Why Unified Systems are More Important Than Ever in the Era of AI with OneStream CEO Scott Leshinski
In this episode of Future Finance, Paul Barnhurst and Glenn Hopper sit down with Scott Leshinski, President of OneStream Software, to discuss OneStream's recent acquisition by Hg Capital and the exciting changes it brings to the company’s approach to finance technology.
Scott Leshinski is the President of OneStream Software, where he leads OneStream’s growth and innovation. With a background in corporate finance at GE Capital, co-founding Bluestone International, and working at Huron Consulting Group, Scott has a wealth of experience in driving major tech transformations. He joined OneStream in 2021 and was promoted to President in 2023 after helping the company achieve impressive growth.
In this episode, you will discover:
What OneStream's acquisition by Hg Capital means for its future
The benefits of using a unified platform in finance
How automation is improving financial processes
The evolving role of automation in financial planning and decision-making
How OneStream’s tools are helping businesses become more agile and efficient
Scott Leshinski highlights the exciting future of finance, where unified platforms and automation are transforming how businesses manage financial workflows. He emphasizes the growing role of technology in making finance processes more efficient and discusses how OneStream is helping companies become more agile.
Follow Scott:
Website: https://www.onestream.com/
LinkedIn: https://www.linkedin.com/in/scott-leshinski-bb5736b/
Follow Glenn:
LinkedIn: https://www.linkedin.com/in/gbhopperiii
Follow Paul:
LinkedIn: https://www.linkedin.com/in/thefpandaguy
Follow QFlow.AI:
Website - https://bit.ly/4i1Ekjg
Future Finance is sponsored by QFlow.ai, the strategic finance platform solving the toughest part of planning and analysis: B2B revenue. Align sales, marketing, and finance, speed up decision-making, and lock in accountability with QFlow.ai. Stay tuned for a deeper understanding of how AI is shaping the future of finance and what it means for businesses and individuals alike.
In Today’s Episode:
[00:00] – Intro
[03:00] – OneStream's Acquisition Impact
[05:00] – Transition from Public to Private
[06:00] – Scott's Career Path
[10:00] – What OneStream Does
[14:00] – The Power of Unified Platforms
[18:00] – Automation in Financial Planning
[19:00] – The Rise of Automation
[26:00] – Building Trust in Automation
[29:00] – Improving Forecast Accuracy
[35:00] – The Future of Finance
[50:00] – Fun Questions with Scott
[55:00] – Closing Thoughts
Full Show Transcript:
Host: Paul Barnhurst (00:00):
Welcome to the Future Finance Show where we talk about treasury management on our future. Finance is brought to you by Q flow.ai, the strategic finance platform, solving the toughest part of planning and analysis, B2B revenue, align sales, marketing and finance seamlessly speed up decision-making, and lock in accountability with Q flow.ai. Welcome to another episode of Future Finance. I'm your host, Paul Barnhurst, and this week I'm joined by my trustee. Did you call yourself Stinky or was it Moose Glenn? That's
Co-Host: Glenn Hopper (00:52):
For our morning radio show. That's You weren't supposed to say that on air.
Host: Paul Barnhurst (00:55):
Oh, that's right. Well, so my co-host Glenn is here with me. How are you doing, Glenn?
Co-Host: Glenn Hopper (00:59):
I'm doing good. I'm excited about today's show.
Host: Paul Barnhurst (01:01):
Do you want to do the honours and introduce our guests? We'll just jump right into things.
Co-Host: Glenn Hopper (01:05):
Absolutely. Scott, welcome. I'm going to introduce you to our audience here and we will dive right in. Our guest today is Scott Leshinski, president of OneStream Software. For anyone who follows the enterprise finance space at all, you probably have seen the news recently that OneStream just completed its acquisition by Hg Capital at a $6.4 billion valuation, now taking the company private after about 18 months on Nasdaq. So, Scott's joining us at a pretty pivotal moment. His path to the president's chair has been a pretty interesting one. He started in corporate finance at GE Capital, then co-founded a consulting firm called Bluestone International. Spent nearly a decade as a managing director at Huron Consulting Group leading finance technology transformations, one of Paul and my favourite topics. And then he joined OneStream in 2021 before being promoted to president in January. He was running OneStream's AI business unit where he drove 60% year-over-year growth in AI bookings. So, he's not someone who just talks about AI and finance, he's been building it and selling it. And today, we're going to dig into what AI actually looks like when it's embedded inside an EPM platform, how finance teams are learning to trust it, and what all this means for the future of finance. Scott, welcome to the show.
Guest: Scott Leshinski (02:14):
Glenn and Paul, it's great to be here with you both and looking forward to the discussion today.
Co-Host: Glenn Hopper (02:18):
I think the timing, it's got to be, it's interesting for us outside as an observer, but for you guys, I imagine it's quite a time the HT app capital acquisition just closed, taking the company private, like I said, after 18 months as a publicly traded company, what is going private at this point? What does it mean for r and d investment in the product roadmap? How does that change what you're doing as a public company?
Guest: Scott Leshinski (02:41):
Yeah, and to your point, it's a really exciting time right now in the journey for OneStream. We obviously have been on a high growth trajectory really since the inception of the company. It was a great opportunity when we took the company public and as you said, had the opportunity to operate in the public markets for effectively seven quarters. But now as we start this next chapter, and particularly in this era of ai, it's an incredible opportunity. First of all, if you look at how we perform to date, we've had strong performance going into this next chapter of the company. And really we look at the new partnership now with Hg as an opportunity to really continue to focus on our AI first strategy as a company. Instead of having to have the adherence to a 90 day calendar quarter, which we did in the public arena, this is now the opportunity to really focus on more of a long-term view on our r and d strategy and do so without the constraints of quarterly earnings pressures. Now obviously Glenn will continue to be disciplined in how we operate the business, but this really allows us to accelerate on the investments in areas that'll be most impactful for our customers. Again, that gets into the AI first strategy that gets into our product roadmap and also allows us to continue to invest in our partner ecosystem. So all of which should be really to the benefit of our customers.
