How BlackLine Turns Finance AI Investment Into Measurable ROI

[00:00:00] Your agents aren't producing accurate answers because they don't have a complete semantic understanding of your data. And Denodo is solving this and solving it through semantic consistency. Through semantic consistency, your agents can start making accurate predictions in real time. So see what else Denodo can do by visiting denodo.com to learn more. But now let me introduce you to today's guest.

[00:00:33] Has finance reached peak AI adoption? Or are CFOs finally pausing long enough to ask whether any of these investments are producing measurable results? Well, recent Gartner figures suggest that AI adoption in finance is moving from 37% in 2023 to 58% in 2024.

[00:00:58] But it was followed by a much smaller increase by only 1% to 59% in 2025. So this slowdown could suggest enthusiasm is fading. It could also mean that finance teams are leaving the experimentation phase and beginning the harder work of putting AI into everyday operations. Or am I getting that completely wrong?

[00:01:24] Well, today I welcome Jeremy Ung, CTO at Blackline, back onto the podcast. He leads the company's product and tech organisation and has a closed view of how finance teams are approaching AI costs, controls and operational deployment. So we'll cover everything today from why token spend can become unpredictable to whether businesses should build or buy their AI capabilities.

[00:01:54] And explore why finance leaders should avoid inventing artificial measurements for AI. Too many acronyms there. So if your company has spent the last two years buying AI tools and now wondering what do you have to show for them all? Today's conversation will be a timely reminder that the business outcome still matters far more than the size of the model. But enough for me.

[00:02:22] Let's get Jeremy back onto the podcast right now. So a massive warm welcome back to the show, Jeremy. For anyone that missed our previous conversation, can you remind them with a little of who you are and what you do? Well, thank you very much for having me back. Jeremy Ung, Chief Technology Officer at Blackline. I run our product and technology organisation and build all things Blackline related. Awesome. Well, thanks for joining me once again.

[00:02:50] A lot I want to be talking with you about today. There's so many big talking points out there at the moment, especially around AI adoption, because in finance it appears to have maybe plateaued after several years of rapid growth. So do you see that as a sign of slowing enthusiasm? Or is it evidence that finance teams are moving from experimentation towards making AI work in everyday operations? Or are they just looking for ROI or something completely different? What are you seeing here?

[00:03:20] I think there's a variety of things that we're seeing. And so build versus buy is a very preeminent conversation that's happening right now. And what finance teams are trying to understand is in light of the regulatory landscape, the controls that they need to have, regulators and auditors, what do they need to do to maintain their posture, to maintain compliance and controls and auditability? And do they build that themselves, which is very difficult?

[00:03:47] Or do they buy it in order to combine that with vendors like Blackline in conjunction with the model providers? The other thing that we're seeing is that the ROI conversation is real. The token spend, I'm sure you've seen the articles on people blowing past their token budgets in months, days, weeks, pick your figure. But people are blowing past token spend and it's hard to track. Yeah. And it's not consistent. It's not deterministic.

[00:04:13] And so people, especially in finance, you have a CFO who's responsible for cost. And so they don't want that cost uncertainty. So they're trying to understand what's the ROI of applying AI to certain workflows. And I think it's turning out in a lot of cases that that uncertainty is resulting in stalled adoption. And so those are the trends that we're seeing. Build versus buy is an ongoing conversation, but is being met by this need for control and governance.

[00:04:41] The need for AI adoption and the benefit of it is perhaps being overshadowed by the cost uncertainty. And so that's where we've really been approaching this from a Blackline perspective with our control console, which we launched in June. Yeah, late June. It's really about how do we take that uncertainty out of the picture? How do we provide controls and governance?

[00:05:05] So whether you want to build on Anthropic or open AI or what is increasingly popular, new models from China, you're seeing in the news. There's ways that we can provide that governance across that spectrum so people have certainty in the auditability, in the controls, infrastructure that they need to have. And then separately, in terms of cost certainty, this is where our approach has always been to provide predictability,

[00:05:30] whether it's how do we manage tokens spent on your behalf, but also how do we turn certain workflows into more deterministic ones so that it improves their auditability and their outcome, but it also improves cost and performance. So those are the things that we're seeing in the industry right now. And looking back over the last few years, I think many finance leaders invested heavily in AI during that initial adoption wave.

