How Al and Automation Is Bridging The Accountant Shortage
Show Notes
TLDR: These show notes were optimized using Claude ChatGPT 4.0 and are designed to make you listen to the full episode. We also asked H4.0 to write a brief limerick summarizing the show notes.
There once was a podcast on AI,
Transforming accounting, oh my!
With Mitchell's insight,
Efficiency in sight,
A must-listen for those on the fly.
Show Notes:
Welcome to Future Finance, with your hosts Paul Barnhurst and Glenn Hopper, where we focus on the technological advancements and emerging trends fueling the FP&A professionals. We bring you vital updates and insights into the technologies that are going to shape not just finance and accounting, but every sector of the world in the years ahead.
In today’s episode, Paul and Glenn engage in a conversation with Mitchell Troyanovsky, Co-Founder of Basis, to discuss the transformative impact of AI on the accounting profession. The discussion highlights the differences between rule-based systems and generative AI.
Highlights of the Episode:
Generative AI vs. Rule-Based Systems: Generative AI offers contextual understanding and reasoning, unlike rule-based systems that rely on rigid if-then statements, making it more adaptable to complex accounting tasks.
AI as a Digital Assistant: AI can create first drafts of accounting tasks, enabling accountants to review and refine the work, thus saving time and increasing productivity. "You'll start having accountants much more of their job, rather than being this kind of monotonous manual work, start to be like, okay, this work has been done for me as a first draft, let me now review it."
Enhancing Strategic Value: AI acts as a digital assistant, supplementing the workforce rather than replacing it, and allowing accountants to provide more strategic value to their clients. "It is a tool for accountants. Explicitly, at the end of the day, the job that they can do the best is to advise their clients, help their clients, help them make better decisions, and be that partner, with them, rather than needing to be stuck in all the manual work needed to get to that actual real value add at the end of the day."
Shifting Roles and Skills: The introduction of AI in accounting will shift the focus from data entry and basic tasks to more analytical and advisory roles, making the profession more appealing to new graduates.
Addressing Trust Issues: While AI will make mistakes, just like human employees, it is essential to implement systems to catch errors and continuously improve AI performance through feedback and iteration.
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.
Learn more at Qflow.ai/future-finance
Follow Mitchell:
LinkedIn: https://www.linkedin.com/in/mitchelltroyanovsky/
Website: https://www.getbasis.ai/
Follow Glenn:
LinkedIn: https://www.linkedin.com/in/gbhopperiii
Follow Paul:
LinkedIn: https://www.linkedin.com/in/thefpandaguy
Follow QFlow.AI:
Website - https://qflow.ai/future-finance
In today's episode:
[01:32] Introduction
[06:35] Real-world applications of ChatGPT 4.0 in financial analysis
[08:50] Evolution of AI models and their impact on specialized financial tools
[11:23] Guest’s background
[13:43] Overview of Basis and the need for AI in accounting
[17:10] Differences between rule-based systems and generative AI
[20:22] How AI handles accounting workflows
[26:42] Addressing fears and trust issues with AI in accounting
[33:11] Fun rapid-fire Session
[37:40] Contact information for more learning opportunities
Full Show Transcript:
Host: Paul Barnhurst:: Welcome to the Future Finance show, where we talk about.
Host: Paul Robotic:: I, for one, welcome our robot overlords.
Host: Paul Barnhurst:: Future Finance is brought to you by QFlow.ai the Strategic Finance platform solving the toughest part of planning and analysis. B2B revenue-aligned sales, marketing, and finance seamlessly speed up decision-making and lock in accountability with QFlow.ai. This week. I have to get something off my chest. Here at Future Finance. We're huge fans of AI and technology, but we're also big fans of Excel. I'm tired of the BS and the hogwash that we're seeing on LinkedIn everywhere that AI has killed Excel. How many years have we heard Excel is dead? Here's the reality Microsoft is investing as much in AI as any company in the world. They've already brought AI into Excel. AI and generative AI will accelerate the ability to accomplish amazing things in Excel. So can everybody stop the clickbait and tell me Excel is dead? The AIs killed it, and can we stop the other extreme? We should use Excel for everything. No, we shouldn't. Excel is not a large-scale database. Yes. Can you use it as a database? Can you do just about anything in it? Yes, could you?, No. I think that's the balance we have to manage. AI is great. Generative AI is going to revolutionize so many areas, and it will make a lot of things easier in the spreadsheet, easier in Excel, easier in Google Sheets, and easier in so many tools.