Co-Host: Glenn Hopper (03:55):
I'm biting my tongue right now to keep from going down the OpenAI, going public philanthropic and all that. It's in recent news you've seen, I did see something that insider, some analysts said, no way OpenAI could even consider going public for another year because if they don't have the controls in place and all that. And I think sometimes companies, I mean obviously with the capital concerns that they have, I understand why they're doing it, but sometimes companies put that level of reporting requirement on a little early, and I know you guys are obviously more mature like that and you're not under the massive spend pressure that OpenAI is, but being able to not have to live and dive by the queues and being able to step back and take that more longer term strategy that has to be a significant, it gives you more ability to think long-term, not just quarter to quarter,
Guest: Scott Leshinski (04:39):
No question about it. And particularly right now when you look at, I think in my nearly 30 years of working professionally, I don't know that I've ever seen a pace as quick or as rapidly evolving as the current markets that we're operating in. And so for us, again, this is just an incredible opportunity for us to really double down on the strategy that we've known to be true that hg, we have a high degree of alignment with our new business partner in hg. And so for us it's really an opportunity to drive further acceleration on that strategy and deliver even more and more value for our broader stakeholders. So that gets into again, our customers, our partners, our employees, and the Hg team as well. So it's a great opportunity and really starts what we believe is going to be the best chapter yet written for OneStream.
Host: Paul Barnhurst (05:26):
I will say Hg Capital, I've had the opportunity to talk to them, some of their analysts every year we talk about the market. I've worked with several of their portfolio companies as anyone knows publicly, they owe several in this space beyond just OneStream. So I've always been impressed with their philosophy here and enjoyed working with them. So I think you're getting a really good partner that probably knows the space as well and as deep as any of the private equity, I mean you just have to look at their portfolio and it's pretty obvious that they big stake in this space and they know what they're doing.
Guest: Scott Leshinski (05:59):
Would strongly agree with that. Paul,
Host: Paul Barnhurst (06:00):
You got a good partner there. So I'd love for you to take us through your career arc, right? GE Capital, the founding your own company to consulting and now making the jump to a software company, kind of what pulled you to OneStream. And maybe just take us a little bit through that career. It's always interesting to hear how that journey goes because I'm going to guess 30 years ago when you started or 25 or whenever it was, this wasn't the path you envisioned if I had to guess.
Guest: Scott Leshinski (06:24):
So it's interesting, and again, each chapter has served a role or has had influence into the direction into where we are today. So I started off as you said in corporate finance. So started off working in fp and a at GE Capital for a number of years, and that really served as my introduction to the world of technology because at that time I had the privilege to be brought on board a team that was really running a finance transformation programme at ge. And what's interesting there is at that time we were going through the intrepidation on do we transition off our Excel spreadsheets into this big world of enterprise software? And we sit here and we laugh at that now, but I would argue there's a similar intrepidation right now when we think about how we drive enterprise adoption of AI in the ecosystem. But I'm going to park that for a second.
(07:16):
So to your point, I started off really in the office of finance and that helped to build a deep appreciation of the job to be done for who is now ultimately one stream's customer, which is really as we serve as the AI powered operating system for the Office of Finance. I then left GE had the privilege to serve as one of the two co-founders of really a technology consultancy. And we figured that if the capabilities that we're bringing to GE were delivering value there, there's a bigger opportunity. Then served there for a number of years that business was ultimately acquired then by the Huron Consulting Group and then spent nearly eight years at Huron helping to build out their bigger technology consulting practise there. And part of that, Paul, really that experience afforded me the opportunity to get a broad range of exposure, not only in terms of the transformation that we were delivering for customers, but also across a broad continuum of different technology platforms.
(08:16):
And so really where that started to form my view on really the broader technology ecosystem is a lot of the work that we were doing, it fell under the umbrella of really finance transformation was being delivered by a lot of point applications and point solutions. And I felt like while we were delivering that initial incremental value for the customer, the overall job to be done left the customer with a high level of technical debt. And so I started to look at it and really there were three things. Number one is again, the recognition on the capabilities and the solutions that we were delivering for our customers was again being done with point solutions. So that being number one around technical debt and looking for ways to better deliver that. Number two, when we start talking about really a platform in the market, it was also the recognition on there are also a lot of bespoke processes that would be better served by being brought into or on top of a unifying platform for the customer.
(09:20):
And then number three was the recognition of machine learning. And even more broadly, artificial intelligence was starting to pick up more and more focus by customers. So I started to have the reflection then on where is the place that is really going to help really deliver on the objective on the job to be done for the customer on a common unifying platform and bring these modern capabilities that the customer has come to expect with artificial intelligence. And that obviously is one stream where at one stream, if you look at us as an organisation, our vision again is really serving as the AI powered operating system for the office of finance. And really the mission as we serve our customers is to really empower every finance organisation and every finance end user to exceed their potential on what it is that they can deliver. And so OneStream really is the perfect place to be in this area.
