Agentic Finance AI: Inside the agentic Finance AI and clustering analysis unveiled at Splash

Hello everyone and welcome to our Innovation Spotlight Series. Today's topic is Agentic Finance AI. Recently at OneStream's Global User Conference, we showed finance leaders what it takes to move from AI experimentation to real impact at scale.

And today we're bringing you those announcements. I'm Jeanette Kozier and I'm here with Drew Shea, our EVP of AI engineering. Hey Drew, thanks for joining us today. Hi there, looking forward to it. Before we dive in, just a quick disclaimer, that this presentation might include future roadmap items.

It's at OneStream's sole discretion to change these roadmap items at any particular point in time. Okay, now let's dive in. When we look at the topic of finance AI, the real question is how to elevate AI from experimentation to enterprise value.

Today, approximately 75% of CFOs are leading their organization's AI strategy, but far fewer have actually deployed it successfully at scale. What we see is a lot of experimentation, but not always clear outcomes.

Forward finance organizations focus on applying AI intentionally within workflows, not as standalone pilots. They define governance. They set clear KPIs and measure value. So AI becomes a real driver of business impact and not just experimentation.

So how do forward finance organizations successfully deploy AI at scale? Well, this is where OneStream comes in. OneStream's unified platform is where organizations use financial management for what organizations must do.

So it's for closing your books, running consolidations, doing your planning, your reporting, and your compliance. And then the operational analytics connects real-time operations with finance. And then, of course, we have purpose-built AI to essentially go and do forecasts, flag risks, and surface insights.

And it's important to note that this isn't a bundle. It's a purpose-built platform for modern finance. In our plug-and-play, click-to-configure no-code library lets users go and create their own dashboards, interfaces, menus, and more, which makes OneStream uniquely unified, infinitely extensible, and AI-powered.

Now, when we think about AI and what finance requires, not wants or needs, but actually requires to apply AI intentionally within these workflows, we think about trusted, secure, purpose-built, and contextual.

Not general-purpose AI, but purpose-built within our platform. So now I'm going to hand it over to Drew to get into OneStream's sensible AI portfolio. Drew, take it away. Awesome. Thanks for the great introduction.

And I'm incredibly excited to be here to be able to chat through some of the latest and greatest sensible AI innovations that we just recently unveiled at our Splash user conference. Now, before we go into all the detail, I want to first just anchor us in on the idea that OneStream has been building production -grade AI systems for nearly a decade now.

And we've learned a lot along that journey that allows us to actually deliver trusted, secure, and purpose-built AI that's specifically meant for the Office of Finance. And today we're going to take you on the extension of that journey.

We're going to go through four key chapters together. So number one, we're going to hit on our sensible AI studio portfolio of capabilities. We're then going to go talk about how sensible AI forecast is continuing to reshape the way you think about doing planning and forecasting.

Number three, we're going to introduce you to your digital finance-focused teammates, your sensible AI agents. And then last but not least, we're going to show you how the sensible AI agent capabilities can actually follow you into any AI product of your choice.

So with that being said, let's go ahead and take a look at sensible AI studio. And exactly one year ago, we announced the general availability of our sensible AI studio. And since then, it's allowed our partners, it's allowed our own development teams to build with a library of over 60-plus AI algorithms that can be captured and integrated into a wide variety of solutions that people already know and love inside of the OneStream ecosystem.

One notable example here is at the tail end of 2025, we introduced AI account reconciliations, which effectively takes anomaly detection capabilities and has infused it into the core workflow of our account reconciliation solution.

Really all focused on allowing the teams to reduce cycle time and mitigate risk between the teams that go from the preparers, the approvers, and the auditors. And with that, we're extending our journey with Sensible AI Studio with the most recent announcement of our Sensible AI Clustering Analysis Solution.

Now, when you think of Sensible AI Clustering Analysis, most organizations today, they either benchmark by region or by org chart. And that's why we rarely actually see benchmarking change behavior. You're never actually comparing apples to apples.

And so Sensible AI Clustering Analysis is meant to completely change that. And it does that by allowing, by leveraging AI techniques, which is called clustering, to ultimately uncover the intelligent, intelligently uncover the peer groups, you know, that, you know, accurately capture the operational and performance potential of your entities.

Whether your entities are stores or products or different services lines, right? You have that flexibility and there's, this product can be for any type of organization out there. So what does that actually unlock for you once you have those, that true apples to apples comparison? It unlocks benchmarking, right? This is all about identifying actual spending gaps and can really easily point to, hey, you know, you're actually overspending on SG&A by $3.4 million for this particular store.

