Discovering New Dimensions In Accounting with Sasha Orloff

Welcome to the Future Finance show where we talk about robots and how they would come for your job.

I, for one, welcome our robot overlord. I think AI is just going to remove the need to be technical, but still get the answers you want because you know what you want to ask for. You just don't know what the answer is. My core underlying belief is one of the things that's been holding the entire industry back is the lack of an underlying data platform that's built for finance.

Native to AI, accounting is the foundation. Without proper accounting, you can't build the financial practices on top of it.

So if we don't set the foundation of like, rules and visualization the right way, it's hard for AI to do anything really that impressive. We focused on two pieces that I thought were really important outcomes. One is to improve the trust and transparency. And then the last piece are the debates that we have about what to do with the capital and the people, the resources of the company, how we think about building trust in ecosystem. There's a lot of really exciting stuff to learn from, like the world of blockchain. Blockchain builds the most trusted system that we've developed today.

Do you have a favorite finance software app?

One that I'm really excited about is.

Future Finance is brought to you by qflow 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 qflow AI.

Yet another podcast from Paul Barnhurst and Glenn Hopper. Man, these guys must really love to hear their own voices. Well, maybe. But also, we're very passionate about the evolving world of corporate finance and the tech that's driving it. Here at Future Finance, our aim is to focus on the technology and trends that are driving our profession, bringing you important updates and understanding around the technology that is going to shape not just finance and accounting, but every profession and industry in the world in the coming years. And you're going to hear me talk about generative AI a lot. And here's why. Unless you've been hiding under a mechanical loom for the past two years, you've probably heard a bit about these LLM chatbots like ChatGPT, Claude Llama, Gemini, etc. These models have the potential to revolutionize the way we do, well, everything. They're more than clever chatbots. And with each technological iteration, we're seeing massive advancements in the technology, from expanded use cases to emerging capabilities. And now we're seeing generative AI incorporated into existing software that we use every day. It's giving us a new way to interact with computer systems and software. A new way not just to automate our work, but to derive additional value from computer systems. These tools not only make it easier to analyze and report on data, but can offer insights and recommendations based on our data. They are becoming resources that can provide insights and inputs. We may not be 100% there yet, but it's coming, and probably sooner than you expect. I'm not saying AGI is right around the corner. Maybe it is. But regardless, we don't need AGI to derive value from these tools. Let me explain why. The latest phase of the AI revolution has been driven by Transformers, a transformer. If you're unfamiliar, it's a powerful type of neural network that learns context and meaning from sequential data. It does this by tracking relationships between elements using a mechanism called attention. Introduced in a 2017 paper called Attention is all you need by former Google researchers. Transformers have driven breakthroughs in machine learning and ignited the Transformer AI revolution, which let's say started in earnest in around, well, in 2022. From that we saw the emergence of LLM powered chatbots like ChatGPT, and 2023 was clearly the year of generative AI or the year of Taylor Swift, depending on where you get your news. These chatbots are great at one shot answers, meaning they're designed to read your question and provide a response. Current chatbots, they're not great at reasoning, they just read the question and respond. And there are some, sometimes there's errors around this problems like not being able to solve logic problems or hallucinating, which I'm sure we've all seen. It's where an LLM generates a wrong answer and displays it with absolute confidence. LLMs also aren't very good at math. There have been some emergent capabilities around this, but as finance professionals, we're not at a point where we should trust LLMs to even do something as simple as, say, build an amortization table. I will throw one note out there on this though. Tools like Chat GPT's Data Analyst, which actually writes Python code under the hood, that that's actually very good at these kinds of tasks. But that's all just foundational for the next step where we're going to see true advancements. This next step is in the realm of agents. If, if 2023 was the year of generative AI, then 2024 is going to be the year of AI agents. I think there's a lot of confusion out there. Now, about what agents are. Here's the best way to think of what an agent is and what it does. Like I said, LLM chatbots read or hear your prompt and then they provide a response. They don't pause to think before they answer. They just start generating novel data. And they're not very good at solving problems that involve multiple steps. They require continued prompting, directions and interaction to complete tasks. But an agent could read your prompt, no matter how complex, and then go work through the steps required to solve it. It possesses a level of autonomy to work through these problems internally, and then they output a response when completed. We haven't seen a completely successful iteration of these tools yet, but leading researchers and entrepreneurs are working on solving for this right now. And in the interim, there are plenty of off the shelf software tools that are out there right now, many of which can be designed with no or low code interfaces that allow users to create agentic workflows. So an agentic workflow is doing what an agent does, but it requires more handholding and prompting. These agentic workflows are capable of performing complex tasks. They string together multiple prompts and outputs to solve for these tasks beyond what a typical chatbot could output in a single response. And when we start to see mass adoption of agentic workflows and ultimately true AI agents, well, that's going to be the moment when even the late adopters among us will have to take notice. About a year ago, I designed a framework for how to build an AI powered finance and accounting chatbot. This chatbot could perform tasks of, say, a junior financial analyst. I'll include a link to the framework in the show notes. Well, I'm talking about agents today because I finally realized how close they are to happening. Now, they may not yet be truly autonomous, but if we can design agentic workflows that complete the tasks that an agent would. These workflows are deployable and scalable and will give us the same result that a than an agent would. The agentic workflow that I built was designed to analyze and report on a company's annual SEC report known as the 10K. And I built this tool using make.com I used make Perplexity and several custom assistants that I built on OpenAI's platform. And I did all of this with zero coding. There's other tools out there that allow users to create these workflows. I'm thinking of tools like Langflow, mindstudio, Zapier to some extent, and one called CREW AI. Crewai does require some coding, but it gives builders even more flexibility to create these agentic workflows. If you aren't familiar with agents, my pro tip for the day is you need to learn about them right now. I'll post some links in the show notes to where you can learn more. Look, I know we're all busy, but if you're listening to this show, we know you're interested in the future of finance and and in learning how you can integrate AI into your professional career. It is incumbent on all of us to not just pay attention to this tech, but to learn how to use it. Because while we may not yet be at a point where AI is going to take our jobs, I do believe that the future will be divided into two camps. One, people who use AI to improve their lives and work tasks, and two those who don't. One of these groups is going to be left behind.

