Finance AI in practice: From forecasting to faster, more confident decisions

So I see there are still a few people joining the webinar, so let's give them a few more seconds. All right. Thanks, everyone, for joining today. My name is Jasper Gartman, and here at OneStream, I am responsible for AI sales and go-to-market, and I'm based in Stockholm, Sweden. With me, I also have my colleague, Peter Gilfalen. So, Peter, do you want to go ahead and introduce yourself before we get into the content?

Thanks, Jasper. I'm Peter Gilfalen. I'm a senior solutions consultant here at OneStream. I've got 15 years corporate finance experience, and I am based in Austin, Texas. Look forward to showing you all the AI capabilities that we have today.

All right. Thanks, Peter. So today, we're going to talk about how finance teams can move faster, make better decisions, and do it with more confidence in a volatile environment. More specifically, we're going to talk about how AI is starting to play a very real and practical role in that endeavor. And to understand why this matters, let's start with what has changed around finance.

And even before you get into that, there's a few housekeeping items that I wanted to get over. So we'll take questions at the end of the webinar. If we don't get to your question, please drop it in the chat, and our team will follow up after the session. Now,

finance is operating in a very different environment than it was a few years ago. There's constant volatility where macroeconomically and geopolitically, the world is changing fast, definitely outpacing the quarterly planning cycles that we're used to.

The expectation is also that finance needs to deliver real-time answers, not just month-end reports. With inflation changes, raw material pricing changes, we need to be able to produce scenario models very quickly, which is sometimes hard. So teams are stretched. There's less time. There's more responsibility and higher stakes. And that pressure is exposing a larger problem where traditional finance, in terms of the operating model, was not built for this speed. Closing always takes priority over producing insights because we need to legally do it.

Static forecasts often miss real-time signals. And when we do these forecasts manually, we can't really keep up with the pace of change. Our manual models can't account for all of the things in the world around us.

So based on this reality, nearly half of CFOs, I think 49% in a recent study conducted by Deloitte, said that automation is a top priority. Moreover, 46% of CFOs, according to PwC, are focused on improving forecasting accuracy. So the direction is clear, but the reality is these critical decisions are still too slow and forecasts are still too fragile. And that is exactly why AI has become such an important focus.

So as CFOs seek to change the way their teams work, AI investment is growing quickly. The issue is business value is not keeping pace. Nearly half of AI proof of concepts do not make it to production. Most AI projects fail to deliver measurable ROI. And generative AI pilots often fail to create measurable business impact. So the issue is not ambition. The issue is more around execution.

I don't think it's a model problem. It's actually more of a system problem. So the question becomes, why do so many AI projects break down? Now, what the research shows is that AI rarely fails because the algorithm doesn't work. It fails because it is disconnected from how the business actually runs. So first, data quality and integration issues. If data is siloed, incomplete or low quality, model accuracy often suffers. Second, governance and trust gaps.

So if outputs are not explainable, if you can't audit them, if they're not controlled, finance will not adopt them. Third, technical complexity and scalability. While models may be easy to build in isolation, whether it be for the purpose of forecasting or anomaly detection, unless they are able to scale across the enterprise, they won't work.

The fourth item is around workflow fragmentation. When these AI projects sit outside of finance and IT, business users cannot easily access or operationalize the insights. They need to be embedded in the systems where finance work. So the result of these challenges is that smart models are created, but they never actually change how finance works. And this is exactly the gap that our sensible AI solutions were meant to close. So we believe that for AI to work in finance, it has to meet a higher standard.

It must be explainable so that finance can trust every number. It must be governed. So audit and compliance doesn't slow down adoption. It must also be embedded so that teams actually use it in their day-to-day workflows.

And fourth, it must be scalable. So it works across the enterprise, not just in pilots. And that is the standard that finance should expect from AI. It should accelerate decisions instantly, not introduce new risk. And that is the foundation of our approach with sensible AI. So the approach centers around four themes. First, it runs on unified, governed, and trusted data. There's not a lot of data moving around. There's no duplicate environments and no delay. Second,

governance and explainability are built into the solutions. So every output is explainable, traceable, and auditable inside of the platform. Third, it is designed for enterprise-scale compute, meaning this is not about running a single model in isolation. It's about running AI across large volumes of data across the business. And that's what allows organizations to move beyond pilots and actually deploy AI in production at scale.