Host: Paul Barnhurst:: I'm super excited about the technology. But let's keep things in balance here. We have a tool set. When I build a house, I don't just use a hammer, or just use a screwdriver, I use a set of tools. It's the same in finance. Spreadsheets are one of those tools. Generative AI is one of those tools we need to make sure we're not on the extremes. We're not thinking, AI, AI, AI, AI, Excel, Excel, Excel. We need to go beyond just Excel. I'm one of the biggest fans you'll find of Excel. But I recognize planning tools can also make a huge difference in our work, as can AI. My message is, "Keep using generative AI, keep finding ways to be more efficient, but don't sleep on the technology that's been around for 40 years." Excel and spreadsheets for even longer. Use AI to become more efficient in spreadsheets. Use other tools to help you to become better at your job, and focus on being well-rounded. Because as FP&A professionals, having a set of tools you can use will make you a better FP&A professional, using the right tools, understanding technology allows you to focus on what is important as a finance professional, and what's important is that we're good at our job, that we're making a difference to the bottom line, that we're helping scale and grow and do things that at the end of the day provide value for the business. If that means using Excel to build that model, great. If that means using generative AI to help you do it, great. So that's my monologue for today, what I wanted to talk about before we interview our guest and before Glenn shares the news story is, I just had to get that off my chest so I apologize for the rant, but I hope you understand the message. Excel is not dead. Spreadsheets aren't going anywhere. Neither is generative AI. Let's figure out how we can all be more productive together, and hopefully, this show will help you be more productive by listening to experts in this field and what they're doing to make our lives easier. I'm super excited about the future of finance and the role that generative AI, spreadsheets, and Excel will play in that future.
Host: Glenn Hopper:: Hey, there, and welcome back to Future Finance. I'm Glenn Hopper, the co-host of the show with my esteemed colleague and partner, Mr. Paul Barnhurst. This week, I want to talk for a minute about the release of ChatGPT 4.0. Maybe it's because everyone is focused on when 5 might drop, but I get a sense that a lot of people thought this release was a bit of a nothing burger. But after spending around 30 hours working with 4.0 Over the last week, I've got to say this release is a big deal. It's got multimodal integration, reasoning, and contextual understanding that will blow your mind. It's like having a Swiss Army knife, but instead of tools, it's packed with PhDs. In one case, last week, I used the tool to create some fairly complex scenario analysis based on three years of historical financial information. ChatGPT created a baseline forecast using a sophisticated statistical technique called Sarima, or Seasonal Autoregressive Integrated Moving Average. For those who aren't familiar, it did this in one shot. I then had it create three different scenarios based on price volatility, government regulations, and shifts in consumer demand. It created these variants, each with mild, moderate, and severe versions, and kept them in memory, which allowed for easy comparison across the models. Then I had it run several Monte Carlo simulations to stress test the scenarios.
Host: Glenn Hopper:: It performed this flawlessly, exhibiting not only better coding and more computing but significantly better reasoning than any other model I've used. I've seen some of the early benchmarking and it's performing pretty well. It demonstrated high accuracy in math, particularly with the chain of thought prompting where it matched or surpassed even Claude's three opus. In several cases, it did better than Gemini 1.5 Pro instruction following, and it's also looking better than Gemini at coding. These foundation models are getting so good, so fast that I'm seeing now that Sam Altman knew what he was talking about. Been several months now, I think, when he said, as AI systems get closer to AGI, we believe general-purpose models will have the ability to handle a wider range of tasks with higher accuracy and efficiency compared to narrowly focused models. The versatility and adaptability of these advanced models will surpass the capabilities of specialized models, ultimately making them more beneficial across various domains. Do you guys remember Bloomberg GPT? Bloomberg leaned in quickly to generative AI, and they built this tool over a year ago. It's a large language model specifically built to handle financial data and tasks. The project required a massive computational investment.