Here's why, and this is what we can do about it. The second is drift analysis. And I want you to think of this as the early warning performance detection system. So you can actually catch entities that are drifting away from their peer group in the negative direction or even the positive direction before it shows up in the quarterly business review.

And then last but not least is peer group analytics. These are peer-aware scorecards that you get for every single entity in your organization. And so when you start the analysis process, you're actually starting it with the analysis.

You're not doing the data wrangling to just get an understanding of what your entities, how your entities are actually performing. So let's make this a little bit more real. Let's dive into a demonstration to showcase what this is all about.

Sensible AI clustering analysis. Let's assume I'm an FBA lead at a manufacturer and retailer of golf clubs and equipment. Our board just kicked off the 2026 planning cycle and has asked me to find another 100 basis points of operating margin without cutting growth investment.

Within one stream, every regional location gets benchmarked against its regional average. South against south, west against west. But here's the problem with that. The stores within a single region don't actually operate the same way.

Footprint size, distance to the nearest competitor, etc. are all over the map, even within the same region. Take these two stores for example. Both in our south region, one is 30,000 square feet with a competitor across the street.

The other is 6,000 square feet with no competitor for 12 miles. They have the same regional target, but those two stores are comparable on anything that drives costs. Now let's watch what happens when we cluster these stores by how they actually operate.

Three real peer groups emerge from the data stores and up ground with their operational twins. Not by where they are in a map, but by how they run. Every feature we use to cluster footprint, competitive isolation, density, product mix was chosen specifically because it drives performance.

Let's zoom in to one cluster, high density, large apparel stores. They are all operationally similar. But this dashed ring inside the cluster marks the top quartile performance in that peer group. How we define that performance matters.

Performance here is the blend of three commercial metrics. Revenue per square foot, revenue per visitor, and gross margin. Better operators will naturally rise to the top. In this store right here, bottom of the cluster on performance runs at 29%, almost double its peers.

Just by closing that gap by 50% means a $935,000 opportunity. If you repeat that process across every store and every cluster, we'll find approximately a $10 million saving opportunity in SG&A across the business.

Let me show you what this will look like inside one string. Let's start by looking at performance scores. This is how we tell clustering analysis what makes one store or one group better than another.

We can set our performance scores by just sliding up and down these sliders. SG&A is intentionally not in that definition. This is purely how well each store monetizes its footprint, converts its traffic, and holds its margin.

Now let's look at the learnings this enables inside a benchmarking workflow. We pre-built and ran this workflow configured with the accounts for our OneStream Cube and performance score we just looked at.

So here's my answer to the board. Closing just the gap to the third quartile of top performers, the most conservative opportunity is a potential savings of $6.4 million just on this one payroll account.

So our EDEO3 is the headline here. It's a dense market large apparel store, and it has a total opportunity identified of $2.6 million. And there is one thing that most benchmarking tools leave out. For every account where this store has a gap, the system has already drafted plain English guidance for how to close it.

For something like office supplies, it says that a vendor consolidation play that other high-endensity large apparel stores have already run could work for you. This means that my controller doesn't start with just a blank page.

They start with a draft that they can challenge or run with. For a specific store, a specific account, they get my specific recommended action. And a human will still own the call. The system doesn't post the entry.

It gives the team a starting point. So back to the board's question. Another 100 basis points of margin for 2026. From this single workflow against one cost area, I have millions of dollars that I can defend with action plans for every store and every account.

And this is just one of the three workflows available within clustering analysis. So there's three things to take away here. One, this analysis ran in minutes, not months. Two, it's finance-owned. That means no data science team and no custom build.

And three, every dollar and every opportunity ties back to a specific entity with a defensible action. From regional averages to operational truth benchmarks the decisions. That's smarter capital allocation.

All right. So hopefully that gives you some perspective into what sensible AI clustering analysis is all about. And keep in mind, we're doing this at production-grade enterprise scale. And sensible AI clustering is now generally available.

So that means you can walk away from this presentation and within six to eight weeks, have this capability ready for your next annual operating plan so that you know where you can find the actionable spending gaps without cutting growth investment.

So let's move on to our next chapter here of sensible AI forecast. And like I said, sensible AI forecast has been changing the way that organizations think about doing planning and forecasting. It doesn't matter what industry you're in or what planning process you're looking to revolutionize.

We see, on average, a 25% accuracy improvement over the traditional forecasting methodology at the organization. And we also see an 80-plus percent faster cycle, which ultimately means rather than planning, you know, maybe once a year, once a quarter, you can now actually start to plan once a month or once a week.

All while having the transparency and trust that you need to understand which drivers, which macroeconomic indicators, which sales promotions are actually moving the needle as part of your planning and forecasting process within each and every scenario that you decide to build.