Co-Host: Glenn Hopper (10:23):
Actually, as you're talking, I realise we're all very familiar with OneStream, but a lot of our listeners maybe in the SMB space or aren't familiar with really the whole EPM model. Can you give us the, because as you're talking about this, it sounds in a lot of ways like the holy grail of yes, that's what I want, but a lot of our listeners may not have ever encountered an EPM before. So if you can tell us a little bit about OneStream, kind of the elevator pitch, two minute version of it so that our listeners understand.
Guest: Scott Leshinski (10:55):
I talked a little bit around OneStream serving as that AI powered operating system for the office of finance. So when you think about OneStream, we really serve as that unifying platform for our customers to bring together not only their consolidations and close capabilities. So if when you think about the office of finance, there are a few core tenants that our customers look to OneStream deliver on. Number one is how do we deliver their consolidations and close process everything around how they ultimately produce the actuals, but also bringing those together and unifying that workflow with how our customers plan and forecast the business as well as drive all of the reporting and analysis. And so part of what's really notable here, and this gets into how we ultimately think about defining or putting clarity around a unified platform. Unified platform is not a bundle of loosely integrated tools or models.
(11:59):
It isn't multiple applications sitting on top of a data lake. When we think about a really a unified platform blend, really what that is is a single architecture where data, metadata, logic and workflows all operate on the same unifying model in real time. And that is what OneStream has delivered is that capability. So when we think about, just to further break that down, Glenn, you think about a data model that gets into how our customers have their actuals, their plans, their forecast, as well as operational drivers, one metadata layer. So all their dimensions, their dimensional members, their hierarchies and their business definitions, one calculation engine. So you're not having to deal with synchronisation of the reconciliation across point tools and also one workflow layer. So specifically for planning, consolidations, reporting all unified together. So the one-on-one stream has never been more important than it is today.
Co-Host: Glenn Hopper (13:02):
You'll notice Paul is wearing his Excel gear, I think he's feeling threatened and this for our viewers, Paul got his function hat on and his Excel MVP as well. But actually Paul, I know you had a question about that, but I think for anyone who hasn't worked in a system like OneStream before, it is maybe hard to visualise.
Host: Paul Barnhurst (13:26):
Yeah, Glenn, as far as the hat and shirt, let's see, I did an Excel community hour and a half and I did a bunch of training around Excel and I didn't change, that's what it comes down to. I actually thought about like probably should change, I don't have time, we're just going with it. So there you go. No offence to OneStream. I love the planning tools as well, but today is an Excel app, so we'll have a juxtaposition for people reminder that no matter what tool you use, you're probably going to use Excel or a spreadsheet along the way. Alright, so as we talk about unified platform, Scott, I'd love in practical terms, what does that look like for a Fortune 500? You mentioned, hey, unified cube workflows are aligned, but comparing that to a legacy stack of tools, walk us through what that looks like.
Guest: Scott Leshinski (14:11):
So I talked earlier around again that common data model, but part of what's notable within OneStream is the flexibility and Paul, to tie back to the Excel connection, it's not lost on us what the customers expect in terms of end user experience and the flexibility. So while we do have that common data model, one thing that's notable about OneStream is what we refer to as our extensible dimensionality. So that allows the customer then to look at the business under a, again, going out, whether it's on a geo basis, whether it's on a business unit basis, whether it's on a product basis, and whether you're looking at it for financial planning or operational planning to have the flexibility in that common data model to support the distinct needs of the customer from a design standpoint. So this all becomes really important because again, when you think about OneStream, and I'll tie that in where that becomes important later when we start talking about artificial intelligence, but I think what's notable here is where and why this becomes important when I start talking about the difference between integration.
(15:19):
So when I hit on an earlier and I talked about one data model, one metadata layer, one calculation engine, one workflow layer really being the keys to the one in one stream. Now the difference though, Paul, is when you start to carry that forward, the difference between will we see point solutions being referred to as more integrated solutions versus OneStream, which is much more of a unified platform, those differences become even more notable now in this new era of ai. So when you think about historically finance teams could survive that level of fragmentation and point systems because they accepted the need to go out and do the reconciliation across inconsistencies, whether it was in terms of calculations or how you were tying out forward-looking either the budget or forecast to actuals. But the point being is in addition to accepting the need to go out and do the reconciliation on those inconsistencies prior to ai, I would also say though Paul, that planning cycles were typically slower in nature, meaning you were going out and you were doing a planning cycle, let's say every month or every quarter.
(16:37):
So the insights, the latency there has historically was tolerated. When you fast forward to where we are now, AI really starts to change the rules on this. And what becomes notable here is that AI really amplifies that fragmentation risk that I was talking about earlier. So the metaphor that we oftentimes use is when you start thinking about fragmentation and how in an AI world it shifts really from inefficiency to ai. Now in this AI era, it really becomes kind of structural failure if you will, and you can't run a high speed train on fragmented tracks. And so now with ai, what's notable about OneStream is we see our customers starting to move from what before were these less frequent processes. And really with ai now Paul, what we're seeing with OneStream and with our AI capabilities is many of our customers are moving more towards what we would refer to as continuous planning, so not periodic or continuous close processes. So forecast update dynamically based on actuals as well as updates to some of the drivers that they're using. And AI models then are operating on this near real time, but also financialized data within OneStream.
Host: Paul Barnhurst (17:59):
Thank you for that. When you mentioned the rail system, I couldn't help think of a comparison between US and Europe when it comes to rail. We have a hard time running high-speed rail because it's pretty disconnected here in the US where Europe it's much easier to run, much tighter, also helps the geographic. But kind of that comparison you talk about, and I get it right in today's age, the AI is that decision layer. And if all your data is fragmented, it's much harder for that decision layer to help you make better decisions. Sure, you can still get some efficiency, but you're missing some things when things are fragmented. So that makes sense to me. I get what you're saying.
