Latest / Extropy on AIR: Finance Transformation, AP Automation & AI in Finance / Execution Beats Experimentation: How Agentic AI Is Delivering Real ROI in Finance
Transcript
- speaker-0: Welcome to Extropy On Air, the podcast from Extropy Advisors, where we dive into the world of AI, digital transformation, and the future of business. Each episode will bring you perspective from leaders, innovators, and experts across industries, all focused on helping organizations, and more importantly, people, thrive in a changing world. Let's get started. Welcome back to Extropy On Air. ready of you new listeners, this is where we're going to talk a lot about how AI is actually transforming business. We don't do this in theory, we don't do this in labs, but we have discussions that ground us in real operating environments. My name is Kevin Zirkel with Xtropy Advisors. I'm going to be your host today. I wanted to open up with a stat that honestly surprised me, but I found this one the other day and it actually comes from MIT. It's a research study and it said that 95 % of AI pilots fail to achieve revenue acceleration, 5%. That stunned me when I saw that number. So, you know, despite all the enthusiasm, all the board conversations that we hear about out there, all the experimentation that's going on in this space, most AI initiatives are not meaningfully moving financial performance. But there is a bit of a twist to this. One of the things that I recently ran across as well is ⁓ some proprietary research done by Bassware. shows something very different happening in finance functions. While overall AI return on investment has risen to 67%, agentic AI is delivering an average of an 80 % return on investment. That's not just incremental improvement. That's true material operational impact. What's really going on here? Why do all these pilots stall, but agents are able to scale? I'm excited to have with me today, unpacked this is Adam Lustig with fastwares. Adam, certainly great to have you with us today. Thanks for joining. speaker-1: Awesome. Thanks, Kevin. Really appreciate it. you know, myself being an AI and an accounts payable nerd, just great to be here with you. you, you put your finger on something that I do think about because there are really two separate AI stories at play right now. Any folks can tend to either mix them up or blend them. One story is experimentation without accountability. I let's proofs of concept, dashboards that typically aren't really acted upon. The other. AI that's actually embedded in how work gets done with measurable outcomes that are tied to it. So the gap between those two has nothing to do with model sophistication or maturity. It's more about anything that actually changes downstream as a result of the utilization of the AI. So, you know, just anecdotally, I was, was talking to a finance transformation lead at a global manufacturer. They'd spent months. building Gen. AI that produced invoice summaries and exception dashboards. Very smart outputs, clean reports, but every recommendation still required a human being to act on it. So the bottleneck never really moved. The moment they shifted to an agentic approach, where the system could actually route, hold, escalate invoices with defined rules, exception resolution, the touch points on the invoices dropped sharply. That's the distinction. Insight sits in a dashboard, but execution is what really changes the close. speaker-0: Yeah, that's great. I think that's kind of the phrase I wanted to hit on that one. I think today's through line is really simple. It's that concept that execution will beat experimentation every single time. if I could, let's shift our framing a little bit here. And before we get into AP and governance and the buy versus build discussions, let's kind of level set something first. When I look through the report, it makes a really important distinction. It says, agentic AI executes work. It doesn't just generate insights like you talked about. So Adam, why does this matter so much? speaker-1: Yeah, yeah, great, great question. Most of these AI pilots, they stop, they end at insights, predictions, anomaly alerts, summaries, useful stuff, but insight alone, as you're saying, it doesn't really move financial performance. And agentic AI really closes the loop. It validates the invoices themselves. It can handle routing for approvals. It escalates the exceptions and it can post actual transactions. all within guardrails that customers can define. That shift from informing work to performing work is where the ROI really shows up. And finance doesn't reward interesting dashboards, does it? It rewards throughput, accuracy, and compliance. speaker-0: I couldn't agree more. That's really powerful as you've stated that. to put it in other words, think of traditional AI. says, here's what you might want to do. Where if you look at agentic AI, it really says, I've done it, but I've done it within the boundaries that you speaker-1: defined. Exactly, exactly. And finance functions love that stuff because finance runs on repeatable, measurable work. speaker-0: Yeah, and that's actually a perfect playground if you will for AP and where AI might sit into that. if we could let's go into accounts payable a little bit. I heard mentioned 72 % of finance leaders they actually say that AP is the most obvious starting point for agentic AI. I think I agree with that but in your mind why do you think AP is such a fertile ground for agentic AI? speaker-1: Yeah, yeah, Kevin. when we think about accounts payable, it's really structured repetition at scale. Invoices are arriving in volume. Data gets extracted, validated, matched, approved, posted, and eventually audited. Rules driven, but exception heavy. And that's exactly where agents can perform well. Think of it like, an analogy, like a factory floor. You don't necessarily have an engineer inspecting every nut. and bolt that