So we're going to hear a little bit more about sensible AI forecast here in a minute. We've got a lot of innovation within that product, but I do want to move on to our third chapter, which is introducing you to your digital finance focus teammates.

And the last year, we introduced the three initial agents within our sensible AI agents portfolio through our private preview program. And since then, we've been able to iterate with a variety of different customers to pressure test the capabilities and rapidly innovate new capabilities into these agents.

So let's take a minute and talk about what these are all about. So when I say and when I introduce the sensible AI search agent here, I want you to think of it as a one -stream process expert in your pocket.

And it's an expert because it has read the 2,000 plus pages of one-stream platform documentation. It's read the books built or written by our world -class architects within the one-stream ecosystem. And the fact that it has that information, that means that the next time an end user goes to ask a question around, you know, what is my, how do I upload my latest trial balance as part of my upcoming workflow? So search is able to give me that answer in under 20 seconds, right? Or if I'm a power user that forgets how to write a working capital ratio in one-stream syntax, again, search is there with that answer.

So search is all about compounding the micro time savings as you interact with one-stream through the one-stream platform. And those time savings compound from, you know, minutes to hours, hours to days, days to weeks, and weeks to months.

Our second agent here, the finance analyst, is all about making it so that every single end user inside of your organization is as capable as the top 1% of power users across all of one-stream. So what does that really mean? It comes down to the fact that the finance analyst is an expert at your specific one-stream financial model.

And the fact that it understands that you decided to put region in UD1 and product in UD2 gives you the ultimate contextualization of our agent to your one-stream financial model. And this unlocks two major capabilities for you.

With the finance analyst, I get narrative-ready reports simply by asking questions to the finance analyst. And number two, with the latest release of our finance analyst, we've introduced what we call narrative analysis automation.

And analysis automation effectively allows you and your teams to deploy hundreds of finance analysts to solve variance analysis, flux analysis, trend analysis, any type of anomaly analysis that you want to do can all be triggered behind schedules that tie to your typical closed consolidation workflows and processes.

So that every single time that you're inside of one-stream, you start your work with an analysis and not a data-ranguling exercise. And then last but not least is our deep analysis agent. So deep analysis was built specifically for the organizations that are constantly going through, you know, vendor contracts, customer contracts, order schedules.

And you're looking for the repeatable clauses, the repeatable sets of information like different types of rebates or early termination clauses, right? Things that you need to be aware of that affects the financial forecast, the financial outlook of the business.

And so deep analysis will comprehensively read every contract word for word and extract exactly what you told it to. So you can focus, again, on doing the analysis and not the reading of hundreds of pages of contract documentation.

Now, to just put this in perspective, right, like I said, this has been out there in the market for a year. And we've had we were fortunate to have a fantastic customer up on stage with us at Splash to share just some of their insights and the validated ROI that they've been able to see and achieve by leveraging sensible AI agents.

And across our three major agents that we just talked about, we've seen some dramatic time savings, you know, per agent per question. Now, one thing I really want to call out here is the fact that our agent capabilities actually track a lot of this information for you.

That means that as you are leveraging sensible AI agents, you get to decide what you expect or what you are seeing your ROI impact to be. And the value model that sits behind sensible AI agents is constantly accumulating misinformation for you.

So you've got a narrative ready report on the value that you're achieving with sensible AI agents that you can report out to leadership. And you might have seen we had one card still left to turn, and that's because we've also announced our sensible AI forecast agent to the lineup of our sensible AI agents.

Now, the sensible AI forecast agent is effectively us layering an agentic surface over the entire product of sensible AI forecast. That means you get to interact in natural language with the deep and rich insights that are available, the scenarios, the different forecasts, the project configurations.

So it's just completely changing the way that end users and power users think about interacting with the product and getting insights that they need to drive their planning and forecasting process. So, again, these are our native agents.

You'll find these agents in the slide-out panel of the OneStream Windows application, the browser. And we're actually really excited to announce that they're finding their ways into other third-party agents.

Before we go there, though, I want to just take a step back and talk about what we've been seeing, I think all of us have been experiencing, around user experience. Now that we've worked with tools like ChatGPT, Claude, Copilot, any third-party agent.

And you don't have to go too far back just six months ago. This is the way that the world worked, right? I would go to four different dashboards. I might have to go to three separate systems to get the insight that I need to inform a business decision at the organization.