Guest: Scott Leshinski (18:36):
And Paul, I would just tack onto that. Now in this AI era, the question isn't whether you have ai, it's whether the archite underlying architecture allows AI to work effectively. And so a unified platform isn't a technology preference per se, it's really the foundation in an AI driven finance function. And without it, AI can ultimately scale confusion within the business. But really with that unified platform, Paul, that's what really helps to scale intelligence within the business.
Host: Paul Barnhurst (19:09):
Let's shift a little bit to what I call the feels like the topic of the year agents.
Guest: Scott Leshinski (19:15):
You
Host: Paul Barnhurst (19:15):
May have heard of them. I'm going to guess we
Guest: Scott Leshinski (19:17):
Wouldn't be a discussion without 'em.
Host: Paul Barnhurst (19:20):
Yeah, exactly. I know you guys have launched a few different agents. You have a kind of finance analyst, the operations analyst. I think it's search and deep analysis. So first, when we say ag agentic AI in the context of finance close process where audibility matters, what does that actually mean? How do we think about it? Is this all a probabilistic process? Is it deterministic? Maybe talk a little bit about agents, help our audience understand what that means in this finance environment.
Guest: Scott Leshinski (19:49):
Let me start with a little bit of some of the background and some of the context. I think what's interesting here is as we move forward, agents are really becoming part of the workflows on how customers expect to experience and interact with the solutions that they use to run the business. And they're expecting to be able to operate leveraging more natural language to interact with their systems and with their data. So that's a little bit of the background here, but I think what's notable and what really makes agents come to life, and I also want to take a minute there Paul, just to also kind of bring in into the discussion the distinction between when we start talking about agenta capabilities, I'd also highlight the distinction here between general purpose AGTA capabilities versus more what I would call finance grade agen to capabilities. And I think it's a notable distinction to bring into the discussion, Paul, because just general large language models out there are obviously trained on large, broad generalised data sets and those are incredibly valuable and they're great at going out and doing things like drafting narrative summaries and commentaries there, maybe doing some general explanations on trends and maybe generating some conceptual perspective for the end user.
(21:21):
But what they really start to struggle with in comparison to more specialised native agents, which is what we've now brought to your point within our sensible AI portfolio, is really that deep contextualised understanding of your data of your business in order to really build that context. That ties back to what we were talking about earlier in terms of that underlying data model within a unified platform. So to further elaborate on that, if we were to go out and ask a generalised or a general large language model a question and we're sitting in the function of finance and you were to ask it a very detailed question around let's say we were looking at something like EBITDA or free cashflow and we wanted to look at that in a line of business in a geography for a specific product line year to date compared to prior year to have that level of contextual understanding, it really doesn't exist in a generalised capability.
(22:25):
So what we've done with our agents, Paul, and specifically our analyst agent rather, is we've delivered that detailed contextualised understanding within OneStream. So what we've come to appreciate is if we were to deliver an answer that was 80% accurate, it's 0% useful to our end customers. We need to deliver capabilities that are highly performant, that are rapidly delivered, the responses are quickly delivered to the end user, but that have near a hundred percent accuracy in the answers that they invoke, that we invoke our agents to go out and deliver for the end user. So I think what I'd start off with is our agents to use maybe a bit of a medical metaphor, we aren't delivering the general practitioner. What we're delivering is the specialist surgeon, if you will, in the medical field that has that detailed contextual understanding of our customer's business informed by that underlying data model that is moving more into a genta capability. So not as it now, not only is it going out and able to produce highly informed, highly accurate answers, but it's also now getting towards the place where it's actually getting into orchestration of capabilities, workflows, visualisations for the end user.
Host: Paul Barnhurst (23:51):
Ever feel like your go-to market teams and finance speak different languages? Dismissal alignment is a breeding ground for failure in pairing the predictive power of forecasts and delaying decisions that drive efficient growth. It's not for lack of trying, but getting all the data in one place doesn't mean you've gotten everyone on the same page. Meet Q flow.ai, the strategic finance platform purpose-built to solve the toughest part of planning and analysis, B2B revenue QO quickly integrates key data from your go-to market stack and accounting platform then handles all the data prep and normalisation under the hood. It automatically assembles your go-to market stats, make segmented scenario planning a breeze and closes the planning loop, create airtight alignment, improve decision latency, and ensure accountability across the team.
Co-Host: Glenn Hopper (24:58):
Paul, I've got to say first off, I'm impressed that you brought up agents before I did this time there's I've on a soap for everything.
Host: Paul Barnhurst (25:07):
Go ahead and ran. Get it off your chest Glenn.
Co-Host: Glenn Hopper (25:09):
No, I'm not going to go down that. I'm not going to bore Scott with that. However, I mean you know the same, I don't mind boring you Paul, you've been listening to it for two years in my whole ran around the agents is people. As we rode the Gartner hype cycle, people were calling everything an agent. They didn't truly mean it was an agent and the idea, and obviously in 2026 these models have gotten much better at going off and performing long form tasks with fewer prompts and able to go do the work. I've been building bespoke agents, I'm doing agents and air quotes now because to get that deterministic result to the end user, it may look like an agent, but I was building pathways that go through a very deterministic process and then maybe there's some AI layers where AI can help and then at the other side, user gets it back, it feels agentic or whatever, but the truly agentic idea that it's a conditional loop and that it's going off on its own and figuring it out and it's going to come back to you, then that's amazing. And really, I've been angry about people calling things agents that weren't doing that, but so far this year I really am seeing that. But what I worry about in that is the same thing when I'm talking to controllers, okay, that's the magic black box problem where the agent goes out, does a bunch of work and then comes back and gives me an answer. How am I going to defend that and how do you guys address that for something that truly is working genetically?
