Latest / The Adults In The AI Room / Earley AI Podcast - Episode 98 Agentic AI in Finance, the Trusted Advisor Advantage, and the Pricing Model Reckoning with Nikita Komarov

Transcript

Earley AI Podcast - Episode 98 Agentic AI in Finance, the Trusted Advisor Advantage, and the Pricing Model Reckoning with Nikita Komarov · The Adults In The AI Room

• 0:28
  SPEAKER_01: Welcome to the Early AI Podcast.
• 0:30
  SPEAKER_01: I'm Steph Early, your host, and we talk every episode about how AI and data is transforming business operations and business strategy throughout the industry and across industries.
• 0:41
  SPEAKER_01: Today we're talking about what happens when you take one of the most time-intensive high-stakes financial analysis work, the kind of work that would take weeks and weeks in a team of CPAs and rebuild it from the ground up as an agentic AI system, financial due diligence, internal audit, vendor overpayment recovery, not assisted by AAI, but done by AI, with humans overseeing the output.
• 1:03
  SPEAKER_01: Joining me today is Nikita Konarov, CEO and founder of Dobbs AI.
• 1:07
  SPEAKER_01: Nikita spent seven years at McKinse advising Fortune 1000 executives on digital transformation and worked with private equity on commercial due diligence and portfolio transformation before founding Dobbs.
• 1:18
  SPEAKER_01: His company serves uh Fortune 1000 enterprise clients as well as top accounting firms looking to automate their core client services.
• 1:26
  SPEAKER_01: Nikita, welcome to the show.
• 1:27
  SPEAKER_01: Thank you, sir, for hosting me.
• 1:29
  SPEAKER_01: Excited for this.
• 1:29
  SPEAKER_01: That's great.
• 1:30
  SPEAKER_01: So let's talk about common misconceptions about AI in finance and accounting.
• 1:35
  SPEAKER_01: When you talk to CFOs and accounting firm partners and financial executives about AI, what did they most consistently get wrong?
• 1:42
  SPEAKER_00: So I would say the financial folks, they are one of the most conservative in the market.
• 1:48
  SPEAKER_00: And I understand why.
• 1:49
  SPEAKER_00: I've been like in their shoes, I'm accounting auditor by training.
• 1:51
  SPEAKER_00: You work with the client data, you need to have a number should be 100% precisely.
• 1:56
  SPEAKER_00: You don't want a lot of creative accounting.
• 1:58
  SPEAKER_00: Yeah.
• 1:58
  SPEAKER_00: Or innovative accounting, right?
• 2:00
  SPEAKER_00: So if you talk about AI specifically, what is a AI?
• 2:04
  SPEAKER_00: That's a large language module.
• 2:06
  SPEAKER_00: How they work.
• 2:07
  SPEAKER_00: They predict the next word.
• 2:09
  SPEAKER_00: That's why these systems they are non-deterministic.
• 2:13
  SPEAKER_00: You can't say what would be like the output next.
• 2:15
  SPEAKER_00: It's kind of black box.
• 2:16
  SPEAKER_00: That's a problem with the financial services.
• 2:18
  SPEAKER_00: You need to have 100% accuracy, but you don't know what's this system gonna tell you.
• 2:23
  SPEAKER_00: It can hallucinate.
• 2:24
  SPEAKER_00: And if you will see every system like hallucinates, every like AI solution, and especially like all our experience with the last couple of years, was okay, it's helpful, but sometimes it provides you something, so you can't rely on it 100%.
• 2:37
  SPEAKER_00: That's one of the biggest concerns of the financial professionals on how to leverage AI in their work.
• 2:42
  SPEAKER_00: And I would say you it's just about the way how you use AI.
• 2:50
  SPEAKER_00: If you use like code, copilot, chat GPT, and others, you will see like time to time these hallucinations.
• 2:56
  SPEAKER_00: But to build a really production-ready system that you can rely on, it's much more complicated.
• 3:02
  SPEAKER_00: And that's what we are focusing on.
• 3:04
  SPEAKER_00: We are making sure that we can take the flexibility of AI systems, but at the same time provide the accuracy that you need so you can trust this number.
• 3:12
  SPEAKER_01: So how do you do that?
• 3:13
  SPEAKER_01: How do you put guardrails around it?
• 3:15
  SPEAKER_01: If it's probabilistic and not deterministic, uh and you need deterministic answers, financial systems uh you know require the numbers to add up, right?
• 3:23
  SPEAKER_01: And sometimes these large language models are not very good at math.
• 3:26
  SPEAKER_01: So, how do you put guardrails around that?
• 3:28
  SPEAKER_01: And how do you take something that's inherently statistical and non-deterministic and make it more bulletproof from the perspective of financial audit and financial controls and so on?
• 3:37
  SPEAKER_00: I would say that's our superpower, and that's where we spend like a lot of time like building IP around.
• 3:43
  SPEAKER_00: So several things.
• 3:45
  SPEAKER_00: One, we orchestrate teams of agents.
• 3:48
  SPEAKER_00: For example, think about financial due diligence.
• 3:51
  SPEAKER_00: You typically have like a director, manager, like analysts that are running it.
• 3:55
  SPEAKER_00: So we reproduce this, but like in the IHS.
• 3:58
  SPEAKER_00: You have a director who oversees like the scope, the core analysis that should be run.