This week's news story is brought to you by your co host Paul Barnhurst, the FP and a guy. What's some of the latest news out there about AI? Well, just recently Claude AI, which is backed by Anthropic, announced they released their new AI iOS app. And one of the things they mentioned so I had to test it, is it will access any of your photos on your phone and analyze them. So of course, what's the first thing I did? I uploaded a photo of myself. Let's see what I had to say. So I said, here we go. Analyze this picture. That's it. And it shows. The image shows a close up selfie of a man with a long reddish blond beard, wearing glasses and a green baseball cap. He has a thoughtful, slightly concerned expression on his face. Pretty accurate. The man is wearing a black jacket with the Amplio logo on it, suggesting he make work for or be affiliated with that company. And then it tells me the picture was taken indoors, probably at a home or workplace. Does a really good job. Then I said, let's have a little more fun. I asked it how old is the man? And it said based on the man's appearance in the photo, including the gray hairs in his beard, technology, some visible wrinkles, an overall mature look, I would estimate he is likely in his late 40s to mid-50s. Then it goes on to know how you can't tell the age from one picture. I'm 48. Nailed it. Next I asked it if I was overweight. You probably all laugh at this one. It said from what I can see in this close up photo, the man does appear to be overweight or obese. Ouch. Based on the fullness in his face and neck area. However, it's important to note the weight assessments from facial features alone can be unreliable. Then I had a little more fun. I said, based on the photo, do you believe the man is well educated? And it's like I can't answer that. There's not enough information. I said, do I appear to be healthy? Again, not enough information. So my new story is Cloud AI now has an iOS app. Not Claude AI. Yeah, Cloud AI just like ChatGPT. So kind of a fun one. Go ahead and upload some pictures, ask it to analyze them. Does a really good job. It did nail that. I'm overweight and it just about nailed my age. I was quite impressed. The technology just keeps getting better and better. So let us know in the comments. What is the technology that you love today? What is you see? What are you seeing out there? What's your news story? Ever feel like your go to market teams and finance speak different languages? This misalignment is a breeding ground for failure, impairing 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 qflow AI, the strategic finance platform. Purpose built to solve the toughest part of Planning and analysis. B2B revenue. QFO quickly integrates key data from your go to market stack and accounting platform, then handles all the data prep and normalization under the hood. It automatically assembles your go to market stacks, makes segmented scenario planning a breeze, and closes the planning loop. Create airtight alignment, improve decision latency, and ensure accountability across the teams. Welcome to Future Finance. This is our inaugural episode. I am your co host, Paul Barnhurst. So I'm super excited to have here with me Sasha Orloff and my co host Glenn Hopper. So a little bit about Sasha. He comes to us from San Francisco. He is the CEO of Puzzle. We'll talk more about what that is in a minute. He hosts the finance podcast by Turpentine and he previously served as the founder and CEO of LendUp and Mission Lane. So Sasha, welcome to Future Finance.

Thanks for having me. Excited to be here.

Yeah, we're really excited to have you. So why don't we start with you telling us a little bit about your background. Tell us about yourself.

Yeah, my name's Sasha, as we just heard before, and I'm building Puzzle, which is modern accounting software for growing companies. And my, my sort of background was I've worked in, in fraud modeling, a lot of math and inclined types of roles. So I worked in predictive modeling and fraud modeling. I worked in banking infrastructure, I worked in lending and payments, worked in venture capital for a little bit and got excited about this idea of building a business where you could borrow money over a mobile phone. And so when I couldn't find the team to invest in as a venture capitalist, I got excited and I joined a startup incubator called Y Combinator in the Bay Area, built Lendup, ended up spinning out a division called Mission Lane. Each of those grew to a couple hundred million of revenue and took some time off and decided to start something new.

How did you pick the exciting world of accounting as your next project?

Well, I mean, I think as most of the listeners will, will know what is more exciting than accounting? Not many things, but effectively, at the end of the day, I did. What I got excited about was, was solving a problem. And the way that this dawned on me was, was sitting around this table as a CEO, I would look at all my direct reports and everybody had beautiful software. They would open up their laptop and they could answer in real time, everything except the finance team. And they would have a pad of paper and a pen and they would write down all of these questions. And I was constantly frustrated and wondering, like, why can't you do this? And then you have one of those kind of moments where you're like, oh my God, it's not the finance team's fault, it's their software. And we realize that we just as an industry have a high tolerance of pain for mediocre software. And, and when that thing hits the mind of an entrepreneur, it's hard to ever let go of that excitement.

Love that. Glenn, you've written a book all about technology and corporate finance. What's your take? Is it mediocre software?

Yeah, well, so, I mean, that's kind of a big struggle in my career and it's why. So first off, the software you have isn't typically great. And then if you want to get something developed for a back office for finance and accounting, you going to the back of the list, you're not getting, you know, moved up to the, to the top because you're not generating revenue. So in my career, it led to a lot of citizen coding where you're trying to make up, you know, you're running an access database on your own laptop or something to try to get information and just been there. Yeah, so that, yeah, so, yeah, no, I, I completely relate and I, and I'M I'm right there with you. What is more exciting than the world of, of accounting and just figuring out depreciation schedules and being sure you're following Gap and. Yeah, it's all, it's riveting stuff. Yeah, we're gonna, we're gonna run off our accounting listeners is what's gonna happen. But if we're, if we make too much fun, bring it on.