Guest: Scott Leshinski (26:42):
It's a great question and to your point, I look at it where now we're entering what I would call kind of the next chapter of ai. So we're now moving beyond pilots and prototypes into true enterprise scaling. And that gets into how you drive end user adoption, how you think about ongoing sustained success for the customer. And you're hitting on a really key piece because in addition to change management, which you would obviously expect to be a key component of that, part of that Glenn gets into how do you drive that level of trust in what these agents are, what they're ultimately producing, the insights that they're delivering. And you don't build confidence in AI outputs in finance by asking people simply to trust the model. It really gets into a heightened level of transparency and explainability. So part of what we've done by design and with a high degree of intentionality, so when we started partnering with our customers, we do early access programmes, Glenn.
(27:48):
So we have a number of customers that engage with us and we've done this both with what I would call kind of our quantitative AI capabilities. So when we released our AI powered planning and forecasting capability called its sensible AI forecast, and then very similarly with our AGENTA capabilities, our agents, whether it's analyst search or deep analysis, part of the consistent feedback there from our customers is the expectation on a heightened level of transparency. So part of what we built into our sensible AI portfolio are a few different things. Number one, you'll see again when you go in and ask a question within OneStream, ask our agents a question. We built in what we refer to as chain of thought reasoning. So you can actually follow the logic that the agent is going through, you can see as it ultimately invokes and starts to interact with the underlying data model and you can literally see the drill through to allow our customers to have the traceability.
(28:51):
And what's notable there, Glenn, is in every AI generated output you need to be able to tie back to a governed data source and ultimately the underlying data model within OneStream. So chain of thought reasoning is incredibly important. The second piece that I would bring up there is really what we would call the validation framework. So when we get responses from our sensible AI solutions within OneStream, we also tie those out and similar to what you brought up in terms of the level of discipline and rigour that you would expect, whether it's in the finance and or controllership activities on the accounting side of the world, we actually tie out, so we benchmark our AI outputs against historical actuals and known baselines for the business. So we're always tying out and we ultimately track performance over time and use very objective metrics things whether it's mean absolute percentage error, mean squared error and use objective quantification capabilities to go out and always be tracking back to ultimately determine how accurate are these model results and then ultimately establish really tolerance or performance thresholds.
(30:09):
So if I tie that back to sensible AI forecast, it could be forecast accuracy bands, Glen, it could be variance explanation completeness. So we actually start to track these metrics and use telemetry within OneStream to make sure that we're delivering again what I would call kind of finance grade AI capabilities. That again, because 80% accurate is 0% useful. We want to make sure that we're always holding ourselves to a high standard there and building these capabilities into OneStream and getting into things like we also build in what we call our the feature and event explanations within. As an example, if you look at our AI powered planning and forecasting, we've built in a feature and event explanation capability to drive transparency to the end user because it would be pretty indefensible to walk in to the CFO's office with a forecast that we simply look at and say, well, how were these results produced? Well sensible AI forecast produced them, just trust us. What we do is we build an incredibly detailed level of explainability into our solutions that allows the end user then to engage with those results and be able to defend and explain those results. Glenn, and
Co-Host: Glenn Hopper (31:30):
I think you didn't say it, but based on the forecasting model that you're using, it's important. The distinction between generative AI and machine learning that is used in a forecast is very different because generative ai, probabilistic and a create something new every time in your chat bot. But the machine learning can be very determined. I mean given the same inputs, you're running it through the algorithm and that's the trust that people have around AI could be based on chat. GPT lied to me about something. And it's different if you're looking at a machine learning forecasting model. And I think that's an important distinction to make.
Guest: Scott Leshinski (32:13):
It absolutely is. And that is again where that term of unification becomes incredibly important because we've now brought, and to your point, making the distinction between generative AI versus machine learning planning and forecasting two very different things. But to your point, there's an incredibly high standard as we move from probabilistic into deterministic capabilities here and the unification is important because what we've done now is we've unified those machine learning capabilities natively within the very business process or workflows that our customers are running on OneStream. So I'll stay with planning and forecasting for now, and the expectation there is you're moving more towards a deterministic model where you have a very predictable behavioural pattern on if I run this forecast, how does it ultimately impact the p and l? There's a much greater expectation there because you're using that to drive data-driven decisions within the business.
(33:14):
So again, where we see that now, Glenn, and I'll just share with you part of the outcomes that we see there, Glenn is now by, we see more and more of our customers moving towards not the exception but the norm where they expect machine learning generated plans or forecast to serve as the baseline plan or forecast for the business. So it's the first forecast to react to what's notable there and kind of the evidential support of how our customers are moving towards that. Glen is we're seeing now on average, and we've had the privilege to partner and deliver at this point nearly hundreds of different use cases for our customers. We've seen our customers now be able to improve forecast accuracy by 24% and all the while reduce the amount of time spent generating the plan or forecast by 86%. And again, that all gets into those traceability and transparency metrics that I talked about earlier. But all of these concepts are really reinforcing to one another.