• 4:02
  SPEAKER_00: You have a manager who decompose the task, you have an analyst who are running them, and you can run like 50, 60, 100 cycles until the output is ready for the human review.
• 4:12
  SPEAKER_00: That's one of the ways, just keeping a lot of the iterations and burn a lot of compute.
• 4:17
  SPEAKER_00: The second is the controls and guardrails, and that's the same how I would work with my team.
• 4:22
  SPEAKER_00: For example, I'm the manager, I have an analyst, I have a set of the patterns how I would review their work.
• 4:27
  SPEAKER_00: That the numbers should reconcile here, the total should be like I know zero here.
• 4:32
  SPEAKER_00: Um I just put in a set of the controls and guardrails that I give to the agents to follow.
• 4:37
  SPEAKER_00: And that that's another way.
• 4:38
  SPEAKER_00: The third, a lot of the problems in the financial space is the quality of data.
• 4:44
  SPEAKER_00: Because garbage in, garbage out, and especially in the financial deal, is just every deal is unique.
• 4:50
  SPEAKER_00: You just can get it like a messy files, and we spend a lot of time on building the models on the data extraction.
• 4:57
  SPEAKER_00: How to digest like 500 pages of PDF bank statements that like were scanned, or like some millions of rows of general ledger in TXT format from some legacy system.
• 5:08
  SPEAKER_00: Yesterday we received like 500 megabytes of like sales data from Amazon, and all these kind of use cases, you just need to know how to extract this information correctly and measure accuracy of the extraction.
• 5:20
  SPEAKER_00: Because if you just put the PDF file in the cloud, it will extract some numbers, but you will never understand the accuracy of it.
• 5:27
  SPEAKER_00: And we measure this specifically.
• 5:29
  SPEAKER_00: So I would say like this three levers has some of the like important components.
• 5:34
  SPEAKER_01: Yeah.
• 5:34
  SPEAKER_01: So so when you have established processes and you have rules around those processes and you know what the outcome should be, so you know the the the accounts should balance, right?
• 5:45
  SPEAKER_01: You should get you should not have variations, right?
• 5:47
  SPEAKER_01: You need you're looking at things that are verifiable, right, from a simple accounting perspective, right?
• 5:54
  SPEAKER_01: So you're not asking it to make predictions necessarily about certain processes, you're simply automating those processes because you can define them very carefully and you can define them.
• 6:05
  SPEAKER_01: There's going to be a clear, unambiguous outcome.
• 6:08
  SPEAKER_01: And I think that's one of the things that organizations need to think about when they're building agentic capabilities.
• 6:13
  SPEAKER_01: It's understanding that that process, it's understanding that workflow, right?
• 6:17
  SPEAKER_01: Because if you don't understand, I like to say you can't automate what you don't understand, right?
• 6:22
  SPEAKER_01: And so the first thing you have to do is say, well, what is the expected output and the outcome?
• 6:27
  SPEAKER_01: And then how can I make sure that I'm verifying that I get to that outcome?
• 6:31
  SPEAKER_01: So it sounds like those are the principles that you're applying by understanding, you know, exactly what that audit process consists of or what that due diligence consists of.
• 6:40
  SPEAKER_01: And as you say, you have agents that are that are that are fulfilling specific roles to monitor the output of other agents.
• 6:49
  SPEAKER_00: Yes, that that's the one piece, and another piece are specific tools.
• 6:52
  SPEAKER_01: Yeah, okay, go ahead.
• 6:53
  SPEAKER_01: Say more about specific tools.
• 6:55
  SPEAKER_00: Specific tools, it means large language models itself, it's a good reasoning layer.
• 6:59
  SPEAKER_00: So they can orchestrate the other agents, they can make it like some judgment and decisions, they can read the context.
• 7:05
  SPEAKER_00: But for like the precise math calculations, you need to have a code that will execute the like the numbers, the math itself.
• 7:12
  SPEAKER_00: Or for example, most of our analysis, they are Excel-based formula.
• 7:16
  SPEAKER_00: So you can trace every number, and when you are building the formulas, it verifies itself that the numbers are like correctly calculated.
• 7:24
  SPEAKER_00: So understanding the process, picking the right tools for every process step, and then yeah, as you started saying, uh like having in the human loop for the critical controls and checks, they are working a lot with accounting firms, and they have in their stamp before can I results to the clients, manage client relationships like hand coding and like some critical judgments.
• 7:46
  SPEAKER_01: Right.
• 7:46
  SPEAKER_01: I I I remember hiring a fractional CFO years ago, and one of his first deliverables was PL statement, right, for the quarter.
• 7:55
  SPEAKER_01: And I remember opening it up and looking at it and going, okay, that's wrong.
• 7:59
  SPEAKER_01: That's wrong, that's wrong, that's wrong, right?
• 8:01
  SPEAKER_01: And I immediately saw these errors, and I am not a finance person, right?
• 8:05
  SPEAKER_01: So I'm not somebody who has, I mean, I I I went to graduate school for business, but I'm by no means a financial expert, or you know.
• 8:14
  SPEAKER_01: And and I said to him, What happened here?
• 8:16
  SPEAKER_01: And he goes, Oh, I didn't look at it that closely.
• 8:18
  SPEAKER_01: I'm like, What?