And most importantly, it is embedded directly in your finance workflows. That means no separate tool, no workflow disruption. And this is what drives adoption. This is why sensible AI is helping organizations move from experiments to real adoption. So let's take a look at the solutions that we bring to our customers. Now, sensible AI supports the full journey from forecast to insight to decision.

So first is our forecasting solution, which is based on time series machine learning, allowing our customers to produce accurate forecasts with clear explanations behind every prediction and the ability to run multiple different scenarios.

Second is Studio. This is a low-code AI workbench that lets finance teams infuse intelligence into more workflows across OneStream. And it leverages capabilities like anomaly detection, regression, natural language to SQL, clustering, and more.

Third is agents. You can think of this as digital assistants that understand OneStream and your organization's unique financial model, which allows them to perform tasks like reporting and analysis. And together, these capabilities help organizations create smarter predictions, faster insights, and scaled capacity.

And that value is already showing up in measurable ways. So with forecast, customers are improving accuracy by 24% on average across the 100 subprojects that we've already completed. And they're reducing the forecasting effort by around 86%. With Studio, customers are accelerating their analysis workflows by 80%. And with agents, the same teams are reclaiming meaningful capacity by offloading tasks to these digital assistants. So this is not just about adding AI features.

It is about helping finance teams improve prediction quality, move faster from data to insight, and scale what the team can accomplish.

So let's bring this to life. In the next section, we're going to walk through how this works in practice. So you'll see Peter demo how finance teams can move from forecast to insight to decision in a single connected workflow. And as you watch this demo, I'd like you to pay attention to three things. How the experience stays inside finance workflows, and how quickly insight becomes actionable, and how the process stays explainable and governed from end to end. With that, let's take a look, and I'll hand it over to you, Peter. Thanks, Jesper.

Let me bring up my screen here. Hello, everyone, and welcome. Today, we're going to walk through how OneStream's sensible AI brings practical, explainable AI directly into finance workflows. What you'll see is not just AI for the sake of AI, but AI applied where it actually drives value. We're going to be improving forecast accuracy. We're going to be increasing transparency and trust. We're going to be accelerating analysis and decision making.

We're going to cover off on three main areas. Our sensible AI forecast, how we generate and explain forecasts. Our sensible AI agents, how AI assists finance users in real workflows. And then we're going to finish off with our sensible AI studio, how we extend AI into deeper analysis like clustering and operational data chat. But let's get started with forecast.

Looking at my screen here, we're starting on a revenue forecasting screen. At a high level, this represents the full forecasting process. It's got historical actuals, it's got driver-based inputs, and it's got forecast outputs.

One important point, though, before we dive a little bit deeper, is that we don't believe AI should be used for everything. Finance teams already have strong deterministic models and business rules. Sensible AI is designed to augment, not replace those processes. So focusing on areas where patterns, variability, and uncertainty benefit from machine learning. In this case, we're using AI specifically to forecast units, which we then roll into revenue. So here, let's drill into our forecasted units.

From here, we can see that sensible AI is starting to differentiate itself. It's not just generating a number, but explaining why that number exists. We're moving from what is the forecast to what is driving the forecast. Now, let's take a look at the model arena.

This is where the modeling process happens. Instead of relying on a single model, sensible AI trains multiple models, evaluates their performance, and then it selects the best fit model for each individual target based on the data. This ensures higher accuracy, reduced bias, better adaptability across different data sets, and importantly, this process is fully transparent to the user. Now, we can look at model confidence scores. This answers a critical finance question. How much should I trust this forecast?

Confidence scores provide a quantitative measure of reliability, visibility into model performance over time, and also early warning signs when forecasts may be less stable. This is key for finance teams that need audibility and defensibility. Now, let's move into feature insights.