goes into every car on that factory floor. You automate the repeatable stuff and you let the experts handle what actually requires judgment. AP is really the same. At that scale, the patterns are clear and those rules are well defined. That's where the agents are in their. speaker-0: So instead of that AI just generating, you know, summary of invoices, if you will, it's actually handling the things that people could historically have done with validation. It's the duplicate checking, it's the routing, it's all these types of things, right? speaker-1: Yep, yep, exactly. And that's why success rates in accounts payable tend to run higher than AI use cases, right? It's concrete. You know what better looks like before you even start. We want fewer duplicates. We want faster cycle time. We want lower cost per invoice, more invoices touched per full-time employee. Overall, we want higher compliance. Clear outcomes mean realistic expectations and real room to perform. speaker-0: Yeah, and I think you see that pretty clearly. think one thing that stood out to me in the report, if I refer back to it, I think this is the correct number. I saw that nearly three quarters, it might have been 72 % or so, say that the deployed AI agents have been successful in AP. I mean, that's a really unusually high success rate for AI. speaker-1: Yeah, it is. It is, Kevin. And some might think it's pie in the sky, but AP success is high because it's not abstract. Fewer errors, faster throughput, lower cost per invoice. When outcomes are truly measurable, the expectations are concrete and manageable. speaker-0: And I think that gets back to the key point that we talked about earlier, that execution always beating experimentation. Obviously that experimentation lacks that operational anchors that are going to be part of it. So I wanted to jump into something a little bit different now. So one of my favorite sections in the report is a split between agentic ROI leaders and followers. And it says about 23 % achieved ROI above 101%. Meanwhile, 42 % achieved 50 % or less. I mean, that's a huge gap, Adam. mean, walk us through what separates those numbers. speaker-1: Yeah. The separation is that high ROI organizations, it's not about smarter AI, it's about how deeply they actually deploy it. So leaders push agents into real workflows. So now you've got compliance checks, reporting preparation, cashflow monitoring. Followers tend to keep AI as an advisory function surfacing insights, but as we said previously, not making them truly actionable. And the mindset there is different too, right? Leaders are executing against defined outcomes where followers are responding to pressure to show that they're doing something. Yep, I can check the box. I'm doing something with AI. Take a look. But these are very, very different places to start. And when I personally, when I speak with finance leaders, I ask three questions. Can you name a workflow where you could define success with a real number today? Do you have rules that define when an agent acts versus when a human steps in? And is AI on your quarterly operating agenda? Or is it just an annual strategy topic that's spoken about at the board level? So the ones answering yes to these tend to have the highest ROI and tend to feel the least amount of pressure because usage builds confidence and confidence then expands deployment. And so the cycle compounds and it becomes a self-fulfilling prophecy. speaker-0: fascinating that you bring that up. the organizations with the highest ROI actually are feeling the less pressure. speaker-1: Exactly. Yep. They're executing. They're not reacting. And that makes all the difference. speaker-0: And again, just to point us back where we started, I think that really reinforces the thesis of the discussion today. It's that execution beating experimentation every time. So let's talk about a question actually we hear a lot at Xtropy. And it always gets back to, you know, we've got some smart people, should we just build this ourselves? And I think this report showed meaningful confusion around buy versus build. You know, how should finance leaders be thinking about this in in your opinion Adam. speaker-1: Yeah, I think overall there's a simple rule there. can think about buy to accelerate, build to differentiate. And so to break that down a little, if AI improves a standard process like accounts payable, buying an embedded solution almost always wins on time to value. If AI creates something genuinely unique to your business model, then building could make sense. Organizations can confuse ownership with value. and owning the code doesn't necessarily create ROI. Execution, as we keep saying, does. And so a real example for you, a financial services firm spent 18 months, this is a customer of BassWare, they spent 18 months building their own invoice automation agent. Great solution for one country, but they operate in 14 markets. Each of those has different tax rules, invoice formats. So a competitor deployed an embedded AP solution trained on billions of invoices globally and hit high automation rates in months. So control matters, but so does time to value. speaker-0: I think that's a really important point. Obviously, we tend to express that to our clients as well. If you look at some of leading solutions in the market, if everyone processes invoices, you probably shouldn't reinvent the wheel to try to build something that's already out there. speaker-1: Yep, yep, exactly. And here's another trap. Organizations often think that they're building AI when they're really just configuring tools that already have some AI baked in. So it's worth getting clear on that before committing 18 months of engineering time. Yeah. speaker-0: Absolutely. Let's kind of move on to something a little bit different, Adam. I want to talk about governance. Again, I this, that 46 % of finance leaders won't deploy an AI agent