• 8:19
  SPEAKER_01: What then?
• 8:21
  SPEAKER_01: This is your job, right?
• 8:22
  SPEAKER_01: This is your job.
• 8:23
  SPEAKER_01: Like, why did I hire you if you didn't look at it that closely?
• 8:27
  SPEAKER_01: This is your first deliverable to me, and you did not look at it that closely.
• 8:31
  SPEAKER_01: When I could open it up and see those immediate mistakes.
• 8:33
  SPEAKER_01: So I think the human judgment is critical, especially if you if you look at it carefully, right?
• 8:39
  SPEAKER_01: So, so the role of the human is even more important.
• 8:42
  SPEAKER_01: And I think people you know underestimate or overestimate the capabilities of the agents.
• 8:47
  SPEAKER_01: I I'll I'll tell you, I had an experience very recently where I what I'm doing is I'm codifying our uh consulting practices, right?
• 8:55
  SPEAKER_01: So we have our methodologies and our frameworks, and I'm I'm building these tools to uh go through and you know process interviews and to apply you know heuristics and to do evaluations based on based on certain inputs.
• 9:10
  SPEAKER_01: And what I what one of my tools, one of the LLMs, said, Hey, you have a problem with the documents, they're not aligned, right?
• 9:18
  SPEAKER_01: So I have a provisioning folder with with knowledge artifacts that I that go into a project.
• 9:22
  SPEAKER_01: So the project has the right context.
• 9:24
  SPEAKER_01: And I said, okay, well, and it gives me this long, long textual explanation.
• 9:27
  SPEAKER_01: I'm like, well, okay, what's the bottom line?
• 9:29
  SPEAKER_01: What do you have to do?
• 9:30
  SPEAKER_01: If and they said, well, I need to do blah, blah, blah.
• 9:32
  SPEAKER_01: And I said, okay, just fix it, right?
• 9:33
  SPEAKER_01: No, I didn't look at it as carefully as I needed to, but I said, fix it.
• 9:37
  SPEAKER_01: So it came back and it and it went through iteration after iteration.
• 9:41
  SPEAKER_01: And I'm asking it questions.
• 9:42
  SPEAKER_01: Well, wait a minute, why is this this way?
• 9:45
  SPEAKER_01: And and and why what about this document?
• 9:47
  SPEAKER_01: And it kept giving me more problems and then saying, okay, I'm solving them.
• 9:52
  SPEAKER_01: And then at the end of it, I looked at the output and I looked at the input and I said, hold on, these documents are exactly the same.
• 9:58
  SPEAKER_01: You didn't do anything.
• 9:59
  SPEAKER_01: Yeah, you're right.
• 10:00
  SPEAKER_01: Like, what?
• 10:01
  SPEAKER_01: And then I had it write an article, a Mia Culpa article.
• 10:04
  SPEAKER_01: What did you do wrong?
• 10:05
  SPEAKER_01: And it was absolutely incredible because it said it it created problems.
• 10:10
  SPEAKER_01: It identified problems that only it could see, and then it solved them.
• 10:15
  SPEAKER_01: And then I couldn't see the output, and then it created another problem.
• 10:18
  SPEAKER_01: And it admitted it.
• 10:19
  SPEAKER_01: It admitted it.
• 10:20
  SPEAKER_01: So so I think that organizations have to realize these things are incredibly powerful, but they will burn tokens and they will they will go through iterations of of work that that you need to be really careful to look at.
• 10:34
  SPEAKER_01: And that's why I think you do need to have LLMs judging the outputs of LLMs.
• 10:38
  SPEAKER_01: Do you want to speak to that?
• 10:39
  SPEAKER_01: Have you seen that kind of thing occurring?
• 10:41
  SPEAKER_00: I mean, that's exactly the difference between efficiency tool and production ready solution.
• 10:46
  SPEAKER_00: I love like cloud called, like codex.
• 10:49
  SPEAKER_00: I'm spending like 10 hours a day using it.
• 10:51
  SPEAKER_00: Just an efficiency tool.
• 10:52
  SPEAKER_00: And you need to like prompt back and forth, you need to control what it's improvised to you.
• 10:57
  SPEAKER_00: And when people say, Oh, we use AI, they like most likely use kind of copilot or chat GPT.
• 11:04
  SPEAKER_00: And that's a like 5-10% of what's actually possible.
• 11:07
  SPEAKER_00: When I'm talking about production-ready system, we are taking all the components, stitching them in a way so the system can validate itself.
• 11:15
  SPEAKER_00: We are orchestrated with the right architecture so you can get a kind of final output, not just kind of back and forth prompting.
• 11:23
  SPEAKER_00: Because exactly as you said, sometimes it's easier to build Excel yourself than like chatting back and forth with the system because it's constantly providing you like the different outcome.
• 11:33
  SPEAKER_00: That's where like you need experts who are gonna understand the process and invest a lot of time and resources to solving one kind of specific problem in the right way.
• 11:41
  SPEAKER_00: And that's I think applicable to everything.
• 11:43
  SPEAKER_00: If you kind of bought, let's assume, kind of good like groceries, but you don't know how to cook, you're gonna like boil something and okay, you you can eat it, but it's gonna be like very far from that cooking for you, right?
• 11:56
  SPEAKER_00: So I would say that's a simple analogy, but it's it's a similar and about any kind of services, about any technology in the same way.