We got a, we got a whole big team ready to build, ready to build, whatever. But I think you, you touched on something which is kind of hits to the point of, of why I get so excited about puzzles. When a somebody in accounting or finance needs to learn how to code to do their job, that's like a fault of the software of the industry, and that's a place where we think there's an opportunity that gets me excited.

Love that. So can you tell our audience what do you see as the future of kind of the general ledger in accounting? I know Puzzle is building a tool to help record all those transactions, but where do you think we're going here in maybe the next, you know, few years?

Well, it's, it's one of those weird industries where it has been decades, like 25 to almost 40 years since we've had a new, meaningful player in the space of a general ledger. And if you were to rethink, how would I build this from the ground up today, with everything we know today and how software can be built and design and all the technology advances, you just would build it differently. And so we did the hard work first of building the core general ledger from the ground up. And we focused on two pieces that I thought were really important outcomes. One is to improve the trust and transparency of the software and the whole concept of accounting, really, aside from, like, the foundation of data for all wonderful finance people to be able to model and think about predicting the future is like, we have to have a trusted, accurate source of, like, what is the financial truth of a company. And today that's done through a lot of manual steps and then the last piece. As a CEO, my favorite and most valuable and most impactful conversations are the debates that we have about what to do with the capital and the people, the resources of the company, and that is typically with a finance team. And right now, so much of the work is manual and rote and repetitive that you're spending so much time and resources and just getting to the truth. You don't have enough time to talk about how do we actually improve the business because we're spending all of our time just reporting on the business.

So I'm wondering, I'm thinking about, you know, double entry accounting is, is double entry accounting goes back to Luca pacioli. There's your 1494 today. Oh my gosh. Yeah, Someone who knows Luca Pacioli. Perfect. The basics of accounting stay the same. But I'm wondering where you're going with this. Is it more like dimensions and tags or. I'm picturing it. You said you know, capital and human resources. I'm thinking about like project accounting and cost accounting. Like how did like what's the vision there? How does all this fit together?

Yeah, I'll talk about it a bit at a high level and then we can dive deep into whatever parts make sense. But just to get, keep everybody on the audience that, that might be less familiar with accounting. So your general ledger is like the backbone of accounting. It's, it's where you actually are rendering your journal entries to generate your financial statements here. And so I think that's fairly well understood. The history of, of the accounting software today that we have online is really anchored around a fixed chart of accounts that have labels in it or your journal entries and those render your sort of financial statements. But I think back to your point, if you thought about it a little bit more as like a horizontal data system in which you can have infinite tags on each money event. So just call them money events because they could be transactions that actually happened. I bought a thing with cash. It could be things that are going to potentially happen in the future. I like am anticipating this accrued expense that probably will happen or might not happen or like something that is likely to happen. I have an invoice or I have a receivable or payable that I'm going to put on a balance sheet. I'm going to pay it off. It's just a matter of when. Like if you're Apple, you're going to wait 90 to 120 days because well you can, but it still sits there. Okay. But if you were to think about what are the changes that you're going to make today, you would think about the architecture slightly differently so that you could have multiple views. So we started off with two things that were me the most frustrating was you have to pick cash or accrual books. But that's kind of crazy because I want both. Like I need my accrual books to see the financial health of company, but I need cash books because I'm doing budgeting and, and forecasting and like, you know, as much as I wish I can't make payroll on accruals, I got to do it with like actual cash. People like getting paid. Second is when you think about how we think about building trust in the ecosystem. There's a lot of really exciting stuff to learn from like the world of blockchain. And you could build a checksum, you could build an immutable ledger that's append only. So it gives you the flexibility to have as many different changes as you want, but you can never delete anything. Could you imagine how do you have trust in a system when you don't know who wrote in what. It's like Excel, like I can write something in Excel and then you can overwrite something in Excel and who knows what that cell came from, where it came from, what was the source. And then also you should have some enforcement mechanisms on accuracy, around supporting documentation, around pulling transactions directly from the source, trying to avoid and tag manual or AI generated content that can't be supported with evidence or all of this stuff brings together something where you have both generated content that can't be supported with evidence. All of this stuff brings together something where you have both trust and visibility into a single system and you're not limited. And then if you have all the underlying data and I'll be quiet because I can talk about this for years and I love it like you can then do all this non accounting related stuff. What's variance analysis? What is spending? How am I seeing revenue by customers or by product or by GL account? You shouldn't have to have these limitations in your accounting software. Just like you wouldn't have limitations on your pivots in Excel. Could you imagine Excel's like, you can only have one pivot. You have to choose it one time and you can't ever change it. You'd be like, sometimes I want to see my revenue by customer, sometimes by product, sometimes by jail account, sometimes by geography, sometimes by department. But like you can't really do that in your accounting software today, which is kind of crazy.

84% of leaders say analytics have underperformed expectations. 44% cite data quality as the biggest barrier to analytical success. 72% of sales orgs report forecast accuracy below 80%. Something has to give. That's where the analyst intelligence platform or AIP comes in. AIP is a data transformation notebook where users can write down their table tasks and the system actually does the work for them. AIP then makes your clean data reusable anywhere by creating a Google data mart and connecting to thousands of sources. A solid foundation is so critical that Daydream Business Intelligence offers discounts for users of AIP. AIP is trusted by portfolio companies of 20 off investment firms like Vista Softbank and Tiger Finvisor. A modern fractional finance team took reporting from days to minutes.

I'm thinking about, I mean just my, my CFO brain is going into overdrive here. I'm thinking about topside entries, you know, around an audit that you have to make. And the confusion around this is what the book said. But then we had to make this adjusting entry and just the, and then multiple people making entries to the GL and it's not happening. I mean you can go into the admin and see who made the entry or whatever, but it's a, it, it's work. It's not, it's not intuitive when you, when you make these entries. So that's, this is fascinating to me.