[00:05:53] But as attention is now turning towards return on investment and measurable difference, improving business outcomes, etc., how should they assess whether those assessments are actually improving their financial operations rather than just simply adding yet another layer of technology or even complexity? I mean, I'll bring this as a parallel to what I get asked a lot in my domain, right?

[00:06:18] How do I measure the effectiveness of AI usage for my developers, for my product teams, etc.? And it comes back to business fundamentals. Don't create artificial metrics to measure AI ROI. You really have to look at the total spend in an organization against the outcomes that are being delivered. So is it improving the rate of your close? Is it improving accuracy? Is it improving business outcomes that you're trying to achieve elsewhere, right? That's how you should measure ROI.

[00:06:47] You shouldn't construct artificial metrics on token spend. It's really just the spend in your organization. Whether you allocate that capital to humans or to AI spend, it's about achieving those end goals and outcomes, and hopefully in a more efficient and faster way than you're doing. So that's, I think, the biggest thing that I would encourage finance and other leaders is to think about the end metrics still. The business metrics haven't changed. AI is a contributor or a driver to that, but you shouldn't be measuring AI in isolation.

[00:07:17] I love that line there. Do not create artificial metrics. Brilliant. Absolutely fantastic. And you also talk about operationalizing AI by embedding it into core finance functions. So what does that look like in practice? And which processes offer the strongest opportunities to deliver that measurable value that we're talking about? Yeah. I think if I look at the breadth of finance teams, a lot of the processes are still manual.

[00:07:47] Whether it's outreaches to vendors for an accruals process, which we've automated through very accruals, or you see other processes like we capitalize our software developer, right? And we go through that process. There's still a lot of human touch in that. And we're beginning to see value and ability to automate more of that with AI. Now, you might ask, why hasn't that been automated in the past through traditional SaaS approaches? Well, there's a lot of variance in how organizations do these processes.

[00:08:14] There's a lot of human touch required to reach out to different stakeholders. And that's something that was difficult for SaaS software to do at scale in a meaningful way. And we're seeing a shift in industry where we're combining the flexibility that AI gives us. We're combining different go-to-market approaches. You might have heard of forward-deployed engineers being a very popularized term these days. But those are the approaches we're leveraging to give people a degree of, let's call it customization, or adaptation to their workflows that is unprecedented.

[00:08:44] And I think that gives them the ability to get the last mile covered and automated. And when we talk about finance, it operates in an environment where accuracy, controls, auditability, and trust all matter enormously. And AI has got its own reputation for hallucinations and making the occasional mistake there. So how do you introduce greater AI automation without creating new risks,

[00:09:10] or even losing the oversight that finance leaders really need? This is where I think the approaches around observability, controls, and governance have changed. Now it's about how do you do this at scale. So we've had a lot of conversations with different audit firms to build a perspective on this. And I think one of the questions has been, how do you see, as an audit firm, the work changing with AI?

[00:09:35] And the answer has been pretty unanimous in that we're seeing an increase in demand for work. It's not going to go down with AI. In fact, people are being asked to look at more data. And so if you think about this, this becomes a scale problem. And so how do we build software as Blackline that meets that scale need? And this is where our control console and our new architecture come into play. It's about understanding AI agents and activity at scale.

[00:10:03] And I think what that means is very specific. One, I think we need to understand all the inputs that are going into AI agents. What are the prompts? The reasoning? What models were used? In what context did these models operate? So for example, what data did they have access to? What tools were they able to utilize? What permissions did they have?

[00:10:24] And this goes and creates a recipe or a list of ingredients so that you can verify what they're doing, how they perform this. Not just now, but in the future, you'll be able to look back at the history of what your agents did and be able to understand were there any biases? Were there any areas or models used that may be problematic that we need to go look at? And that level of observability is critical to being able to leverage AI at scale in finance. And I think you see this very commonly.

[00:10:53] To draw a parallel, you see this very commonly in the food industry, right? It's very common now. You'll see the list of ingredients on a package. How many calories does it have? What ingredients are in there? And you'll even get, are these products sourced in a specific country, right? Like, where is the meat from? And I think that's kind of what we're building for AI agents and what people are expecting to come out of this is to get that level of control and governance. You need to be able to do that, but at scale across hundreds of thousands, if not millions of AI agents operating in an enterprise.