• 12:04
  SPEAKER_01: Yeah, yeah, it's interesting.
• 12:05
  SPEAKER_01: I I I like your distinction between looking at efficiency and looking at productivity, right?
• 12:10
  SPEAKER_01: Because you can you can be efficient at a task that is not producing value, right?
• 12:15
  SPEAKER_01: And people get hung up in, hey, I'm using Claude for all these things, or I'm burning tokens and I'm and I'm you know producing these deliverables.
• 12:21
  SPEAKER_01: Well, are they really necessary and are they solving a problem?
• 12:25
  SPEAKER_01: And at the end of the day, it is about those business outcomes rather than the uh activities, right?
• 12:30
  SPEAKER_01: And so I think a lot of organizations are looking at this and saying, well, what are people really doing?
• 12:35
  SPEAKER_01: And what is the AI doing?
• 12:37
  SPEAKER_01: And that's why we're seeing, you know, layers of people being removed from the process, because many times the work that they're doing could be emulated or or replaced by an agentic system.
• 12:48
  SPEAKER_01: But then the question was, well, what were they producing in the first place in terms of actual value and productivity?
• 12:54
  SPEAKER_01: I want to I want to ask another question about, you know, you mentioned that accounting is so conservative, right?
• 13:00
  SPEAKER_01: And and and it has to be trusted.
• 13:02
  SPEAKER_01: So when you think of of resistance to AI adoption, what what is moving organizations past that?
• 13:09
  SPEAKER_01: Because again, they're so conservative and this has to be trusted, and there's there are these problems and challenges with with just the stories that we we just articulated.
• 13:18
  SPEAKER_00: So I would say following from like decades working in management consulting and deploying like techs for Fortune Hunter enterprises, I would say even if you have in the best technology in the world, it's a lot about change management and adoption.
• 13:32
  SPEAKER_00: Unfortunately, most of the companies in the market, they don't see the distinction between like we are using some sort of TI, like we are using Copan, and they don't understand the art of possible and how to properly like implement it.
• 13:45
  SPEAKER_00: When we are working with accounting firms, we also are not bringing just technology, we are helping them transfer transform their business model because historically these are organizations that are like you have a pyramid of the professionals.
• 13:58
  SPEAKER_00: You have like analysts, like managers, directors, partners, and historically all this work will done by like human experts.
• 14:06
  SPEAKER_00: Right now you're bringing AI that's gonna do like some work for you, and you need to understand okay, how gonna teach my like junior people, how now I like will review the work, how I'm gonna price these services, because now it's not like time and materials, but it's like something else.
• 14:20
  SPEAKER_00: So there are a lot of complicated questions that you need to navigate.
• 14:24
  SPEAKER_00: But I I think it's in the same like in any industry, you have the guys who are ready to take risks and try to win the market, or the ones who are gonna wait for like years until they will be kind of struggling and potentially will out of the market.
• 14:40
  SPEAKER_00: So the same as here.
• 14:41
  SPEAKER_00: I think it takes a while to get the folks to be comfortable to start using and adopting it.
• 14:46
  SPEAKER_00: But right now I see more and more people they understand that okay, that's not another like kind of bubble, it's actually technology that's working.
• 14:55
  SPEAKER_00: We see like anthropic open AI who are making like 60, 40 billion dollars revenue right now.
• 15:02
  SPEAKER_00: We are seeing like the efficiency boots, like probably like a month ago, we probably saw open AI model escape like from the sandbox environment and like hacked like hugging face.
• 15:13
  SPEAKER_00: So, all these pieces I just showing, guys, that's gonna be like a new reality, and you either embrace it or don't.
• 15:19
  SPEAKER_00: So, like in short, it's it's complicated to persuade the folks in the beginning, but we are trying to run the whole transformations and not just be the tech vendor, but the ones like who are helping them to go for the whole transformation.
• 15:32
  SPEAKER_01: Yeah, and and you mentioned the business model has to change.
• 15:35
  SPEAKER_01: I mean, financial due diligence for emergent acquisition transaction would take you know weeks and weeks and weeks and and and a team of CPAs.
• 15:42
  SPEAKER_01: So walk us through how you compress that to hours.
• 15:46
  SPEAKER_01: And then what does it take to produce that output so everything is traceable?
• 15:50
  SPEAKER_01: And then again, how do you go back and price those those uh services?
• 15:54
  SPEAKER_01: Because time and materials doesn't make sense.
• 15:57
  SPEAKER_01: You know, you you're you now you have to pay for an outcome, and how do you price that?
• 16:00
  SPEAKER_00: Yeah, so let's start from the last one.
• 16:03
  SPEAKER_00: So we're talking right now about providing this infrastructure to the accounting firms, so we help them to be more efficient.
• 16:09
  SPEAKER_00: Their clients, if you talk about the financial diligence, probably like private equity or like any kind of buyers who want to learn like or sell side, uh, who want to prove their like a kind of financials correct or find any potential kind of discrepancies, like ibida adjustments.
• 16:24
  SPEAKER_00: Their goal is to save millions by having an accounting firm who can verify financial.
• 16:30
  SPEAKER_00: There's like 100,000 like Ibida adjustments, like with the 10 mix multiple, that's the million dollar.