And it's pretty common across almost all other softwares where there's level, there's audit levels and there's transactional. You can like slice and dice the history of things and version controls and manipulate. You're not stuck to like one view of your data. Like, where else are you? Like, if you're doing product analytics or whatever type of other website analytics, you can't just like, oh, we get this one view and this one view only. Like, how can I possibly make an effective decision about improving a sales funnel or a conversion funnel or product engagement or tour if I'm only allowed to have like one view of the data analytics? Like, that's crazy. That just doesn't exist. So we're trying to bring a lot more of that visibility control and flexibility. Like if you're doing product analytics or whatever type of other website analytics, you can't just like, oh, we get this one view and this one view only. Like, how could I possibly make an effective decision about improving a sales funnel or a conversion funnel or product engagement or tour if I'm only allowed to have like one view of the data analytics? Like, that's crazy. That just doesn't exist. So we're trying to bring a lot more of that visibility control and flexibility to people so they, they can ask questions in multiple ways, not just the way that gap decides to do it in this one way. You have to do that too. So we do that. But like you shouldn't be limited to one view.

I love that the, the, the end statement. That's one I use a lot in Excel, you know. So you talked, you've Talked about a number of different things, and I think that event driven and the different approach to accounting and being able to look at a number of different things is really important. Excited to kind of see how that grows.

But another two people, so they, they can ask questions in multiple ways, not just the way that gap decides to do it in this one way, you have to do that too. So we do that. But like you shouldn't be limited to one view.

I love that the, the, the end statement, that's one I use a lot in Excel, you know. So you talked, you've talked about a number of different things, and I think that event driven and the different approach to accounting and being able to look at a number of different things is really important. Excited to kind of see how that grows. But another area talked about is improving the transparency, the security, those type of things. So are you using blockchain with that or how do you think about blockchain? Because I've heard a lot of, of that being a tool to use with accounting. Right. To be able to lock down those transactions and really have that history. So how do you think of that in all this?

Yeah, I mean, there's a fairly simple similar approach from kind of blockchain to sort of a data transformation like accounting. Accounting's effectively as a data transformation. Not, not to dumb down. It's, it's a lot of rules, it's really complicated. It's important. But at its core, if I think about it more like a modeler or a mathematician, there's a starting point and there's a set of transformations and there's an ending point. And blockchain builds the most trusted system that we've developed today. To order to go from like, here's the origination of a coin or whatever.

Sort of talked about is improving, you know, the transparency, the security, those type of things. So are you using blockchain with that or how do you think about blockchain? Because I've heard a lot of, of that being a tool to use with accounting. Right. To be able to lock down those transactions and really have that history. So how do you think of that in all this?

Yeah, I mean, there's a fairly simple similar approach from kind of blockchain to sort of a thing that you are tracking. And here's all the changes that have happened over time. In that case, it's a change of ownership. But at its core, like accounting is kind of the same thing. If we take a computer that you bought, like there was some transaction that generated into something that Generated into some treatment that generated into a slew of other components. And so in its simplest core for cash, you would maybe just trace that computer to like a computer expense, a data transformation, like accounting. Accounting effectively as a data transformation. Not, not to dumb down. It's, it's a lot of rules, it's really complicated, it's important. But at its core, if I think about it more like a modeler or a mathematician, there's a starting point and there's a set of transformations and there's an ending point. And blockchain builds the most trusted system that we've developed today to, in order to go from like here's the origination of a coin or whatever sort of. But obviously if you're going to depreciate it as a fixed asset, it goes to the balance sheet and then it goes, and then it gets depreciated sort of over a period of a couple of years or so. All of that is just a series of events of moving this money event through a system around. You should be able to trace that whole thing from the end state. Whether it's your fixed assets on your balance sheet or your computer expense on your profit loss statement. All of those sort of can trace all the way through. So if you thing that you are tracking and here's all the changes that have happened over time, in that case it's a change of ownership. But at its core, like accounting is kind of the same thing. If we take a computer that you bought, like there was some transaction that generated into something that generated that same level of transparency from the original source, who did everything along the way into some treatment that generated into a slew of other components. And so in its simplest core for cash, you, you would maybe just trace that computer to like a computer expense. Hey, whether it's software or a human, whether it's been split, whether it's on your cash books, whether it's on a balance sheet, when it transfers back over to expense, all of that is just a series of traceable events. So we brought that level of visibility and transparency to every single thing that happens in the system. And so that brings a level of like transparency where you can not only just, but obviously if you're going to depreciate it as a fixed asset, you, it goes to the balance sheet and then it goes, and then it gets depreciated sort of over a period of a couple of years or so. All of that is just a series of events of moving this money event through a system around. You should see and have a higher degree of trust of what your financial statements are today. You can trace it all the way back through the schedule down to the transaction everybody that touched on way. But something else that's really cool is you could also see the version of your financial statements at any point in time. So there's a state today. But you know, if you're looking back at 2023, let's say there's a state of what 2023 looks today, but there might have been a state that looked slightly different a couple months ago. Be able to trace that whole thing from the end state. Whether it's your fixed assets on your balance sheet or your computer expense on your profit and loss statement, all of those sort of can trace all the way through. So if you maybe had to make some adjusting entries or you know, something else came in that you have to modify, but you're now limited to like one view as opposed to like show me what this looked like and then show me the variance between these two things. Show me everything that just happened between these time periods. And so what that does is it just gives confidence again to like what happened, what changed? What's different? Did we expect it? And if not, how do we fix it?

So I feel like if I'm an.

Auditor, I brought that same level of transparency from the original source who did everything along the way. Whether it's software or, or a human, whether it's been split, whether it's on your cash books, whether it's on a balance sheet, when it transfers back over, all of that is just a series of traceable events. So we brought that level of this.

Because the, the tracking and the control and understanding or if I'm a company preparing for audit and it's, you're, you're trying to have your, your narrative of what happened here. And if, if an entry changed for some reason or something is off, that wasn't, you know, it looked one way in. Okay. And certainly you don't want, you're not going to go back and change locked periods, but things, things happen over visibility.