Well, with the advent, like I said, of these third-party agents, ChatGPT, Claude, and Copilot, as an example, right, my expectation and interaction with software is fundamentally changing, right? I expect to be able to see an answer or a result in the way that I want to see it, in the language that I speak, and I want to see it in under 30 seconds, right? And fundamentally, that's causing a user experience gap, right, with typical software systems.

And so one of the ways you can go about solving that, right, is you might put a third-party agent between yourself and the data, right? And there are tons of use cases that are unlocked just by doing this, right, that can give yourself, your team, your organization fantastic efficiency gains.

But within the Office of Finance, it's mission critical that we always know that we're interacting with the right data, the right workflows to, you know, to ultimately make business decisions within the Office of Finance.

And doing just that process leads to, you know, might solve part of the user experience gap, but the governance gap emerges, right? And the questions you've got to be able to ask yourself are, well, how do I know that a third-party agent is going to be working with the, you know, source of truth information, source of truth financial information? Or how do I know that as an agent starts doing financial calculations that it's going to abide by my intercompany relationships or the third-party? Or you think of if you're constantly into mergers and acquisitions, how do I know that it's going to abide by the changing ownership structures within my organization, right? All these little detailed nuances really matter to the, you know, as you start to do financial analysis, financial processes inside of the Office of Finance.

So that is exactly why we built what we call the OneStream agentic layer. This effectively is about us decomposing our agent capabilities, our native agent capabilities, into tools that other third-party agents can interact with, so that we can extend the financial intelligence of OneStream outside the four walls of OneStream.

And ultimately, what does that mean for you and your organization? That means that you get to have agents your way, right? The OneStream agents can follow you, the capabilities, the financial governed source of truth can follow you and your teams to Excel, to ChatGPT, to CoPilot, to Google Gemini, any third-party agent.

We have the capability for our agents and the OneStream financial intelligence to interact with. So that is the power of the OneStream finance agentic layer. Now, like I said, there's a lot of different agents that we have the ability to connect and interact with, but one I want to highlight more notably is Microsoft CoPilot, right? And we've got a fantastic partnership with Microsoft.

We've got, you know, as part of Splash, we announced the latest version of our Office 365, Excel 365 add-in. We introduced, you know, deep and native integration of our agents directly into Excel. And we also talked about and showcased how our agents can interact seamlessly through the variety of form factors of CoPilot.

And so, as you think about all that, I actually want to take a moment to be able to share and showcase what this ultimately looks like, both from what our power of native agents work, as well as what your experience looks like as you interact with the OneStream agentic layer through third-party products.

Now, let's hop in and take a look at the OneStream agentic layer in action. For the sake of this demo, we're going to be focused on our OneStream KubeData MCP. These new tools now allow third-party agents like ChatGPT, CoPilot, Gemini, and Claude to both search and execute existing OneStream KubeData reports, as well as have the ability to pull any slice of OneStream KubeData.

Let's hop into CoPilot and take a look. We can now ask CoPilot, Can you pull my income statement actuals versus budget for North America equipment? Now, CoPilot, now having access to those OneStream tools, is first able to search across all of the relevant reports defined within our OneStream application.

Along the way, resolving any of the necessary inputs, like North America equipment mapping to our specific entity, and finally pulling the data directly into context. Along the way, each of these individual data pulls are flowing through our OneStream agentic layer, ensuring they follow your existing security and access models.

Now that the data is pulled in, CoPilot can display it directly to us within the chat, where we can then continue to ask questions or drill in further as needed. Let's now hop over to Claude to take a look of this in another third-party agent.

Here we can see we pulled the same income statement report and now want to drill into this large negative compensation variance. We can now follow up with Claude asking, Let's drill into our compensation variance to determine which cost center is driving it.

Please include a chart along with the data. Cool. And now similar to what we saw with CoPilot, Claude is going to start reasoning over which tools to leverage for this question. In this case, choosing to use our dynamic query and capabilities.

What you'll notice here is the first tool we always leverage is this get application context tool. This is an essential part of the process as it ensures we bring in all of your relevant OneStream specific application context, like your relevant cubes, your specific fiscal year offset, which we can then use to help us search across your dimensionality, traverse your metadata hierarchies, and finally execute this targeted slice of cube data.

And as it's finishing up here, you'll see the agent pull that data directly into context, using it to generate a highly visual and interactive result here. And as you can see, our operations cost center was the largest driver.

And as we scroll down here, you can see some detailed explanation as well. This should start to show you just what these third party agents are capable of when powered by our OneStream agentic layer. Now the ability to search and analyze your OneStream financials is not just limited to third party agents.

We have these same underlying tools and context now directly within the OneStream platform. However, we've now taken things a step further. With the next generation of finance analysts, businesses are now able to automate hundreds of recurring analysis flows to execute in the background on your behalf.