• 16:36
  SPEAKER_00: And like we are talking about like the smallest piece.
• 16:38
  SPEAKER_00: That's why for the buyers, it's not that important.
• 16:43
  SPEAKER_00: Like, are they paying like 10, like 30% more or less?
• 16:47
  SPEAKER_00: For them, it's much more important to have an accurate output.
• 16:50
  SPEAKER_00: And I I would rather pay you more, but you figure out these potential discrepancies and adjustments that like seller sides hide from me.
• 17:00
  SPEAKER_00: So that's why accounting firms should understand that they should price this as a value-based price, not like time and materials.
• 17:08
  SPEAKER_00: Sometimes it's hard to transfer there because that's historically how the industry operated.
• 17:12
  SPEAKER_00: Lots of the folks, like CFOs on the on the other sides, they are previously working in accounting firms, and they know that's that's how the services were priced.
• 17:21
  SPEAKER_00: That's why like you still need to have these conversations.
• 17:23
  SPEAKER_00: I I believe that they will be only much more work, but the quality of it and granularity will increase.
• 17:29
  SPEAKER_00: That's why you need to kind of lock the value-based price, but like start to deliver more and more value.
• 17:35
  SPEAKER_00: And for me, I'm looking for if I'm private equity, I'm looking for the good service and outcome.
• 17:41
  SPEAKER_00: And for me, it doesn't matter like how many hours your analyst spent.
• 17:45
  SPEAKER_00: So, yeah, that's that's a piece around the kind of pricing structure.
• 17:50
  SPEAKER_00: Uh, around compressing the cycles, typical deal takes like four to six weeks.
• 17:54
  SPEAKER_00: The first week is data exploration and like data book prep.
• 17:58
  SPEAKER_00: Probably closer to the end of the second week, you typically have a management meeting.
• 18:02
  SPEAKER_00: We can compress this time, we can take the whole like data rooms, all unstructured documents that you receive from the clients, and in three, four hours you will get the data book that's ready for the management meeting.
• 18:15
  SPEAKER_00: So that most of our clients they cut the cycles from like a week and a half to 72 hours to have a management meeting.
• 18:22
  SPEAKER_00: And that's how to accelerate the tip.
• 18:24
  SPEAKER_00: After that, any changes, any adjustments, any roll forwards, we can do in an automated way.
• 18:31
  SPEAKER_00: You just need to provide additional data that you receive from the client or like upload the transcript of your call, and the system will automatically incorporate all these changes.
• 18:40
  SPEAKER_00: That's why it's accelerated, like any additional adjustments and changes if it's needed.
• 18:44
  SPEAKER_00: It also helps you to provide the data gaps and manage back and forth uh with your target, with your client, that typically takes a lot of time.
• 18:52
  SPEAKER_00: Waiting for the data, making sure, okay, like we received it.
• 18:55
  SPEAKER_00: I I hear a lot of cases from the private equity guys, they say, Oh, I we are just annoyed sometimes because we are send these materials like this data several times, and they like continue asking us.
• 19:07
  SPEAKER_00: So we are helping to eliminate this friction too by understanding, okay, the data provided that that's missing, that's critical.
• 19:14
  SPEAKER_00: So to help to like streamline the whole process and get visibility on like what's happening in the data.
• 19:19
  SPEAKER_01: Yeah, I think that that makes a lot of sense, uh, especially the data visibility.
• 19:23
  SPEAKER_01: I mean, that's what the lifeblood of all of this is.
• 19:26
  SPEAKER_01: And uh and and uh you know, we we've seen that as well.
• 19:29
  SPEAKER_01: You know, you you get repeated requests for information, and you know, you people lose track of it, and and the AI systems will not lose track, right?
• 19:36
  SPEAKER_01: Especially if you are uh you know using it in a in a way that is uh is traceable and auditable, right?
• 19:44
  SPEAKER_01: That's that's uh critical.
• 19:45
  SPEAKER_01: And when you think about this, this is more than efficiency, right?
• 19:49
  SPEAKER_01: This is building an an employee, an AI employee, a digital employee, when you think about it, right?
• 19:54
  SPEAKER_01: And the big difference between an AI tool that makes a human faster and an AI system that takes a piece of work end-to-end.
• 20:01
  SPEAKER_01: How do you think about that distinction and what architecture makes that possible?
• 20:06
  SPEAKER_00: That's a story um around an efficiency tool as a quote-quad and have a production-ready solution.
• 20:12
  SPEAKER_00: The difference is here that with the efficiency tool, you prompt back and forth every time the output is different.
• 20:19
  SPEAKER_00: You can try to sanitize like some prompts or like skills, but again, there will be like no consistency, and every time that you have an edge case, you need to kind of solve it somehow.
• 20:28
  SPEAKER_00: With EI employees, we are taking all your input data and we provide you the final output.
• 20:34
  SPEAKER_00: It can be like 70 pages, Excel workbook, everything kind of formula-based, like proof of cash expenses, networking capital, like revenue waterfalls, KVD adjustments, like vendor customer concentrations, and like yarn yar yarr.
• 20:48
  SPEAKER_00: So we are providing you the final output that you can take to the client.
• 20:53
  SPEAKER_00: It's ready for the human review.
• 20:54
  SPEAKER_00: You don't need to kind of back and forth like interacting with the eye to get to that.