And transparency to every single thing that happens in the system. And so that brings a level of like transparency where you can not only.

Just time and being able to have this trail seems like that would be a very important piece of information for companies and for. And for auditors as well.

Yeah. See and have a higher degree of trust of what your financial statements are today. You can trace it all the way. We're excited. We're actually kicking off a few audits and that's part of our company goals for this year. So actually if any listeners are sitting on QuickBooks or Zero and like, oh my God, I got to prepare for my first audit, we would love to bring the power of 50 product managers, engineers, lots of formula back through the schedule, down to the transaction. Everybody that touched on way. But something else that's really cool is you could also see the version of your financial statements at any point in time. Public auditors, CPAs, tax lawyers, tax experts to the table to help you pass your audit. So there's a state today, but you know, if you're looking back at 2023, let's say there's a state of what 2023 looks today, but there might have been a state that looked slightly different a couple months ago for the first time because we think this could be really powerful. And the flip side is, again, as a CEO, you guys actually probably had to work in this and understand. All I saw from the CEO was a bunch of people dressed in suits in our worst conference room for months with a stressed out finance and accounting team going, oh my God, what are they going to ask for? What are they going to do? How are they going to do like, because maybe you had to make some adjusting entries or you know, something else came in that you, you have to modify but you're, this is a stressful time for everybody. And, and, and like, if we can bring the stress level down, you're now limited to like one view as opposed to like, show me what this looks like and then show me the variance between these two things. Show me everything that just happened between these time periods. And so what that does is it just gives confidence, then I think it has a better outcome. What we all want is we want accurate financial statements. We also want to review what are the controls and what are the risks in an area so much of the audit fees are caught up in. Sam's again, to like, what happened? What changed? What's different? Did we expect it? And if not, how do we fix it?

So I feel like if I'm an.

Auditor, I leveling and reviewing and doing stuff where software can help, that we think can help alleviate that a lot, that we can spend more of the time just like back to my favorite conversation with the finance team. Let's talk about how we improve the business. Let's talk about how we improve controls. Let's talk about how we improve risk.

To the business of this because the, the tracking and the control and understanding or if I'm a company preparing for audit and it's, you're, you're trying to have your, your narrative of what happened here.

Let's spend less time worrying about is this accurate or not. We should be able to improve that with software. Um, so that's kind of our goals for 2024. What year are we in? 2024. Yeah.

Want to kind of shift gears here a little bit and Glenn and I have a couple AI questions. I'll kick one of these off and then I'll give you a couple here to ask Glenn, but how do you.

Envision AI and if, if an entry changed for some reason or something is off that wasn't, you know, it looked one way in. Okay. And certainly you don't. You're not going to go back and change locked periods. But things, things happen over time and being able to have this trailer seems like that would be predictive analytics.

Changing finance. So not just accounting, but the broader finance in the next five to 10.

Years, I would add to that. I mean especially a very important piece of information for companies and for, and for auditors as well.

Yeah, we're excited. We're actually kicking off a few audits and that's part of our, with the.

With blockchain and with the traceability, you're adding a whole new level of data to, to the what's in the general ledger. So more data means more analytics and more and, and more everything that you can do around that goals for this year.

So actually if any listeners are sitting on QuickBooks or Zero and like, oh my God, I gotta prepare for my first audit. We would love to bring the power of 50 product managers, engineers, lots of former.

So I'm very curious as you guys are building this out, how, how you.

Envision, you know, public auditors, CPAs, tax lawyers, tax experts to the table to.

Help you pass your audit to what you're doing, but also maybe just the industry in general.

It's a fun topic of conversation because it's hard to think about how much, how fast AI is evolving and the capabilities get just mind blowing results. Every like six months something like insane is happening. So it's a really fun time for the first time because we think this could be really powerful. And the flip side is again, as a CEO, you guys actually probably had to work in this and understand all I saw from the CEO was a bunch of people to be learning about and playing with us. So it's hard to know all of the things. All dressed in suits in our worst conference room for months with a stressed out finance and accounting Team going, oh my God, what are they going to ask for? What are they going to do? How are they going to do like this is a stressful time for everybody and like if we can bring the stress level down. But I'll share a couple things that I'm growing in confidence are going to be changing. So as new companies emerge like Puzzle and there's others out there, actually the core underlying data architecture has to be built in a way that allows at least LLMs in the version of AI to be able to read and understand it. So aside from just the then I think it has a better outcome. What we all want is we want accurate financial statements. We also want to review what are the controls and what are the risks in an area. So much of the audit fees are caught up in Sam hallucinations. Just the retrieval of like which revenue are we talking about, which burling and reviewing and doing stuff where software can help that we think are we talking about which profit are we talking about? If these things aren't structured and labeled the right way internally, then it's hard to retrieve it. It's just like, you know, putting together, you know, one of these like things you buy on Amazon that comes with like random left hand drawn descriptions of like instructions on how to put together a table and you're like, man, this is even worse than ikea. Like I help alleviate that a lot that we can spend more of the time just like back to my favorite conversation with the finance team. Let's talk about how we improve the business. Let's talk about how we improve controls. Let's talk about how we improve risk to the business. How can I possibly do this? These instructions are terrible. So if we don't set the foundation of like rules, let's spend less time worrying about is this accurate or not. We should be able to improve that with software. So that's kind of our goals for 2024. What year we in 2024. Yeah. And visualization the right way. It's hard for AI to do anything really that impressive. So that's kind of the underlying architecture and database design. So things will be designed differently today. And I think that's going to be a challenge for a lot of the legacy companies because it's hard to re architect systems that already have lots of people on them. I think today there's a lot of opportunity.

Want to kind of shift gears here a little bit and Glenn and I have a couple AI questions. I'll kick one of these off and then I'll give you a Couple here to ask Glenn, but how do you envision AI and predictive analytics changing finance? So not just accounting, but the broader.