Co-Host: Glenn Hopper (34:21):
Watch this humble brag, Paul, go ahead, get it off. I love everything you just said. So during COVID I wrote a book about using machine learning and finance and guess how popular that was? It was maybe a little bit early, but
Host: Paul Barnhurst (34:40):
I have it on my shelf, Glenn,
Co-Host: Glenn Hopper (34:42):
Only because I gave you a copy, I made you,
(34:46):
I kept sending it to you and you kept nobody wanted to talk about AI back then. And it makes complete sense that you're in this role now coming up through the AI side and it's people only start going back to Netflix, movie recommendations or Amazon product recommendations or self-driving cars that have been around somehow was just whatever, something happening behind the scenes. But then generative AI that people can talk to and get these results. It's a jagged frontier. So some can be amazing and some can be completely hallucinated, but we all see the potential of it. And I'm wondering from, there's plenty of applications for generative AI in financial planning and analysis and in finance you just have to know where to use them and what guardrails to put on them. But I'm wondering if people, you could be the greatest controller, the greatest CFO in the world with your lane specific to finance and accounting, now you have to be a pseudo technologist and figure out this technology that at times can look like magic.
(35:57):
I mean, what is the difference between, I don't know, I do encourage experimentation with off the shelf tools like chat GPT, you can see that and understand it, but a lot of times people have a hard time seeing the difference between that and what you guys are doing. I know we've already covered the machine learning piece and you referenced this earlier, but maybe dig a little more into the case for AI that's purpose built for finance and how that differentiates. And I think you've already covered this a little bit, but it's just, I think it's really important to talk about that difference between, well I've vibe coded this thing or I've got Claude to do a Sima forecast for me and it came up with a chart that looked pretty cool. There's a big difference between that and the way you guys are delivered.
Guest: Scott Leshinski (36:44):
Yeah, so again, when you think about, and I'll use that same metaphor where there are general, what I would consider to be kind of the generalised large language models, which obviously are what generative AI is built on top of those LLMs that again are really trained on broad generalised data sets. So a lot of the publicly available information and they have a really strong capability of internalising that you can take some of your financial data, typically static data extracts out of your data lake data warehouse and push it over there. But really what we've come to appreciate is, particularly in the office of finance, there's a greater expectation there, an expectation that these capabilities when you're asking questions number one, are highly accurate. So as I said earlier, 80% accurate is 0% useful. So there's an expectation that the results that these generative AI capabilities are delivering for end users are incredibly accurate, number one.
(37:57):
Number two, there's an expectation that they're highly curated, highly contextualised on your business, meaning they have that very detailed, very explicit understanding of how you look at how you think about how you model your business to really get that detailed contextual understanding. It has to understand how you roll up your actuals meaning or how you prepare your plan or forecast to get that detailed contextual understanding. It has to have a deep detailed understanding of the data model that you use and how you report how you plan or how you close the books as a company. And then the third aspect of that then Glenn, is it has to be trained on your proprietary data set and be able to do so in a secure, trusted environment. You don't want to push your data out outside of and push your data out to these publicly available models.
(39:01):
He'd rather bring the models to your data and make sure that you're not exposing particularly finance and or accounting data. Obviously I think we can all appreciate the level of security that goes around that and OneStream serves as the book a record for our customers. So that is part of what our customers have come to expect. So just to close that off, Glen is we've delivered that highly curated, highly contextualised deep understanding in contrast to these general models. So when our customers want to go out and they either want to orchestrate some type of capability, whether it's a reporting or visualisation in OneStream and they want to do that and ask very detailed discovery questions around their data to unlock informational insights that before were very hard to get at. Now you can leverage the enta capabilities in OneStream to automate the orchestration and the creation of those insights. So creating any form factor visualisation of those insights, but not sacrificing or subjecting the business to trade-offs around the level of detail or accuracy of those insights. That is what's fundamentally different in what it is that we've delivered here. So
Co-Host: Glenn Hopper (40:23):
There's the context that it gets during training and the guardrails that come with that, but also the context that it gets from one, understanding the system, the universe that it lives in, and two, that very unique customer data that it knows the schema and the semantic layer of how to access all the information and everything. So keeping it within that universe, it's not going to go off and hallucinate because the best trained model in the world that is trained on publicly available data isn't going to have the context of your business and your company and all that. So I guess what you're saying, so a lot of that I guess is training time, but then it's happening at inference as well just because it knows the system and it's already been given the context, right?
Guest: Scott Leshinski (41:13):
That's right. And that is part of the rationale. Obviously we share a strong partnership with Microsoft and we look at our partnership with Microsoft is really a force multiplier when you start thinking about that with Microsoft RSIs and OneStream, the value that we're delivering to our customers. But with Microsoft, I think what's notable here is that power of three model, but this is also why Microsoft is now bringing our agents into their productivity tools. So into O 365 there is to surface our agents in some of their productivity tools as well. And I'll tie it back, Paul, to your Excel shirt from earlier, that is the power of three when you start thinking about OneStream, Microsoft and our expansive SI network as well.
Host: Paul Barnhurst (42:05):
Microsoft has its hands in just about everything, always a good partner to have. They've done a lot of really good work, so thank you for sharing that. So I want to shift gears a little bit and there's some research that you guys did recently and in there they mentioned that 89% of finance students kind of feel AI ready. If you look traditionally when someone came out of their career, students generally didn't feel ready for the real workplace. There is a real learning curve, but you ask senior professionals and only 54% say they're ready. What do you think that inversion means for companies? How's that going to change things? I'd love and I get, there's no right or wrong answer here, but just your thoughts on that and what it means for talent upskilling and just the future in finance as we see a lot of uncertainty and a real change on how prepared people are for certain things.