• 20:59
  SPEAKER_00: Especially if you have a bigger team, we are helping to standardize the best practices.
• 21:03
  SPEAKER_00: What's the template, like how how the formatting should be done, what typical analysis should be done.
• 21:09
  SPEAKER_00: And that helps to keep a consistent level of quality, granularity of insights, it helps to onboard the new employees, it helps to train junior people.
• 21:18
  SPEAKER_00: So you just have a consistency and you have a system not just providing you something half-baked, but it's drives this until like the ready like result where you can start providing an additional judgment, isn't it?
• 21:31
  SPEAKER_01: Yeah, and and I think that you know you you make the distinction when people say, Well, I can do this with Claude, or I can do this with ChatGPT, or I can do this with Gemini, or whatever your tool of choice is.
• 21:41
  SPEAKER_01: There's a very there's a big difference between a task and a process, right?
• 21:45
  SPEAKER_01: A task is something that you can execute in isolation in a standalone way, but the process is really that integration with other uh activities, other processes, other inputs and outputs, your your technology ecosystem.
• 22:00
  SPEAKER_01: You know, understanding the bigger picture business activity and business goals and uh making sure that you're building this in a way that is reliable, repeatable, and it does not cause additional friction in upstream or downstream processes, right?
• 22:17
  SPEAKER_01: So it's really trying to think of building this as an application that goes into an ecosystem.
• 22:23
  SPEAKER_01: Because you hear a lot of noise about you know software companies going away and you know, and and coding agents being able to replicate that functionality.
• 22:32
  SPEAKER_01: But there's a big difference between being able to replicate functionality and actually building a functional system and application because there's so much more to it.
• 22:41
  SPEAKER_00: And even if you can build fast, like, or let's assume write code fast, you have a crisis of the product thinking and expert thinking.
• 22:50
  SPEAKER_00: You need to know how to architect the systems, and you need to put a lot of resources of the product people who will learn the process, understand it, decompose it, like streamline, standardize, and then put this like in the kind of software or like technology.
• 23:06
  SPEAKER_00: And everybody thinks, okay, we have an AI, we can it can write like code fast, but you are missing the biggest piece.
• 23:14
  SPEAKER_00: That's um right now software, it's not about the coding agent, but it's actually having an ability to orchestrate multiple layers and get to something kind of production rate.
• 23:24
  SPEAKER_00: That's I think the the common kind of misconception because I I buy codes like a lot and I don't have like engineering the ground, but I play kind of really close and do my some personal efficiency automation.
• 23:35
  SPEAKER_00: But realistically, sometimes you have like 10x times more like of the tech depth than the actual competitions.
• 23:43
  SPEAKER_00: You can do something small, like for yourself, but to give something to use that, it takes me like probably five weeks more time to release like the open source, like 50x more time to get to the production ready, like 500 ticks more time.
• 23:55
  SPEAKER_00: That's a piece of it, it's about the product thief.
• 23:58
  SPEAKER_01: Product thinking and requirements.
• 24:00
  SPEAKER_01: I I I uh posted something on LinkedIn the other day where I was uh reposting a uh uh someone from uh Substack who was talking about that this is now causing a crisis of requirements, right?
• 24:12
  SPEAKER_01: You can do these things and you can say, well, this is what I want to accomplish, but what are the overarching requirements?
• 24:17
  SPEAKER_01: What are the business requirements that account for multiple stakeholders, multiple perspectives, multiple systems, right?
• 24:24
  SPEAKER_01: Because again, it's it's defining what the business is trying to do, but defining that in the broader sense of a of a technology and business ecosystem that has multiple players, multiple roles, multiple systems, multiple technology, multiple data flows, right?
• 24:40
  SPEAKER_01: So it's not just about speeding up this task, making this task more efficient.
• 24:43
  SPEAKER_01: It is really about that, as you say, that product thinking and that product mindset and the systems thinking, and then building the requirements that that enable that systems thinking and that product thinking.
• 24:55
  SPEAKER_01: Yes.
• 24:55
  SPEAKER_01: And I guess those are great, those are great points.
• 24:58
  SPEAKER_00: Is I right now is a very energetic and capable intern, can do like a lot of work, but it doesn't know how to run things properly.
• 25:08
  SPEAKER_00: And that's what they're building software nowadays.
• 25:10
  SPEAKER_00: And what's the domain specifics?
• 25:12
  SPEAKER_00: How to orchestrate all these things in the right way.
• 25:14
  SPEAKER_00: Then you can put like all these interns there.
• 25:17
  SPEAKER_00: Uh, but again, it you you can't give a task like build me like amazing product in the case, go drink tea.
• 25:24
  SPEAKER_00: It will not work.
• 25:26
  SPEAKER_01: Yeah, it's a good point.
• 25:27
  SPEAKER_01: And and even though the tools are getting better, there is still that that need for human judgment and understanding, you know, what is important to people, what is important to the business, because otherwise, you know, these these tools and these technologies can get very um narrow in their focus and and and uh blind, you know, have blind spots, right?
• 25:48
  SPEAKER_01: Because they they are they're they have too too narrow a focus on whatever that task, whatever that initial outcome is, and can miss a lot of things and do a lot of really dumb things, right?
• 25:58
  SPEAKER_01: So that's what what we we find.