[00:11:23] And so that's the challenge that we're seeing. That's the problem that we've solved through our control console. And I think that's an interesting and new shift in unlocking AI adoption. And many businesses and people listening will work with organizations that still have fragmented financial data and manual processes that are spread across so many different systems. They're making it even more complex.

[00:11:46] So how does that underlying complexity limit what AI can achieve and what foundations need to be addressed before companies can scale successfully? Any advice that you can help anyone listening that's in a situation like this right now? Look, garbage in, garbage out, right? It's a phrase for a reason. And so having a strong data foundation and accurate data means that you're going to have better outcomes.

[00:12:12] The one thing I wouldn't block on, though, is that, you know, look, disparate systems is a challenge that's solvable. And that's where we're applying AI to bring together data into a singular way, a singular source of truth. And I think AI has really accelerated our ability to understand different source systems. You know, we have many customers who have mainframe systems who, you know, they don't, you know, like the talent shortage is there for COBOL and other developers.

[00:12:40] So to get data out of those systems or to change those systems is difficult. AI is actually great for that. It allows us to build rapid connections to these systems so we can create the connections and can bring it into a centralized repository that you can use to build off of and automate. And so that created that central source of truth has never been easier. And I think it becomes critical foundational infrastructure for automation through AI.

[00:13:06] And I think there's often a pressure to invest in the very latest AI capabilities. But I'm curious, do you think finance could generate even better returns by improving and expanding the AI systems that they've already deployed? Rather than just continue adding new tools, new features, et cetera. Are they moving too fast sometimes? I think you're seeing maybe some of the pullback from just buying tools to, you know, point solutions to solve the problem.

[00:13:34] I think you are seeing platforms really be what people are anchoring on. Platforms that can provide that breadth of capability but also the flexibility to build on top of. So proving capabilities that provide that foundation like our clothes, our transaction matching capabilities. But then we allow you to extend that and build on that with your own agents and others.

[00:13:55] I think that's the key thing that we're seeing here is that, you know, it is hard work managing that fragmented ecosystem because it creates the same governance problem, right? That we had with SaaS, like you had, you know, shadow IT. Everyone was procuring SaaS software on their credit cards. It created a governance problem. It created an information security issue.

[00:14:16] And so as we're beginning to, you know, as people expanded the AI aperture and consumed all these different models, they're seeing that this creates the same governance problems. And so this is where platforms like Blackline are addressing that need through control and governance layers, but also providing that stable platform that you can build on without having to manage that model complexity. And as AI becomes embedded into financial workflows, how does the role of the finance professionals change?

[00:14:45] We hear a lot around AI replacing people, et cetera, but I think it's more about working alongside one another. So which tasks should increasingly be handled by technology and where will human judgment become even more valuable? Do you see a clear split there? I do. I mean, I think the shift has been to AI agents doing work in human steering. If I look at how my software development teams build software today, they're defining specifications, goals, and outcomes.

[00:15:16] And they're steering the AI agents, guiding, nudging, and approving the work or reviewing the work. And it changes the nature. And I think the same is true of finance. You have your goals. And I think, again, back to the metrics point you raised earlier, these are tied to business goals and outcomes. Yeah. Right?

[00:15:32] So define those outcomes and have agents operate against those, whether it's accelerating your close process, whether it's improving your transaction matching accuracy, or reducing the amount of time it takes to collect on debt and things like that. These are ways that we can give AI the goals and have it operate. And then as humans, you get surface the outputs that need to be approved. You get shown the chain of thought and reasoning, just like you would have a team of people working for you.

[00:16:02] Right? And so that gives you scale and leverage. And I think the tools change. So if every employee is now a manager of a team of agents, the tools that you need, the skills that you need to manage that team changes. Maybe you're used to being an individual contributor and just doing the work yourself. Now you need to understand how a team of people or a team of agents, in this case, is going to work for you. And I think that's the mindset shift that we're helping finance professionals come to grips with and help them make that transition.

[00:16:32] Because I think it's so critical in this era. I love that. And I always try and give people listening a valuable takeaway. So if we were to take, let's say, a CFO that could be listening today, they want to get more value from their existing AI investments over the next 12 months. Any practical steps you think they should take or that might help them move from isolated use cases towards those measurable improvements in finance operations and business performance and everything that we're talking about here today?