Guest: Scott Leshinski (43:01):
Yeah, history doesn't repeat itself, but it certainly rhymes. The early part of my career, I started off in corporate finance and back then really the superpower was how good you were at building Excel models and doing financial modelling in finance. You fast forward then and that next big chapter out there was the ability to have superpowers in enterprise software. So were you a distinguished architect in the OneStream world that was capable of doing really robust things in OneStream? And I think the next chapter that's opening up now, Paul, is really an AI driven chapter which is now allowing us to really compress the value of traditional fp and a work. So things like when you start thinking about some of the tasks that historically we spend an inordinate amount of time working through and dealing with whether it's or reporting or generating a forecast or creating narration or narrative documentation.
(44:06):
Now with ai, a number of those activities become automated and now it allows us to kind of rethink where and how we spend our time. And so to your point, the level of AI proficiency and some of the re-skilling that's required with this becomes a real thing and it's important. But I think what's notable though, if we were having this discussion two years ago, I think the approach, the metaphor I would use here is you'd have to go through a lot of training to be able to go and climb the mountain. I think now with the way ag agentic AI has evolved and accelerated, and particularly when you think about your own personal use and when you have questions or you're looking for information, where do you go nowadays? You go out to some form of ai, whether it's Claude, whether it's open AI and chat GT, and you ask questions there.
(45:06):
And so my point is, is now I think instead of having to build up this great capability to go and climb the mountain, I think we've lowered the mountain now by bringing these capabilities into where our customers work. And that is again, the importance again of unification is now in addition to our native agents that we've delivered in OneStream, we've also opened up using MCP server capabilities. We've now brought some of these large language models and generative AI capabilities natively within OneStream as well. So you have a different form factor to help drive some of that knowledge base for our end users. So you can access, it's almost bring your own agent now within OneStream where now you can bring Claude or chat GPT into OneStream. We've trained it now on a lot of the publicly available product documentation that's available and it helps to inform those end users and give them the knowledge base they need for self-sufficiency. So just to close it off, Paul, I would say two years ago if we were having this discussion, the approach would've been fundamentally different. It would've been a pretty significant overall. And don't get me wrong, there's still some re-skilling that's taking place in this era of ai, but I think with the rapid advancement of some of these ag agentic capabilities, we've lowered the hurdle to become proficient and self-sufficient with ai.
Host: Paul Barnhurst (46:37):
Go ahead Glenn. I could tell you're chomping at the bit.
Co-Host: Glenn Hopper (46:40):
Yes. Well actually I'm thinking I want to ask you the question that I'm asked every day, the pace of change, and we've referenced it a couple times during the show, it's so rapid and it's also so as I said jagged before too, but it we've rocket forward and then you hear about some really dumb mistake and you hear there's pressure now, it's been growing each year probably since chat GPT-3 0.5 came out for not just finance, but all work functions to change. How are we going to get more efficient? What does this mean? Are we all going to lose our jobs? Are we going to be replaced by robots? And it's hard to see where that is right now because maybe I'm too much of an optimist, but to me right now I see AI being very good at specific tasks and if your sole function was to swivel shared data from one system to another, that's probably been kind of on something that was ready for automation for a number of years now.
(47:45):
But for the key contributors, I picture we're it's moving, I'm going to do this Paul, I'm going to reference another quote I say all the time, but I think it's really relevant here. Cliff STO has a quote that I use all the time and I use it sort of inversely to the way he said it because I think that as AI gets better, it moves up the chain. But the quote is, data is not information. Information is not knowledge, knowledge is not wisdom and wisdom is not understanding. So I picture AI just getting better, the automation RPA did some level of work transforming data we could do and we're kind of moving up, but it's happening quickly. Tools like OneStream and we're now connecting more and more. I mean, I don't know, I'm asking you to put on your futurist hat here. What do you see happening with EPM and the FP and a function in the next three to five years? And I know there's not really a single answer on this, but I'm asked it every day, so I'm leaning heavy on you to help me out here
Guest: Scott Leshinski (48:55):
And maybe to draw off the metaphor that you're using there. I do believe that particularly when you start thinking about performance management and OneStream in that area, we are really moving now from systems of record to systems of intelligence. And I think we're seeing whether we're seeing automation of some of those highly repetitive that are just time intensive activities, those highly repetitive activity, we're seeing automation now of some of those highly repetitive activities, whether it's things like account reconciliations, we're seeing our customers now being able to compress some of those down. But what I do think is notable though is when you look at it, there's still the human in the loop, even in this future version for finance. I mean at the end of the day, you need humans in there that understand the context, that have judgement , that drive decision ownership on some of these activities, Glenn, and that really understand the system, what I'd call the overarching system design and the orchestration across these areas.
(50:01):
And you do need someone who can go out and ultimately understand the logic behind how some of these decisions are being driven and be able to make decisions on top of that data. I think part of what's also notable here though is at the end of the day, Glenn, the future isn't another layer of tools. This is where we really believe, again, it's a unified intelligent platform and we do believe that the winners in this next era won't be the companies with the most features necessarily. It'll be the ones with the most intelligent, unified platform powered with ai. And that's the future that we're ultimately building towards at OneStream is bringing those capabilities where we're trying to allow our customers to really focus on the job to be done, the most efficient way to do it with the most effective and performant outcomes from that
Co-Host: Glenn Hopper (50:58):
Exciting times. And I do the going private right now makes a lot of sense and being able to really lean into everything that you guys are doing and it really is going to be required to move quickly and to go more towards that long-term thinking to continue riding this wave and be on the front of it. So this has really been a fascinating episode. We do at the end of it, we have a couple of, I dunno, Paul, do you want to explain? I think you do a much better job explaining our Sure,
Host: Paul Barnhurst (51:29):
I'll be happy to explain this. So basically we throw some random questions at you and see what you say. Now lemme get a little more context. There's the very short version. We feed ai the internet, which has access to the questions. We came up with your bio and we ask it to generate five kind of fun, unique. Sometimes we'll throw in the word quirky questions and we see what it comes up with. So we use a different LLM, we just pick one every week. We don't have a rotated series. I feel like doing this one or that one. We come up with questions, we don't use the exact same question or time. So Jen or Glenn came up with these with ai. Which model did you use this week, Glenn?