• 26:00
  SPEAKER_01: Let's talk a little bit about the problem of junior count talent and especially in accounting firms, they've always developed senior talent by running juniors through years of foundational work.
• 26:10
  SPEAKER_01: If that foundational work is now being done by AI, how do the next generation of managers, directors, and partners get built?
• 26:17
  SPEAKER_00: It's just gonna be different.
• 26:19
  SPEAKER_00: And I would say every profession right now will go through like the same cycle.
• 26:24
  SPEAKER_00: Let's talk about software engineers.
• 26:26
  SPEAKER_00: You don't need any more junior roles, but you need to have junior people who will progress and become the next seniors.
• 26:35
  SPEAKER_00: But the pattern is just different.
• 26:37
  SPEAKER_00: I want people to be to have an ownership over like the something that they deliver.
• 26:42
  SPEAKER_00: I would love them to use AI, but they should be the ones that are responsible for the like the final outcome.
• 26:49
  SPEAKER_00: Um they should know how orchestrate AI, how to verify that it's kind of working in the right way.
• 26:56
  SPEAKER_00: Um, they should know how to learn much faster with AI.
• 27:00
  SPEAKER_00: So, before, for example, if you talk about accounting, you needed to do multiple deals like yourself in Excel, get the feedback, but right now you're getting much more exposure, so you can do much more deals faster and see like iterate with the eye and see the reasoning logic.
• 27:17
  SPEAKER_00: Learn like certain partners.
• 27:19
  SPEAKER_00: I would say like the educational system or like the professional development system is gonna be transformed, and that's what we are talking about.
• 27:27
  SPEAKER_00: Uh I think every company and leadership understand like how they retrain their people and then word like the new talent, but it it's it's just a new way of learning.
• 27:38
  SPEAKER_00: Yeah.
• 27:39
  SPEAKER_01: Yeah, it is interesting.
• 27:40
  SPEAKER_01: Uh all sorts of businesses are going through that because you know, senior people are retiring, uh, they're they're taking with them institutional knowledge and tacit knowledge and expertise.
• 27:51
  SPEAKER_01: And uh number one, how do you capture that and and embed it in in your systems and processes?
• 27:57
  SPEAKER_01: And of course, AI is very good at that.
• 27:59
  SPEAKER_01: I like to tell people record everything, right?
• 28:02
  SPEAKER_01: Record all your meetings, your client discussions, your your coaching, your interactions, your problem solving, just you know, uh brainstorming in your own thinking, right?
• 28:12
  SPEAKER_01: And record it's because AI can extract that and build it into things that are reusable and harvest those practices and harvest that knowledge from that and and turn them into things like skills that that can help you uh accelerate and and again accomplish those tasks.
• 28:29
  SPEAKER_01: One of the things that that we realized is that accountants and and and and have something that uh lawyers and consultants sometimes don't.
• 28:37
  SPEAKER_01: And that's deep trusted access to the entire financial body of a client.
• 28:41
  SPEAKER_01: How does AI change what's possible with that access?
• 28:44
  SPEAKER_01: And what new services does that unlock?
• 28:47
  SPEAKER_00: That's actually an interesting point.
• 28:49
  SPEAKER_00: Yeah, there was a survey about trustforwardness of different professional services.
• 28:55
  SPEAKER_00: And interestingly enough, accountants they have much more trust than management consultants or lawyers.
• 29:02
  SPEAKER_00: We can try to think why it's happening, but that that's a fact.
• 29:05
  SPEAKER_00: And it opens a lot of opportunities for accountants.
• 29:08
  SPEAKER_00: Why?
• 29:09
  SPEAKER_00: Because now you have an AI.
• 29:11
  SPEAKER_00: For example, consultants, their superpower was to structure information, highlight like the important pieces, run certain analysis, but most of this work right now can be done by them.
• 29:22
  SPEAKER_00: That means accounting firms they can start extending the amount of the services they provide.
• 29:27
  SPEAKER_00: But what's cool with accounting firm, they have a recurring work with the with their clients.
• 29:32
  SPEAKER_00: In the consulting, it's mostly one-time projects.
• 29:35
  SPEAKER_00: You may have like a kind of bigger transformations, but the accounting firms they work with you like every month, every year.
• 29:42
  SPEAKER_00: They buy your search service.
• 29:44
  SPEAKER_00: That's why they can use this access and recurring services to start extending the amount of like the work and offerings that they per lock.
• 29:52
  SPEAKER_00: And like with the eye, they becoming kind of much more capable.
• 29:55
  SPEAKER_00: That's why I see the big opportunity for like accounting firms to start to become like the advisory firms and support across multiple elements.
• 30:04
  SPEAKER_01: Yeah, that's that's very interesting.
• 30:05
  SPEAKER_01: And it is such a great business model when you think about the fact that you have to do it every year, right?
• 30:09
  SPEAKER_01: You have to uh periodically audit and produce tax returns and all of that.
• 30:14
  SPEAKER_01: You uh believe that the long-term outcome is a better world, but the short and midterm transition is a problem because this technical revolution is moving faster than people can absorb, right?
• 30:27
  SPEAKER_01: The technology is moving so fast, it's even for the experts, it's very difficult to keep up.
• 30:32
  SPEAKER_01: What does that mean for organizations and individuals who are not moving as fast or fast enough to be able to leverage all of this?