Host: Glenn Hopper:: It was something like 1.3 million hours of GPU time on Nvidia A100s. It was trained on a proprietary data set of more than 350 billion tokens, and it was augmented by a nearly equal amount from general data sets as well. The model is great at specialized financial tasks like sentiment analysis and named entity recognition, and it would outperform even the latest foundation models and specialized financial analysis. But I'm wondering how long can they maintain this lead. I think about what Altman said about these massive and now multimodal models and wonder whether the high cost and complexity of building Bloomberg, GPT, or similar models are justified for broader use cases for anyone beyond these tech giants. That got me thinking about all these custom GPTs that were built. Before GPT released the ability for users to create their own custom GPT, I was slaving away in front of a blinking cursor in Python, trying to fine-tune GPT 3.5 for finance-specific tasks, and I was using Rag to link relevant texts, and it took a long time. But then these GPT came along and anyone could do this without coding and in a fraction of the time. So I built a bunch of them. A whole army of finance and accounting chatbots is available from my website, robocfo.ai.
Host: Glenn Hopper:: But seeing where we are with GPT 4.0, I wonder if they're even needed. Don't get me wrong, there are still plenty of scenarios where specialized GPT is the way to go if you're dealing with proprietary or super-niche information, it's not out there for everyone to see. That's when techniques like Retrieval-Augmented Generation, or Rag and fine-tuning and building GPT all come into play. I'm thinking of things like company workflows, call center scripts, and maybe working with specific data sets. But for professional specialization, I don't know. I mean, you can say these generative models are generalists, but if you're a generalist who knows more about every specialization than the specialists themselves, it's like hiring a chef who can also fix your car and do your taxes. So if you haven't tried 4.0 Yet, I strongly encourage you to do so. Grab some financial statements and maybe the latest SEC filings from your favorite public company, load them into the tool, and then ask it the questions you'd ask your financial analyst, and then maybe fire your financial analyst. I'm joking, of course. It's not like GPT is going to be able to hold your feet while you do sit-ups.
Host: Paul Barnhurst:: 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. QFlow 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, closes the planning loop, creates air-tight alignment, improves decision latency, and ensures accountability across the team.
Host: Glenn Hopper:: Hi, I'm Glen Hopper, co-host of Future Finance with my esteemed colleague and broadcasting partner, Mr. Paul Barnhurst. Our guest on this week's show is Mitchell Troyanovsky. Mitchell is the co-founder of Basis, an AI platform for accounting firms. Mitch, welcome to the show.
Guest: Mitchell Troyanovsky:: Thanks, guys. Thanks for having me.
Host: Paul Barnhurst:: We're thrilled to have you. So I'll kick it off with the first question here. I'm sure Glenn will pipe in a minute and share some wisdom for all of us. So tell our audience about Basis. Maybe we'll start there. I know it's an AI platform for accounting firms, so tell us a little bit about Basis and why it's needed.
Guest: Mitchell Troyanovsky:: Absolutely. We strongly believe that accounting is not only one of the most important professions in the US for the functioning of the economy, but also one of the most under-resourced professions, especially, today. I mean, everyone always hears about less accountants going into the field coming out of school, everyone hears about all the retirees of firms, and you end up having this massive, massive shortage of accountants, which already goes on top of an existing shortage of accounting work needed to account for all the complexity. So what Basis is, is trying to leverage the absolute most cutting edge techniques in AI to start to take a lot of these manual existing processes and be able to have these digital assistants who can come in and start to take on parts of the work.