This gives a holistic view of our relationships and which drivers are influencing the forecast, how strong those relationships are, how they interact with each other. And here's where we enhance this further using generative AI. We can then use an LLM to translate these relationships into a natural language explanation.

For example, the forecast is primarily driven by seasonal demand patterns and pricing changes with promotional activity contributing moderate uplift. This bridges the gap between data science outputs and finance user understanding. Finally,

we can look at our tug of war chart. This is one of the most powerful visualizations because it shows what is pushing the forecast up versus down at a specific point in time. For a selected point in time, we can see the positive contributors, you know, the demand increasing pricing or negative contributors, market slowdown or supply constraints. It's called tug of war because each feature is effectively pulling the forecast in a different direction. This allows finance teams to explain variances, communicate drivers to stakeholders, and just make more informed decisions.

So instead of just saying revenue is up 8%, we can say revenue is up 8% driven primarily by volume growth, partially offset by pricing pressure.

So that's forecasting. Accurate, transparent, explainable. Now let's shift gears and look at how AI can actively assist users in their day-to-day work.

Here we're introducing the finance analyst agent. This is not just a chat bot. It's an AI agent that understands finance context. It's also going to be able to execute tasks and provide traceability reasoning. Think of it as a digital finance assistant. Let's give the agent a task here.

I'm going to ask it to show me a full P&L for March and Q1 comparing budget versus actuals. The agent is going to interpret the request, select the appropriate dimensions, time, scenario, accounts, etc. And it's going to construct the report. What's important here is the agent is not guessing. It's following structured logic within one stream.

Now let's look at the results the agent produces. We get a full P&L view for March, Q1, and budget versus actuals.

But what really matters here is how this is done. First, the agent inherits core security. It only accesses and returns data that the user is authorized to see. Second, we can validate the numbers just like any other one stream workflow.

We can review the point of view, right, that's going to have our time, scenario, and entity. We're also going to be able to drill down into any of the data to trace back values to their source. This is going to ensure full transparency and auditability. Third, and this is critical, the agent does not attempt to calculate variances itself. Instead, it leverages core one stream calculations.

As you'll see down here, where we have the actual variance logic, the account formulas, and the existing financial rules. So rather than recreating the logic, the agent is calling the trusted financial engine that already exists. This avoids calculation inconsistencies, logic duplication, a risk of incorrect results, and what we get is the best of both worlds. We get that AI driven speed and automation, but also with enterprise grade accuracy and governance.

Now let's talk about our Sensible AI Studio. In Sensible AI Studio, we can build and apply advanced models and AI experiences. Two examples we're going to highlight here are clustering for pattern discovery and also our operational data chat for conversational access to large relational data sets. This is where AI expands beyond forecasting into broader analytics and exploration.

Looking at clustering here, we're going to help group similar entities or regions together. We're going to identify patterns in data. We're going to segment the business in new ways. For example, we can look at customers, products, and regions. Here we're looking at the clustering output.

Each cluster represents a group of regions with similar characteristics. Instead of predefined segments, these are data-driven groupings. This allows us to uncover hidden trends, behavioral similarities, and outliers. And we can connect us back to finance because we can evaluate each cluster on revenue contribution, margin performance, and growth trends. This helps answer which regions are driving profitability or where should we focus our attention. So clustering turns raw data into actionable financial insights. Finally, let's look at our operational data chat.

This allows users to interact with the transactional data using natural language. For example, we can ask things like, which regions had the highest transactional costs?

The system's going to query the relevant data, interpret the relationships, and return this clear contextual answer. This dramatically lowers the barrier to analysis. Instead of going out and having to build reports or navigating across multiple screens, users can simply ask questions and get answers. So it becomes a unified conversational layer across all of your data.

To wrap up today, we've seen how sensible AI forecast delivers value across three areas. With accurate, explainable predictions with our forecast. It's going to have AI-assisted workflows with governance through our AI agents. And then it's going to provide advanced analytics and conversational data access with our studio. And the common theme across all of this is really just that practical, transparent AI design for finance. So

we are going to wrap up today. And I thank you for your time. And we look forward to your questions.