All with a focus on auditability and transparency. Let's now click in and take a look at this income statement variance analysis plan that we have created. As we drop in, we land on our planning agent where users work iteratively to generate a highly accurate and detailed analysis flow.

As we scroll through on the right here, you can see all the key things this contains. The high level objectives, the assumptions that we have. Scrolling down into the specific investigation approach, dimensions we want to drill into.

Again, a very, very detailed and thorough analysis plan, capturing everything all the way down to the output requirements. And from here, users have the ability to then set these plans to run on schedules.

Let's click in and see the one that we set for this specific plan. Here you can see we've now set this plan to run on the first of each month at 8am. From here, we can click in to take a look at the generated report.

At the top of the report here, you're going to see a summary of the core underlying drivers that were surfaced for this specific analysis. As we scroll down, you'll see the process the agent went through, starting with the high level income statement.

As we drill down depending on where the variances are here, going into some account level findings and then further down going into an operating expense drill down. Along the way, not only including high level tables of the data, but also some core commentary to support as well.

And at any point, we can click into any of these cited sources to see exactly where this data was coming from within OneStream. This ensures users have that auditability and transparency into where their numbers are coming from at all times.

Now this should start to show you how OneStream is now allowing your finance teams to leverage agents your way. All right, so hopefully that shows you the power of what the agentic layer and how OneStream's financial intelligence can meet you for whatever agent that you ultimately use.

And also how you can leverage OneStream's native agent capabilities to transform the way that you think about doing work inside of OneStream. Now, with that being said, I'm going to turn it back over to Jeanette to take us, you know, take us through some of the closing remarks.

Thank you. Thanks so much, Drew. Wow, that's a lot of information in 30 minutes. So that was definitely action packed. It's very exciting to see how OneStream's agentic layer can give finance organizations that's secure, that trusted, that governed in contextual AI.

So thank you so much for going through that. At this point, why don't we take a couple of questions? First and foremost, people want to know how do you recommend they get started with sensible AI agents in the OneStream agentic layer? Yeah, I think there's, I think I'll kind of hit this as a two part question.

If you're a prospect, and you're considering, you know, starting your OneStream journey, I would make sure to be starting with the OneStream AI agent capabilities, right? Why do I say that is because as you build out your core OneStream financial model, you're going to want your teams having the sensible AI search capability to help them get enabled on how to leverage OneStream.

Right? And then as you think about the, your OneStream financial model being built out, you're going to want the sensible AI agent finance analysts to be able to sit on top of that model immediately to give them the natural language interaction with all the insights that you've taken the time to curate within your OneStream financial model.

And if I hit this from the angle of your, if you're already a customer, there's no time like the present to get started, because it's really easy to overlay the, you know, the finance analyst on top of your OneStream financial model.

It's just a couple of weeks implementation, and it's really focused on just having a conversation with your teams to ultimately understand how do they speak, you know, through natural language when they're talking about a financial model.

We want to be able to capture that information and overlay it on top of your OneStream financial model. So, finance analyst speaks the language that your teams speak. That's really what the implementation is all about.

Got it. Okay, thank you for that. Second question is, how should organizations think about a rollover strategy for sensible AI agents? And then how should they look to manage usage? Yeah, I'm sure, you know, one of the big topics right now is, you know, AI token and token spend, we wanted to make sure that the AI capabilities of OneStream are highly auditable and highly governed.

Right? So that means, you know, we've seen customers take a wide variety of different rollout strategies for, you know, for sensible AI agents, where, you know, I'd say one of the leading strategies is start with a small, you know, power user centric team.

And then, you know, we've seen a lot of confidence in the ability to understand how are we actually leveraging our agents and how we want to effectively roll this out to a wide variety of different user groups over time.

Excellent. Yeah, you touched upon something that you, you know, we've both talked about this highly governed. There's one quote that our CEO likes to say, which I find is completely relevant for the finance organization, is that your agents won't go to jail, you will.

So definitely getting a platform that has purpose built AI that's governed and, you know, trusted and secure and all of that. So thank you, everyone, for taking the time today to listen in on how OneStream can help you on your AI journey.

So if you'd like to experience more of what was discussed at Splash 2026, you can take a look in your docs tab here on this webinar. And essentially, you'll see a link for the Splash 2026 experience site.

And then secondly, if you'd like to learn more about the Forward Finance Blueprint, also check out the docs tab and you'll see a link there as well. So we thank you for taking the time today. We appreciate it and wish you great luck on your AI journey.

And if you have any questions, just feel free to reach out to us. Thanks much. Have a great day. Thank you.