Opportunities to where AI can help improve accuracy. So that's a place where we really focus because we accounting and finance in.

The next, you know, five to 10.

Years I would add to that. I mean especially with the with blockchain and with the traceability you're adding a whole new.

So we sort of think about us. We're the data underlying your financial insights and modeling. And so we spend all of our time focusing on the accuracy and integrity of the data. And so when that happens that makes everything like garbage in, garbage out. We're solving the garbage in sort of problems. And then there's a lot of sort of helpful things where right now a.

Lot of level of data to what's in the general ledger.

So more data anomaly detection is done with either a set of very discrete software rules.

If this means more analytics and more and more everything that you can do around that. So I'm very curious as you guys are building this out how you envision.

You know than this and or humans and you only have so much time in a month or in a time period to like do all of the potential scenarios. But where AI can do is it can look at almost every type of change or anomaly in the system and highlight and say hey, this is something that you should look at now. So I think that's the state of where we are today. A lot of in our experience specific.

To what you're doing, but also maybe just the industry in general.

It's a fun topic of conversation because it's hard to think about how much how a lot of the finance or FPA's job is like doing a lot of one ad hoc requests. And that is just about there's a specific question, this person knows how to ask that question. They know what they want and the finance person has to then allocate this across all of their time and all the things they have to do and then go like retrieve a custom thing. And so if we can either allow the retrieval of data or, or in a permissioned way the sharing of certain data. Like the CTO should know what their budget is and they should know how they're spending against that budget and should need to ask for a bunch of analysis and have a 30 minute meeting along the way. I know that's just like one small example but we can connect people to data in a way that like you know, when you learn SQL it's easier to like ask and get answers to data. And if you learn one of these or more modern data visualization tools, you can kind of play around a little bit more. And I think AI is just going to remove the need to be technical but still get the answers you want because you know what you want to ask for. You just don't know what the answer is. And it's kind of gatekept in a secret place that the fast AI is evolving and the capabilities get just mind blowing results. Every like six months something like insane is happening. So it's a really fun time to be learning about and playing with this. So it's hard to know all of the things, but I'll share a couple things that I'm growing in confidence are going to be changing. So as new companies emerge like Puzzle and there's others out there actually the core finance team has access to.

Yeah, I love that because I've, you know, for years we've been talking about democratization of data, but to me, what AI does, and generative AI, specifically using LLMs, it's democratization of data science. Now you still have to know, you know, kind of, you have to know the questions to ask. Just like in, in finance and accounting you have to know the difference between EBITDA and net income or what, you know, there's components that you, you, you have to understand. But in the, the if Python and SQL were a barrier to entry to that exclusive club of data science, those now get knocked down when you can interact with the data through natural language. That's kind of what I'm, what I'm saying.

Yeah, so just the underlying data architecture, it has to be built in a way that allows at least LLMs in the version of AI to be able to read and understand it. So aside from just the hallucinations, just the retrieval of like which revenue are we talking about, which burn are we talking about, which profit are we talking about if these things aren't structured and labeled the right way? The easy application that we use in terms of sort of accounting to help make the core underlying value of data. And the example I like to give because it's fun and it happens, you know, all the time in a fast growing company full of engineers is, you know, when you buy something from Apple, you know what you bought from Apple and it might be a computer, but it might be a bunch of other things that aren't a computer. And if it's a computer and it's, you know, these current apples that are, you know, three, four, $5,000 you like you should, you have to depreciate them. So in order to know that you have to one provide the receipt. But you know, you know, not everybody always does or you have to have context because the thing that holds up your books closing a lot of times is these bookkeepers and accounts on the other end. Like I don't know what this Apple thing is and they got to trace down who the person is and ask them along the way. So if you could just let them respond to an email or a Slack message and say this is a computer internally then it's hard to retrieve it. It's just like you know, putting together you know, one of these like things you buy on Amazon that comes with like random left hand drawn descriptions of like instructions on how to put together a table. And you're like man, this is even worse than Ikea. Like how can I possibly do this? These instructions are terrible. So if we don't set the foundation of like rules and, and, and visualization the right way, it's hard for AI to do anything. You don't have to know anything about accounting, you just have to know that I know what it is. It's a computer. I write this computer. Oh, this is a. We just hired Sasha, he's a developer and he needed this like special computer that's like really expensive and so he bought a good. I know what it is. LLMs are really good job. Okay, well let's now look at the chart of accounts. What should this be? Now let's look at the accounting policies. Oh, it looks like this qualifies for depreciation policy. Let's move it over to the fixed asset register. Let's automatically depreciate it according to the policy of like a three year depreciation policy. No terminal value or whatever, you know, the company is for things and let's expense it every month. So just by this person saying this is really that impressive. So that's kind of the underlying architecture and database design. So things will be designed differently today. And I think that's going to be a challenge for a lot of the legacy companies because it's hard to re architect systems that already have lots of people on them. I think today there's a lot of opportunities to where I can help improve accuracy. So that's a place where we really focus because we're accounting and so we sort of think about us. We're the data underlying your financial insights and modeling. And so computer. We've now taken care of over three years of journal entries with the Words, this is a computer because of AI. I think that's a really powerful sort of example and it brings it to light in a really, really cool way. Fun way.

Yeah. And I think everyone right now, it's so, you know, machine learning has been used to, to great use from fraud detection to classification prediction. I mean, there's a million use cases for machine learning. Everyone's talking about generative AI because it's so accessible to the public. We're trying to explain a random forest classifier to someone who's not. And it's like, that's, you know, no one cares. But when you can type into a chat bot, you know, write a limerick.