Co-Host: Glenn Hopper (52:04):
I used Opus 4.7 for these because I realised this morning and I was in Opus, which I am most days in Claude, which I am most days. I realised that we didn't do the AI generated questions. I quickly dumped in LinkedIn profile and told Claude to go search the internet and see what we could come up with. So that's where the questions came up with this.
Host: Paul Barnhurst (52:23):
So one time I gave it the LinkedIn profile, it told me I can't search LinkedIn, I don't have a user account. I'm like, guess you can look at a profile.
Guest: Scott Leshinski (52:31):
So we
Co-Host: Glenn Hopper (52:32):
Do it slightly different. Paul gives you an option, I just let the bots choose completely. So Paul, what's
Host: Paul Barnhurst (52:38):
The option? Yeah, so one of two options. We call it human in the loop or AI all the way. So how the human in the loop works is you get to pick a number between one and 25 and I'll ask that question or two. The random number generator picks a number between one and 25. And I'll ask that question.
Guest: Scott Leshinski (52:57):
Let's go with lucky number seven.
Host: Paul Barnhurst (52:59):
Seven, alright, you've spent your entire career in Chicago, read these ahead of time. So if something's wrong, that was ai, not us.
Co-Host: Glenn Hopper (53:08):
The ai.
Host: Paul Barnhurst (53:09):
What's the one restaurant you take every out of town? Client
Guest: Scott Leshinski (53:14):
Only one. Well, I've primarily lived in Chicago. I've had the opportunity to work and live in a variety of different places over the years. If I was going to one restaurant, now keep in mind I don't get out as often anymore. Now we have two young kids so our life is a little more limited on that. But I would say if I was going to go to one restaurant, it might be Twin anchors in Chicago.
Host: Paul Barnhurst (53:36):
Alright, twin anchors. I'll have to try that next time in. I'm in Chicago, whereas
Co-Host: Glenn Hopper (53:41):
Twin anchors.
Guest: Scott Leshinski (53:42):
Twin anchors, it's a staple of Chicago. It's been around for probably a hundred years now and it's a old school barbecue and rib restaurant in Chicago. It's been in a number of movies so it does have some notoriety there, but it is a staple in the city.
Host: Paul Barnhurst (53:57):
Glen, we might need to keep this podcast on the road.
Co-Host: Glenn Hopper (53:59):
Yeah,
Guest: Scott Leshinski (54:00):
There we go.
Co-Host: Glenn Hopper (54:01):
Yeah, I'm in Memphis, so I hear about a good barbecue place in another city. I'm immediately ready to fight now I've got to check it out. Let's see. So my approach to these is I feel like wrote the questions, I just ask it to pick the best one and hang on, let me submit this really quick. And
Host: Paul Barnhurst (54:18):
Who knows what AI uses for best, right?
Co-Host: Glenn Hopper (54:21):
Yeah, no, actually I just told it pick one, not pick the best. So it pick number one, Paul, I said pick one and it gave me number one.
Host: Paul Barnhurst (54:29):
I think there was a prompt issue there, but we'll let it go this time.
Co-Host: Glenn Hopper (54:31):
Not a bad question here. I think assuming that it didn't hallucinate this, you are a Michigan state, Spartan, do you have a go-to game day tradition?
Guest: Scott Leshinski (54:40):
So we have two young kids. So I will say when it's a college football game day, you'll find that there's a lot of green being worn in our house. So big fans of Michigan State football, which has been tough over the years, but more importantly Michigan State basketball. So if I'm home, you'll find that oftentimes I'll be watching the game. We are with our daughter and with our son wearing full green and white.
Co-Host: Glenn Hopper (55:02):
Love it.
Host: Paul Barnhurst (55:02):
They got a good basketball programme for sure. It was
Guest: Scott Leshinski (55:05):
One of
Host: Paul Barnhurst (55:05):
The best coaches out there, no question.
Guest: Scott Leshinski (55:06):
That's right. Now this is fun guys. I appreciate we as OneStream also appreciate the partnership on this, the opportunity to join the two of you today. Whatever I or we can do to help and particularly obviously as we build towards the forward finance panel at Splash, let me know if there's anything more that I can do to help in preparation for that.
Co-Host: Glenn Hopper (55:25):
Really appreciate it. So Scott, thanks so much for coming on. Yes,
Guest: Scott Leshinski (55:29):
Thank
Co-Host: Glenn Hopper (55:29):
You
Guest: Scott Leshinski (55:29):
Paul. Great to be with you both. Appreciate your time. Look forward to the next discussion. Thanks so much.
Host: Paul Barnhurst (55:34):
Thanks for listening to the Future Finance Show. And thanks to our sponsor, QFlow.ai. If you enjoyed this episode, please leave a rating and review on your podcast platform of choice, and may your robot overlords be with you