without governance properly in place. So some might see that as caution. Others, I think, might see it a little bit as friction. How do you think leaders should really think about this governance issue? speaker-1: Yeah, I think a lot of this is it's subject to interpretation, but I think I think governance gets misread pretty consistently and people hear the word governance and think friction where I think of it more as infrastructure. So when you hire somebody into accounts payable, you don't tell them post whatever you think is right. You define approval levels, you define escalation paths, you provide oversight. And so you build trust. through having some structure. Agents are really the same. The guard rails should not be what slow you down. That's what makes autonomy possible at scale. speaker-0: Yeah, and that governance, it really isn't going to slow scale. It actually is going to enable scale, right? Well, one other thing, Adam, we can't ignore this in the report. says 30 % of leaders say that AI agents are already displacing finance jobs. I know this is a big one, particularly maybe for some of our listeners, but how do you address that tension within your organization as people start to see numbers like that? speaker-1: Exactly. Yep, yep. see a lot of things posted on social media and elsewhere about the jobs that are the safest and those that might get replaced by AI. But I think it's really important to separate tasks from roles. Agentic AI, again, think about the factory floor. Agentic AI handles the repetitive stuff, the data capture, the invoice matching and coding, the compliance checks. Those were never the strategic center of finance. And so what we see now is a... an evolution of the role. so another example, anecdotally, one global consumer goods company, they had their AP team spending a lot of their time on data entry, chasing approvals. After deploying agentic automation, that dropped sharply. The team shifted into supplier relationship management, early payment programs and focus, recovered hundreds of thousands in missed early pay discounts in year one alone. So it's not about job loss, it's about role upgrade, repositioning people to be more strategic from the tactical day to day. And organizations that communicate it that way tend to see a lot smoother adoption from their workforce. speaker-0: Yeah, and I think that's the absolute right narrative. mean, it really shouldn't be that AI is replacing finance. I don't ever see that happening. It should be AI is looking to upgrade finance and it will. It's going to lead to a revolution and potentially a change in how we do work moving forward. speaker-1: Exactly. Yep. should be framed more as a backup, not a replacement. And if you frame it correctly, I think people will tend to lean in instead of resisting. speaker-0: you right there. I'd like to maybe just step back a little bit and look at the bigger picture. mean, let's kind of close at altitude, if you will. AI return on investment. You see that overall it rose from 35 % to 67%. Agentic AI, I think, sits at 80%. But the most important sentence in the report, and I pulled this one out specifically, was that agentic AI is becoming less of a technology choice and more of an operating model decision. What does that mean in this BASFAR report? speaker-1: Yeah. So overall, think, you know, one of the key points is the fact that AI, it's no longer a side initiative. And when agents coordinate across accounts payable, procurement, treasury, audit, the value moves beyond task efficiency and into real operating leverage. Fewer exceptions, faster close cycles, better working capital visibility and execution. And at that point, It stops becoming a technology experiment and it starts to become more how the finance function runs today and in the future. So organizations that feel great about their AI journey in five years won't necessarily be the ones who ran the most pilots. They'll be the ones who picked one workflow, made it work, built confidence around that, and then scale it from there. Again, going back to the theme of execution beats experimentation. speaker-0: I love that you closed with that. We opened with it, let you close with it. And really, I do think that's the difference between a pilot and a platform. But again, Adam, really appreciate you joining us today. To recap, know, if 95 % of AI pilots fail to accelerate revenue, the lesson isn't, don't use AI. We certainly don't want you to walk away with that. The lesson really is, stop piloting in isolation, make sure that you truly define measurable workflows, make sure you embed execution into every one of these initiatives, and absolutely install governance. It will make all the difference in the world. And then at the end of the day, scale deliberately. Because as Adam just mentioned, execution does beat experimentation every single time. Adam, this has been fantastic. I think our listeners are going to get a lot out of it. So I'd certainly appreciate you joining us today to cover some of these. speaker-1: Awesome, Kevin. It's mutual. really enjoyed it. Thanks. Thanks for your time. speaker-0: Awesome, you bet. And as we close again to our listeners, if you want the full data set and the analysis that we've been going over today, I certainly encourage you to go to BassWare's website, BassWare.com, download BassWare's report entitled AI to ROI, Unlocking Value with AI Agents. I think you'll really enjoy it. So until next time, again, this is Xtropy on Air. Remember to like, listen, refer a friend. We'll see you guys next time. Thanks so much. Thanks for listening to Xtropy on Air. We hope you found today's conversation insightful and inspiring. Be sure to follow or subscribe so you never miss an episode and connect with us at Xtropy Advisors to continue the conversation on AI and digital transformation. Until next time, I'm Kevin Zirkel and this has been Xtropy OnAir.