Host: Paul Barnhurst:: Got it. So the idea I hear is to have a tool that can supplement the shortage of people in many ways. Is that kind of the way to think of Basis to help augment the staff or do I think of the two working together?
Guest: Mitchell Troyanovsky:: Absolutely. I think it will differ by workflow, and also how good Basis gets over time, how much of things they can do end to end versus where it needs to be working with existing accountants. But I think it is a tool for accountants. Explicitly, at the end of the day, the job that they can do the best which is to advise their clients, help their clients, help them make better decisions, and be that partner, with them, rather than needing to be stuck in all the manual work needed to get to that actual real value add at the end of the day.
Host: Glenn Hopper:: That's a tall order. I mean, if you look at everything that happens in an accounting department or at an accounting firm, and I'm wondering when you set out when you were putting everything on the whiteboard and you were designing, what Basis does, did you have you always seen it as ultimately the goal is to be an end to end assistant? If so, maybe you originally envisioned it as a smaller product but then realized that there were other areas you could move in. How do you tackle something as big as providing AI assistance for accountants? What did you start with, what are you doing now, and what's on your roadmap with it?
Guest: Mitchell Troyanovsky:: So we think generally of Basis as this holistic intelligence that can take on sort of different workflows in the same way that if I were to hire outsourced help, let's say, I can go throw them at different workflows. They may not be amazing in all those workflows from day one. Of course, they have different skill sets, but broadly, they are sort of holistic in their ability. We want Basis to, in the same way, be this kind of holistic intelligence that you can take on different workflows. So when you think about trying to build something like that, you almost have to think, it sounds sort of funny, but about how you train accountants generally. That normally starts with debit and credits. You do that before you're doing full month closes and you do that before you're doing full, let's say, tax work, or one-day assurance work and things like that. So we very much started in the kind of outsourced accounting world, the casting world goes by different names. Primarily, because that was the area that is fastest growing in accounting firms. Two, the area that had, I think, the largest shortages, also the area in which the business model is not hourly. So people can get a much more pure margin. Then lastly, and most importantly from our perspective, in terms of building this vision, that's where you do the debit and credits. That's where you're quite entering things into the ledger. So if you start from the absolute foundation, it makes a lot more logical sense from an intelligence perspective to kind of build up the pyramid of skills that you would need to have that holistic intelligence.
Host: Paul Barnhurst:: Isn't that also where it's the most process-oriented when you start with debits and credits? Because the exceptions are always going to be the hardest to train, whether it's a human or a machine. Is that fair to say?
Guest: Mitchell Troyanovsky:: I think that is fair to say. But I would just add that. One of the big unlocks of AI today compared to what everyone's heard about ML for a decade now is that you have things that can be relatively generalizable and intelligent rather than just being, hey, here's this one exact workflow, and this is the only thing you can do. You're automating this workflow. It's taking on workflows that might have many, you can call them edge cases, but it's just there's so many. They're not even edge cases. That's just the work. Being able to automate that requires real intelligence. So even though debit and credits are very procedural, the reason you can't automate debit and credits with like a rule or with an if statement is because it isn't actually like binary. It's very intelligent and kind of artful almost.
Host: Glenn Hopper:: So I talk about this all the time. I think you mentioned machine learning which we've been using for over a decade now, and finance has been in fraud detection and machine learning is from product recommendations to spam filters and all that. So it's tested technology. People have been using that for years I think for finance and accounting until generative AI came along and there isn't a whole lot of, you know, we're still early in it. So there's not a lot of broad adoption for it yet. But I think one thing that people have to get their head around is that I know a lot of people use RPA and, robotic process automation for anyone who hasn't used that before, and RPA is very much rule-based, very difficult to set up, it's very picky on how things go with generative AI, you're not just getting look here for this to find out what the sales tax is, you're getting actual contextual understanding. For people who are not familiar with what's going on with generative AI, how would you describe the difference between a rule-based system like you just alluded to versus what generative AI and these LLMS are able to do?