About depreciation, we spend all of our time focusing on the accuracy and integrity of the data. And so when that happens, that makes everything like garbage in, garbage out. We're solving the garbage in sort of problems. And then there's a lot of, sort of helpful things where right now a lot of anomaly detection is done with either a set of very discrete software rules, if this, then this and, or humans and, and you only have so much time in a month or in.

A time period or whatever, and it spits out something. Data scientists and people who've been using machine learning and even some, you know, sort of hybrid finance people who've realized, oh wow, I can do more with a, with a Python program than I could in Excel. I know Paul May, Paul may disagree.

With that, but no, I wouldn't disagree.

So it's funny that I feel like the machine learning and the AI applications that were pre generative AI and that are, you know, more tested and true at this point are almost getting overlooked. But would you say, you know, you mentioned kind of the limitations of where we are with generative AI and I think they're being, we're overcoming limitations every day. If it's hallucination, we're addressing it through retrieval, augmented generation, and we're, you know, putting citations and whatever we're doing to address those. But is there anything like right now if I'm in finance and accounting, and I'm sure you guys are thinking about this as you develop out the product, but are there, are there like, do.

All of the potential scenarios, but where AI can do is it can look at almost every type of change or anomaly in the system and, and highlight and say, hey, this is something that.

You should look promising. AI and machine learning applications that are specific for finance people that they could.

Be looking at now. So I think that's the state of where we are today. A lot of, in our experience, a lot of the finance or FPA's job is like doing a lot of one.

Ad hoc at right now, today to help them in their job to improve their forecasting and decision making.

I have two answers. I have a biased answer called Puzzle IO and then I have a slightly, you know, less exciting but less biased answer as well, which is my core underlying belief is one of the things that's been holding the entire industry back is the lack of an underlying data platform that's built for finance native to AI. And so if I think about what a couple years from now looks like and then I'll bring it back to today, we think that there is an unbundling opportunity for the ERP I.e. design requests and that is just about. There's a specific question, this person knows how to ask that question. They know what they want and the finance person has to then allocate this across all of their time and all the things they have to do and then go like retrieve a custom thing. And so if we can either allow the retrieval of data or in a permissioned way the sharing of certain data like and in the world of AI.

So for our last segment, this is something we're trying out.

We were, the CTO should know what their budget is and they should know how they're spending against that budget and should.

Paul and I were spitballing right before the show on what we wanted to do for our, our, you know, different customer segment each time. So what we've done, we have a bank of lightning round kind of personal questions and rather than have sort of canned questions every episode, we thought well, let's put them into a database and we'll, we'll just generate, we'll pull one random but then we added another layer to it. So we're going to give you three.

Questions, need to ask for a bunch of analysis and have a 30 minute meeting along the way. I know that's just like one small example, but we can connect people to data in a way that like, you know, when you learn SQL it's easier to like ask and get answers to data. And if you learn one of these.

For each question before you hear them, you can either take one of our canned questions or the one our modern.

Data visualization tools, you can kind of play around a little bit more. And I think AI is just going to remove the need to be technical but still get the answers you want because you know what you want.

The LLM is going to generate for you.

Yes, I love it, that's what I was hoping for, who gets it? So, and I think we'll try ChatGPT. Who has, who has the harder question, the more fun question.

So this week we're using Gemini, and I don't know, I wonder if we should not expose that until after. But I'll go ahead and say we're using Gemini this week. We'll kind of roll through the LLMs and see what kind of questions that they come up with. So for, for question number one, would.

You rather ask for. You just don't know what the answer is, and it's kind of gatekept in a secret place that the finance team has access to.

Yeah, I love that because I've, you know, for years we've been talking about democratization of data, but to me, what AI does, and generative AI, specifically using LLMs, it's democratization of data science. Now, you still have to know, you know, kind of you have to know the questions to ask. Just like in, in finance and accounting, you have to take a random selection from our canned questions or see what Gemini comes up with.

Oh, I mean, I feel like I have to go with Gemini. No offense, but your first episode, I don't want to insult you guys in your inaugural episode, but like, let's go with Gemini.

Okay, so let's see. Let me push the magic button here and say. Okay, it's a. All right. So Gemini wants to know, how do you know the difference between EBITDA and net income or what? You know, there's components that you, you, you have to understand. But in the, the, if Python and SQL were a barrier to entry to that exclusive club of data science, those now get knocked down when you wind. How do you unwind or recharge after a long period of focused work?

Oh, I mean, that's a good one. I, I, we'd have to ask.

You can interact with the data through natural language. That's kind of what I'm, what I'm saying.

Yeah. So just think, Sasha, from 15 years ago, because I don't unwind. All I do is work, work, work, work, work. That's a long way. But if I were to have a time, I would say my favorite thing, if the easy application that we use in terms of sort of accounting to help make the core how to do anything would be to go skiing. Of course, you can't really do that after long days of work because I live in the Bay Area and be too long of a drive. I actually just, I, if I don't have Anything. I take a bike ride or I walk around the block and I know it sounds really cheesy, maybe this is just because I'm in San Francisco and that's kind of life. I actually take off my shoes and socks and I just stand in grass like at the park. There's just something about having underlying value of data. And the example I like to give because it's fun and it happens all the time in a fast growing company full of engineers is when you buy something from Apple, you know what you bought from Apple and it might be a computer, but it might be a bunch of other things that aren't a computer. And if it's a computer and it's these current apples that are 3, 4, $5,000 like you should, you have to depreciate them. Um, so in order to know ass, but maybe a bit of, of human and dog feces on your feet and standing there to really just like kind of connect. I just somehow it grounds me and I like it. And I'm that crazy person that people are like, who's, why is this person in the morning drinking or afternoon drinking coffee or tea, sitting, standing barefoot in a park with a bunch of stuff. But that's what I like to do.

I, I, I would add to that. You could take the advice that Bruce Willis got in Die Hard where you take your shoes and stuff that you.

