I will have links to absolutely everything. I encourage people listening to check that out. And we will be speaking to you, if not later this year, early next year, find out some more about those announcements and teasers you left us with today. But just a big thank you for your time and sharing everything today.

[00:22:40] [SPEAKER_00] Thank you for having me. It was a lot of fun. Hopefully it was valuable for the listeners.

[00:22:45] [SPEAKER_01] I think Jason leaves us with two questions every AI program must answer. Who owns the decision and can the company predict what autonomy will cost? Treating agents as digital teammates, providing a useful framework is a great start. Start them as advisors. Validate their behavior. Give them greater authority on a gradual basis and record every single decision. But at the same time, train employees

[00:23:14] [SPEAKER_01] to manage boundaries, investigate exceptions and most importantly, question outcomes. Otherwise, human oversight can quickly become just an impressive phrase like human in the loop where nobody's quite sure exactly what it means, but it sounds great. So thank you so much to Jason for talking about this in a language everyone can understand, especially governed autonomy, AI pricing and the changing role of finance teams.

[00:23:43] [SPEAKER_01] And you can learn more at basware.com and follow Jason and Bassware on LinkedIn. But before approving another AI pilot, could your organization explain who remains accountable, how your success will be measured and what the bill might be? Lots for you to think about here. Let me know techtalksnetwork.com but I'm afraid we're out of time today. I'll be back again real soon with another guest. Bye for now.