Guest: Mitchell Troyanovsky:: It's a great question. I think one of the reasons it's hard for people to conceptualize this is, I don't like the term generative AI personally, because I think it makes a lot of people focus on the generation aspect of things, and that is obviously powerful and important, but it sort of hides the true power under that, which is this ability to reason over context and make decisions. That is on its own the kind of true power of it and in the same way that, if I have an if statement, let's say, and I have five if statements and I run everything through those if statements, those if statements can triage between the various outcomes of that kind of like list of rules. If someone you know has had to manage a thousand rules in a QuickBooks file, they're very familiar with this, and you end up having all these exceptions and things don't work well. The thing was capital versus not, there's no reasoning. There's no intelligence inherently. It's like you'd look at that as somebody looking at the rules and you as someone who is intelligent, can see that, oh, this is ridiculous. Of course, you would never put a $5,000 Uber expense into a taxi ride because there's no way a taxi ride ever costs $5,000. But like, you had a rule that said if Uber then do a taxi. So the rule itself isn't smart and I don't want to anthropomorphize too much. But I do think the easiest way to understand how intelligence works is to think of it like a person a little bit, where if you work with a person they may not do the if statement in the same, perfect way, but they're going to understand it much better. That allows you to have nuance. So that's, I think, the kind of big difference.
Host: Glenn Hopper:: You mentioned anthropomorphizing and I know everyone says don't do that, but you and I have talked about this before, and the example I always use is whenever you're prompting these things, treat them as if they're a very bright but very green intern. So when I talk to them give them very clear instructions and walk them through. I'm, I'm wondering if as you fine-tune and train, I know you're not building your own foundation models, but as you get the prompts right and get the workflows right and everything, how do you treat the technology and how do you sort of guide what you're trying to make it do and what you're delivering to your customers?
Guest: Mitchell Troyanovsky:: I think a lot of it goes into techniques like how you fine-tune, and making sure you're doing things very specific. One thing we're very big on is doing things extremely locally, in the sense that, the model and the workflow for one client is very client-specific. It's not being trained across clients for many reasons. But I think one of the big ones that people sometimes don't fully appreciate is efficacy. The way one transaction is handled for a client, the way one client does a close, or how all this works is different than it is for other clients. You have all this variation, both like how accounting should be done, across the chart of accounts, what the policies are, there's tons and tons of variation across clients. If you're just thinking, oh, I'm going to train a model that knows that like taxi equals Uber, then you're sort of missing the force from the trees, because you're just looking at things at a very high level, not at the client level, which is what's useful and what's important.
Guest: Mitchell Troyanovsky:: Then in terms of anthropomorphizing and how to think of it, I think the most important thing as people get more comfortable working with these models is trying to treat it more like that intern, or maybe like an open-eyed new hire or something like that, where they're smart and they have an enormous amount of potential, and they are extremely valuable. But it doesn't mean that they're always going to get every single little perfect thing. You need to work with them to give them feedback, and iterate on how you work with them and build a rapport, as to what things they can do and what things they can't do. You'll get that intern who will surprise you with how good they are at certain tasks that you didn't expect. Then there are other tasks where you're not there. So you need to build out that rapport when you kind of integrate this into your workflows.
Host: Paul Barnhurst:: 84% of leaders say analytics have underperformed expectations. 44% cite data quality as the biggest barrier to analytical success. 72% of sales orders 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 tasks, and the system 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 top investment firms like Vista, SoftBank, and Tiger. Fin visor, a modern fractional finance team, took reporting from days to minutes. This leads me to a question, as you mentioned, where do you think we're at today? How much of the workflow are you seeing the tool able to do? Maybe the typical accounting setting and I get it's local, every company is different. But what do you feel like they can do today and what percentage or how do you envision it 3 to 5 years down the road? Because I know we're early. Ai is good at a lot of things. It has room to improve. So where do you think you are on that journey? What do you think it can do today and how do you think about it, and maybe 18 months, 3 years down the road?