Thanks, Peter. I'll pick it back up. So what you just saw in the demo is not theoretical. Organizations are actually already using these capabilities in production and seeing measurable results.

As an example around forecasting, we have a Fortune 500 logistics company that is using us to forecast items across their P&L.

Their challenge coming to us was that their revenue and margin forecasting were inconsistent, both across facilities and customers. So they were trying to do forecasts at a very operational level, but saw a lot of error. And they spent thousands of hours putting together the forecast. Now, with AI, we were able to give them a 98% forecasting accuracy on revenue and 87% accuracy on contribution margin. And we cumulatively reduced the error by 51%, saving around 13,000 hours annually on the process of forecast creation.

So that's one example. Further, when it comes to studio, we work with a dental services company in the US who leverage the AI studio for benchmarking across their practices.

Their challenge coming to us was with hundreds of locations, benchmarking performance across various operational and financial KPIs was very hard to spot underperforming practices. They had to go into a lot of manual analysis.

So the outcome we gave them using the clustering capability that Peter just demoed was 80% faster insights, accelerating that corrective action across the network to drive margin expansion. Further, on agents, we work with a Fortune 500 global conglomerate that are leveraging these digital assistants for instant FP&A answers. So the answer is, they're going to go into a lot of the way to do that. They're going to go into a lot of manual ad hoc reporting and variance explanation requests that slowed down decision making and created a heavy workload for their FP&A team.

With these digital assistants where there's this self-service capability to generate ad hoc reports in natural language, they're saving around 19,000 hours annually.

So the next question is, how can you get started in a way that is practical and focused and get some of this value yourself?

Now, here on stream, we have a team of dedicated experts who can help our customers get started on their AI journey.

And the question is not whether AI will change the function of finance. It's who will lead that change in your organization? And the opportunity is to forecast a lot more accurately, continuously, get a more scaled intelligence across more workflows, and empower lots of finance leaders with faster answers. And to make sure that you can reap those benefits, we're offering a sensible AI roadmap session where we facilitate a practical conversation with our team of AI experts, where we focus on your specific use cases and priorities, and we devise a plan to unlock value across your finance team.

And what I will say is that when organizations think about taking that next step, there's usually a few common concerns that we'd like to address.

And we hear these a lot. I think a lot of them are valid. And I'll take you through how we think about them, how you can think about them. So the first common concern that we hear is our data isn't clean enough.

The reality is you don't need all of your data to be clean to start getting value out of AI. You start with the data that is required for a high-impact use case, whether that's a particular set of items that you forecast, whether it's a particular set of financials underlying reports that you want to create. And if your data supports that reporting, it's typically enough to begin forecasting with AI.

Second, we don't have the team or the bandwidth. That's a common objection I hear. Now, the reality is that these solutions are designed to deploy quickly. They're pre-built and configured around your use cases and built for finance. So we've already completed a lot of the hard work for you before the project even starts. And you do not need a large data science team to execute on this. This is built for finance users, leveraging the expertise your team already has.

Moving on to the third type of objection. A lot of organizations will think, well, we already have tools like Copilot. What do we need this for? So Copilot is a strong productivity tool for content creation.

Sensible AI is tackling a different set of use cases. It's built to drive financial decisions, leveraging all of the financial and operational context that sits in your OneStream environment. Consolidation hierarchies, scenarios, multiple types of dimensions.

A lot of context is needed to have these digital assistants be effective. And that's exactly what we built. So the key is to start with a high impact use case. Start with a problem where better predictions, faster insight or automation creates clear value. From there, the data requirements, the implementation timeline and the adoption plan will follow. And that brings us to the main takeaway.

If you want to continue the conversation with us, we urge you to join us at the OneStream World Tour, where you will get the ability to meet our executives, a lot of existing OneStream customers who have already started exploring and getting value out of these capabilities. So if you scan the QR code below, you will find a city near you where we'll be hosting this world tour. And with that, I want to thank you for your time today. We will follow up on the questions and wish you a good rest of your day. Thanks.