[00:17:01] Any tips or advice on where they should begin? Yeah. I think the cross-section of what you can do with AI has become vast and perhaps overwhelming. But the platforms have been the key to building how you extend and accelerate your scale. And so platforms like Blackline are a great leverage point. And don't just think about where you operate siloed use cases like Close as a singular process.

[00:17:28] Think about it as a horizontal process that spans your business and the business metrics you're trying to achieve. And think about how you enable agents to work in that process as a scale function for your workforce. And I think that's a critical mindset shift is you don't want to think about these as discrete, isolated processes. You want to think about this as an end-to-end process that you're optimizing for your business. And you think about how you leverage platforms. Obviously, I'll advocate for Blackline in that.

[00:17:56] But how do you leverage those platforms to enable that scale? And these platforms, we're expanding our capabilities. We're giving you building blocks. Because I think now it's changed to agents helping individuals scale, giving those agents tools. And this mindset shift, I think, is unique in that we're not equipped for a workforce that is 100x larger than the human workforce powered by agents. I think it's a different world, one that we didn't grow up living in.

[00:18:23] And so we're all building that skill set, those tools to do that. I think as CFOs get educated about what that means, I think the possibilities get unlocked in terms of what end-to-end business goals that you can solve or outcomes you can optimize for. Fantastic advice. So what about yourself and Blackline there? Anything you can share about what excites you at the moment? You said you had a big release in June. Anything you can share around?

[00:18:47] What is really changing in software right now is really exciting to me as a technology professional in that we are designing and building software that is really agent first. And humans play this role of steering governance and guidance. And there's this intersection of human-agent interaction that we're pioneering in the world of finance. There's this, you know, the ways agents interact in our platform that are governable and controllable that are critical that I don't see elsewhere in the industry.

[00:19:16] And it's exciting that we are at this cutting edge of technology. The teams are operating in this model. And I think for me, it feels like waking up every day with a fresh set of challenges, with new learnings and new insights into how technology is evolving. And I think it's exciting to be able to translate this to a domain with so much impact to businesses worldwide. And I think that's a powerful moment to end on. And for anyone listening that would like to talk about anything that we've covered today in a little bit more detail,

[00:19:43] or just keep up to speed with some of those announcements coming out of Blackline or connect with you and your team. Where would you like me to point, everyone? Stop by our website, blackline.com. You'll see we do have a lot of blog posts and different content. We are also evolving rapidly, and we're sharing that knowledge on blackline.com. Stop by our website. Obviously, our LinkedIn contains a host of other information as well. But definitely stop by the website and check out that for more.

[00:20:09] Well, we covered a lot today from what operationalizing AI investments will require from finance companies, how to extract ROI from existing AI investments by embedding AI into core finance functions. So many big takeaways there. I'll add links to both the website and the LinkedIn profile page as well, including your own LinkedIn. I will encourage people to keep that conversation going. But a pleasure as always. We really appreciate your time today. Thank you so much, Neil.

[00:20:39] I think the big takeaway from this conversation, certainly in the moment I'll be walking away and thinking about, is Jeremy's warning against creating artificial measurements for AI or OI. So finance leaders don't need another collection of impressive looking metrics. Metrics that nobody outside of the AI team can understand. What they need to know is whether the financial close is faster, transaction matching is more accurate,

[00:21:07] collections have improved, and employees complete their work with less manual effort. And I think Jeremy makes a convincing case for treating AI spending like any other allocation of capital. Whether money is assigned to employees, software models, or even tokens, the result should be measured against the outcome that the business needs.

[00:21:31] And I think this becomes especially important as AI agents begin performing that finance work. And businesses need records of the prompts, models, data, tools, and permissions involved. And I think Jeremy's food label comparison makes it surprisingly easy to understand. If we expect a packet of biscuits to list its ingredients and calories, it seems reasonable to expect an AI agent handling financial information.

[00:22:00] It should provide its own operating history too. So a big thank you to Jeremy for returning to the podcast. I'll include links to everything that he mentioned there. But over to you. Is your company measuring AI because it is AI? Or is it measuring whether the business is genuinely performing better? Let me know. TechTalksNetwork.com. You can find out how to contact me there, work with me, or browse through 4,000 interviews.

[00:22:30] But that is it for today. Thanks for listening as always. See you tomorrow. Bye for now. Bye for now.