Guest: Mitchell Troyanovsky:: The wide majority of tasks that happen today inside, let's say, doing the outsourced accounting kind of work, You will start to have over the next 18 months, for sure, AI doing the wide majority of those, that doesn't mean they are now doing them fully autonomously or that they are in some ways replacing accountants, but they are not. But what it means is you'll start having accountants much more of their job, will stop from going like, okay, I have to do this work and I'm going to start from scratch and do it, okay, this work has been done for me as a first draft, let me now review it and let me now understand. Then as people kind of get up level like you'll have junior people who a larger percentage of their work, rather than being this kind of monotonous manual work where start to be like, okay, we need to think about how the results of this impact the client. Let's write some advice. Let's look into projections and start thinking about what sort of decisions will this client need to make based on this information. Things of that, things that are much more strategic, things that add a lot of value to the client. So I think not only will people save a lot of time because you'll be able to have AIs come in and do a lot of very base-level work rather than doing it from scratch. But also accountants will be able to start adding a ton more value to their clients because they'll have more capacity to do that kind of high-value work, especially, if the low-level work that is normally blocking that starts to be done much faster in your closes start speeding up and you start having a lot of the transactions in at a much more timely fashion, etc.
Host: Glenn Hopper:: I think, talking about replacing that low-level work, one of the things that, we talked earlier about is the accounting shortage, and that increases the need for products like Basis. I know nobody goes and gets a degree or a master's degree in accounting because they're perfect for finance because they're excited to go do data entry. But that's kind of where everybody starts at that low level. So I'm wondering and also if fewer people are going into accounting right now it may be that the appeal of that sort of entry-level work doesn't sound very great. So in some ways, we're making the work more attractive, that it's less of that mindless and more mindful work out of the gate. But the other side of that coin, I guess, is it's an old paradigm that you have to have that level of experience to fundamentally understand it. So it's like you're immediately a fresh grad out of school, instead of coming in and kind of learning the ropes, you're immediately going to be expected to add greater value. Do you think that's a new challenge for recent grads, or do you think it makes the job more appealing to them? I don't know if that's something you guys have. That's a great question to be considered or talked about.
Guest: Mitchell Troyanovsky:: No, it's a good question. I think in the new world, there will be lots more skill sets that we don't think about today that will start to become valuable. You'll have to learn coming in. So for example, how to work with these AIs is something that people don't even think about as something you'll have to learn, but people will have to learn. How to work with it, when to trust it, how to set up good task sets, how to build rapport, and that will be a skill in the same way that writing Excel formulas is a skill. Do you learn to write Excel formulas when you first join the job or do you learn it like in school, before? It kind of depends on the program, and I think you'll have some similar dynamics there. I also think in the same way, that most junior employees 30 years ago would read a bank statement and key in amount values, today, they don't have to do that. You don't need to do that, bottom-level work to understand how accounting works and how the client works and things like that. If you are looking at outputs and helping review proposed journal entries and you're understanding how the debit and credits fit in for this specific balance sheet account. What happens normally is you still are learning a lot of that work, even if you may not be the one keying it in. I think you'll see a similar shift to what happened when you started having cloud-based software and you didn't have to do as much like pure data entry.
Host: Paul Barnhurst:: It'll be interesting to watch. I do share Glenn's concern a little bit, that there's something to be said when you've done a process a lot of times you start to learn it versus just reviewing it. I sometimes wonder, just AI in general, not just accounting, but other people are going to lose some of that. I think everybody worried about that with the calculator and other things. We'll figure it out at the end of the day. But it'll be interesting to watch. I'll be that old person complaining when I used to have to do it, although I was never an accountant, so I can't claim that one. I did process journal entries, but I have a finance degree, not an accounting degree. But it will be interesting to watch because I see both sides of that. I agree with you. We've always figured it out. We'll figure this out, but I do worry that there may be some stuff that's lost from going through that process, some experience that they may lose that we'll have to address.