You have to one provide the receipt but you know, you know, not everybody always does. Or you have to have context because the thing that holds up your books closing a lot of times is these bookkeepers and accounts on the other end. Like I don't know what this Apple thing is. And then you got to trace down who the person is and ask them along the way.

And you make fists in the grass and feces with your, with your, you.

Could just let them respond to an email or a slack message and say this is a computer. You don't have to do anything about accounting. You just have to know that I know what it is. It's a computer. I write, this is a computer. Oh, this is a feed.

I don't know if you remember from that movie. He wasn't in the grass. He was on the hotel carpet or whatever it was. But that may just be a new addition to help you relax even, even more. And see, we didn't even need AI for that. We just needed the magic of an 80s of an 80s action movie.

All right, we're going to do, do another question here. So do you want us to pick a random question or do you want AI to pick question number two?

Pick a. Now let's. Now let's go human form.

All right, so we use the random number generator. And here is the first one we had for you. This one's pretty straightforward. But what do you do outside of work? And I think you kind of answered that in the first one. All you do is work. So I might just pick a different question. Give me a number between one and eight.

Number seven. It's my old soccer jersey.

Do you.

Okay. And you can't answer your company to this. Okay, do you have.

We just hired Sasha. He's a developer and he needed this like special computer that's like really expensive. And so we bought a computer. I know what it is. LLMs are a really good job. Okay, well, let's now look at the chart of accounts. What should this be? Now let's look at the accounting policies. Oh, it looks like this qualifies for depreciation policy. Let's move it over to the fixed asset register. Let's automatically depreciate it according to the policy of Leica. Three year depreciation policy. No terminal value or whatever, you know, the company is for things. And that's expensive every month. So just by this person saying, this.

Is a favorite finance software app that you like to use.

Do I have a favorite finance software app? Well, we are like the Switzerland of finance software. We have to work and play well with everybody out there. But if I had to pick one, I would say one that I'm really excited about is Runway.com.

I knew you were gonna say that. I know seek you really well. I know you guys.

Computer. We've now taken care of over three years of journal entries with the words this is a computer because of AI. I think that's a really powerful sort of example and it brings it to light in a really, really cool way. Fun way.

Yeah. And I think everyone right now, it's so, you know, machine learning has been used to. To great use from fraud detection to classification prediction. I mean there's a million use cases for a machine work together.

Yeah, it's, it's so. It's so much of financial planning and analysis looks the same and Runway looks very different now. Whether that's good or bad and whether or not finance you want to adopt it, you know, tbd. I wish them obviously a lot of success as I do a lot of other people. But that feels like one of the most different pieces of finance software that I have seen in a long, long time. And that gets Me excited because I'm learning.

Everyone's talking about generative AI because it's so accessible to the public. We're trying to explain a random forest classifier to someone who's not. And it's like, that's, you know, no one cares. But when you can type into a chatbot, you know, write a limerick about depreciation or whatever, and it spits out something. Data scientists and people who've been using machine learning and even some, you know, sort of hybrid finance people who see.

If it works that love learning.

Love it. What do you think, Glenn? Should we give them one more? You want to do one more? We'll wrap up. Let's go. All right.

Yep, let's go. One more here. So human question or AI question?

Your question this time. Human.

All right, so what we've got for this one is what's the most surprising or unexpected outcome you've gotten from interacting with an LLM?

What's the most surprising? Let's see.

You are now out of time.

I, I, One of the things that I like to do, I guess we've sort of talked about this before, but that I like to do is surprising as a theme is taking sort of large sample data sets and just putting it in to sort of chatgpt. So we did a survey, for example. We surveyed a hundred and realized, oh.

Wow, I can do more with a Python program than I could in Excel. I know Paul May, Paul may disagree.

With that, but no, I wouldn't disagree.

So it's funny that I feel like the machine learning and the AI applications that were pre generative AI and that are, you know, more tested and true at this point are almost getting overlooked. But would you say, you know, you mentioned kind of the limitations of where.

We are, 50 CFOs about what bank they used and why they used it. And we got all sorts of crazy results, as you can imagine, of like, large, unstructured language. Like, my CEO just said, I have to do this and, but, but I don't really want to. And I had to fight with this along the way. And we just fed it all into Chat GPT and I said, I put it on one tab and I put what is the key themes in here? And then I did myself and was like, all right, let's go through and see what I can come up with.

Along our regenerative AI. And I think they're being, we're overcoming limitations every day. If it's hallucination, we're addressing it through retrieval, augmented generation, and we're you know, putting citations and whatever we're doing to address those. But is there anything like right now if I'm in finance and accounting, and I'm sure you guys are thinking about this as you develop out the product, but are there, are there promising AI and machine learning applications that are specific for finance people that they could be looking.

And at the end, I sort of shared it along the way and it found a couple, like, really interesting insights and themes that I missed and I didn't pick up on it. We're going to publish that hopefully soon and so we'll, we'll share that. But, but, yeah, but that, that was, I think the most fun was like, it beat me as it's not. That's, you know, a little humbling, but it was fun. It was fun to see.

Thank you for sharing that one. We totally enjoyed having here. Honestly, I think we could probably talk for another hour, but I'm not sure the guests would want to listen that long.

Right now today to help them in their job to improve their forecasting and decision making.

I have two answers. I have a biased answer called Puzzle IO and then I have a slightly, you know, less exciting but less biased answer as well, which is my core underlying belief is one of the things that's been holding the entire industry back is the lack of an underlying data platform that's built for finance. Native to AI.

We'll go ahead and end it there, but we just want to thank you so much for joining us, Sasha, being on the first episode. 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.

And so if I think about what a couple years from now looks like and then I'll bring it back to today. Hey, we think that there is an unbundling opportunity for the ERP that is designed in the world of AI.

So for our last segment, this is something we're trying out. We were, Paul and I were spitballing right before the show on what we wanted to do for our, our different customer segment each time. So what we've done, we have a bank of lightning round customers.