• 30:41
  SPEAKER_00: Yeah, we're we're getting philosophical here.
• 30:43
  SPEAKER_00: Yeah, there are like multiple books, like Singularities Near, or like others that shares the perspective on the world.
• 30:50
  SPEAKER_00: So, my honest belief, I think in the long term, we're gonna live in the better place.
• 30:55
  SPEAKER_00: Yeah, you're gonna be like everywhere, you'll gonna be happy, we'll have enough kind of products and services for us.
• 31:03
  SPEAKER_00: So that's where we're going to.
• 31:05
  SPEAKER_00: But I'm concerned about the midterm in terms of an ability of economic to transfer in new reality.
• 31:13
  SPEAKER_00: Because if you think about any technical revolutions before, they were going like over like hundreds of years, right?
• 31:20
  SPEAKER_00: Like in different parts of the world.
• 31:21
  SPEAKER_00: But right now with AI, like we can do this tech revolution in the five, ten years.
• 31:28
  SPEAKER_00: Yeah.
• 31:28
  SPEAKER_00: As we talked, like change management is important, it's it's much harder within this time frame to completely change the way how the people work, how organizations are structured.
• 31:39
  SPEAKER_00: That's why I'm concerned that a lot of people will stay without jobs.
• 31:43
  SPEAKER_00: Yeah, while I believe that with the eye we will have much more opportunities and like work to do, there is a risk in the midterm, the people who are not ready to be flexible enough and adjust, they can start losing their jobs, especially kind of the white colors.
• 31:59
  SPEAKER_00: What it means for organizations, I would say there is no opportunity to like bury your heads in the sand anymore into not going to touch you.
• 32:08
  SPEAKER_00: You need to embrace it.
• 32:10
  SPEAKER_00: And if you are not ready to embrace it, be ready to be out of the service and out of the market, because there are a lot of the companies who want to take your market share, either incumbents who are ready to risk and move faster, or like AI native companies who just rebuild the whole business model like from the scratch.
• 32:29
  SPEAKER_00: That's why the best recommendation to any organization is start doing this today.
• 32:34
  SPEAKER_00: It can be like even smaller like use cases, but you need to start doing this today.
• 32:38
  SPEAKER_00: The people need to learn how to work with this.
• 32:41
  SPEAKER_00: That's your like one of the biggest assets.
• 32:44
  SPEAKER_00: The people who are smart enough, capable enough of how to leverage the technology.
• 32:48
  SPEAKER_00: For the people or like for the students, I have a younger brother, and like my best advice is just spend a lot of time every day, like with the eye.
• 32:58
  SPEAKER_00: Try to burn as much tokens as you can, but try to learn how to work with it.
• 33:03
  SPEAKER_00: Be curious enough.
• 33:04
  SPEAKER_00: I'm uh I I have a nephew, and so yeah, we are building some computer games with him.
• 33:09
  SPEAKER_00: So I think try to be a creator and test these systems every day until you and understanding how to work with the new technology and we will well position you for the future.
• 33:19
  SPEAKER_01: That's terrific.
• 33:20
  SPEAKER_01: Well, listen, Nikita, thank you so much for joining me and sharing what it looks like to rebuild the financial analysis from the ground up with agentic AI and what it means for professionals and organizations that are managing that and navigating that shift.
• 33:35
  SPEAKER_01: And to our listeners, thank you for tuning in to the Early AI podcast.
• 33:38
  SPEAKER_01: Be sure to subscribe for more conversations on how AI is shaping the future of business.
• 33:44
  SPEAKER_01: And one last thing I wanted to mention.
• 33:46
  SPEAKER_01: Uh, we are now trying out a new platform called DLogos.
• 33:50
  SPEAKER_01: And I think at D L-O-G-O-S.com, our podcast will be there.
• 33:55
  SPEAKER_01: Uh, it's going, you know, you can uh learn about other podcasts.
• 33:58
  SPEAKER_01: Uh it's a little experiment we're trying, working with them, they're a startup, and uh so check that out.
• 34:03
  SPEAKER_01: There'll be show notes there and and other things.
• 34:05
  SPEAKER_01: And Nikita, how can people connect with you?
• 34:08
  SPEAKER_00: So, my linked in you can come off or like the website is dubs.ai, glbs.ai.
• 34:14
  SPEAKER_00: Yeah, happy to be a third partner like uh in your AI journey.
• 34:19
  SPEAKER_00: I know that's complicated and it's exciting and scary at the same time, but yeah, happy to be like the third partner in this journey.
• 34:26
  SPEAKER_00: Thank you so much for organizing.
• 34:27
  SPEAKER_00: That was wonderful.
• 34:28
  SPEAKER_00: Enjoy the conversation.
• 34:29
  SPEAKER_01: Thank you.
• 34:30
  SPEAKER_01: And we'll talk to people next time on the next early AI podcast.
• 34:35
  SPEAKER_02: Thank you for joining us on another deep dive into AI innovation.
• 34:39
  SPEAKER_02: Tune in next time when we introduce another industry expert and discuss how to maximize AI in mobile.
• 34:46
  SPEAKER_02: The early AI podcast is sponsored by early information science investors.
• 34:51
  SPEAKER_02: To learn this, visit early.com.com.
• 34:56
  SPEAKER_02: Thank you for tuning in.