Guest: Mitchell Troyanovsky:: You didn't think that if initially, your job was on manual spreadsheets to do the math, you would lose experience in adding up the numbers, but you gain the need to learn how to use Excel. So I think you'll have very similar dynamics.
Host: Paul Barnhurst:: There's a lot to be said for that. There are trade-offs. You learn some new things, you might not pick up other things.
Host: Glenn Hopper:: I know we're very excited to get to the future finance bag of fun questions, but one, kind of going on that while we're staying serious for a minute, I run into and I do a lot of speaking to finance and accounting groups and it's interesting to watch reactions in the room when I show them a demo of what you can do with generative AI and show them sort of these agentic workflows where we can replace manual processes. Maybe it's growing a little bit, but not much. There's fear of missing out and then there's fear of the technology itself. Then there's fear of what happens to me when AI gets more and more powerful. You don't even have to have AGI. You just need increasingly, incrementally better models than what we have now. So you encounter the same, I imagine if you're going into a client at an accounting firm, there are people who are scared or distrustful of using AI. So how do you assuage their fears? How do you talk to these people or maybe if there still is a negative now, but what is your interaction with people when they tell you I'm scared to death, either that I'm going to lose my job because of AI or because I don't trust the output from it?
Guest: Mitchell Troyanovsky:: Absolutely. Maybe I'll start with the output and then I'll talk about the larger questions. So on the output, honestly, I think the most useful thing is here to get people to click it is would you like if you could hire, a thousand master new hires, let's say, who you didn't have to manage, who did work exactly the standardized way you had it, and you could do that effectively for no cost would you trust the output for every single one of them? And the answer was obviously no. You don't trust necessarily new hires out, but you do necessarily trust if you go and outsource the workflow the output of someone overseas. You may not trust your 20 years of experienced tax partners' output, depending on the workflow, you may have someone review what happens. So the reason that you hire people in general is because people are good at doing very intelligent tasks, but they make mistakes. Them making mistakes doesn't mean you don't hire people, and it just means you need to put in place systems to catch those mistakes, to make sure that the most important workflows, you have the most reviews to make sure you set all those things up. So I think broadly, once you kind of start to think of the new paradigm in AI, in that kind of mindset that it's extremely valuable, but it will make mistakes and it will.
Guest: Mitchell Troyanovsky:: If someone comes and tells you, hey, this thing will never make mistakes ever. That's just not possible. That's not the expectation I would want to put on somebody. So that's the thing for one is just helping them understand that it is this thing that is almost closer to human than a rule. But then on the larger thing, okay, if it's going to make mistakes, how do I know when to trust it? The key is it can't make dumb mistakes and it can't make mistakes all the time. It has to make mistakes a very small percentage of the time and mistakes that are predictable. So it's just more like the proof is in the pudding. If you try it and it gets everything wrong, of course, you're not going to use it. You need to be able to hold to certain benchmarks and you need to do the work, and if it doesn't, just like if you hired a bad employee, you would fire that employee, you would fire Basis. So I think in terms of convincing people that it works, they just have to try it, and we can do case studies and things like that to help with that. But it's so new that at the end of the day, it's tough to fully get it until you use it and see the outputs.
Host: Paul Barnhurst:: Thank you for that answer. We're going to move into our fun section. This is kind of the surprise section. So we use ChatGPT to generate 29 different questions to ask you. Some of these are getting to know you. Some of these might be future predictions, just different categories. We used it to ask. So we're going to use a random number generator for the first one. I'm just going to go ahead and randomly we'll let you pick on the next one. So give me a moment here I'm on my random generator. You get question number 22. Here we go. Are you ready?
Guest: Mitchell Troyanovsky:: What's it?
Host: Paul Barnhurst:: What's your favorite finance-related joke or meme?
Guest: Mitchell Troyanovsky:: I'm a big Matt Levine reader. I don't know if other people, who are listening or you guys are big Matt Levine readers. My favorite running joke is that everything is securities fraud and that anytime any public company does anything wrong even if it has nothing to do at all with the security, they'll get sued for securities fraud because they didn't disclose whatever random thing it is.
Host: Glenn Hopper:: I've got to say with these questions. So we built a GPT that generates questions, and we just were testing it out and we thought, well, we don't need to generate them live. We've got a bunch of them here. So we included them. But last week or on the last episode, we used Gemini and tried to do it live. The questions were a lot of duds, I don't know where that is on the benchmark of being able to create interesting questions, but I was happier with what we got from ChatGPT, and in future episodes, we may do it live. But there were a few duds in here. So there was some editing, but we thought maybe since we're a future finance show, we should use technology to come up with these. All right, so let me see. Let me spin the wheel here and see what we could do for our next question. Let's see. If I could take over one of your daily tasks, which one would you happily give up?
Guest: Mitchell Troyanovsky:: Oh, my God, I'm currently in charge of stocking the fridge of spin trips and the boxes where the spin trips come in from Amazon are pretty heavy. They're very cheap, and we drink a lot of them. So if we get one of those Tesla robots that come in and start unboxing boxes to put in the fridge, that'd be nice.
Host: Paul Barnhurst:: I like it. When you get that, please, put a photo, send us a photo. We'll put it out there to let everybody know we're taking over your favorite task, or least favorite, I should say. This time, I'm just going to let you give me a number. I have 29 here on my list. Pick a number between 1 and 29. So this will be your turn to randomly pick.
Guest: Mitchell Troyanovsky:: Let's do 11.
Host: Paul Barnhurst:: All right. Give me one second here. Let me go up toward the top. Number 11. I like this one. If you had to explain blockchain to a five-year-old, how would you do it?
Host: Glenn Hopper:: Are you a Web 3.0 guy, Mitch?
Guest: Mitchell Troyanovsky:: I'm well aware. I would tell them that it is like that board game that we always play called Monopoly and just imagine that every time we did a transaction, like I bought something for you or whatever we wrote that down on a piece of paper, and then we always wrote that down. But this piece of paper was like sacred, stored as if it's like the Constitution in Washington DC, and nobody can touch it and it always stays there. just like Monopoly, the money is fake.
Host: Glenn Hopper:: Well, wow, that was well done, sir.
Host: Paul Barnhurst:: Love it. We had sacred, the Constitution, fake, and Monopoly for blockchain. Well done. I think, on that note, that's a good note for us to close here. Just kind of wrap-up. We hope you had a little fun there at the end with our random questions. Of course, we have to keep it, we have to bring technology into everything we do. We're excited to see where Basis goes. We wish you the best of luck because we know there's a shortage that's not going away in accounting. We know technology needs to help fill that gap. So we, here, at Future Finance, are rooting for you. We'll just give you kind of a last word. If someone wants to get a hold of you or anything you'd like to kind of share with our audience before we wrap up, it's all yours.
Guest: Mitchell Troyanovsky:: No, I appreciate it. If you run an accounting firm and you have a cast division or outsource accounting division, and you want to hear a little bit about what we do, we're still small and growing. But please reach out. You can find my email on the website or some of the contact info, and we're constantly working with more and more firms, especially, those who are forward-looking about how technology is going to change the profession in the nature of this podcast, of course.
Host: Glenn Hopper:: Perfect. We'll put all that in the show notes, too, so everybody will be able to reach out to you.
Host: Paul Barnhurst:: Thank you so much for joining us. We enjoyed having you and love learning a little bit more about what you're doing.
Guest: Mitchell Troyanovsky:: Awesome. Thanks, guys.
Host: Glenn Hopper:: Thanks, Mitch.
Host: Paul Barnhurst:: 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.
