Founders know AI has potential, but what does early success really look like when humans and agents work together? John Ikosipentarhos shares how his firm is building a living knowledge base powered by AI. By observing how the team interacts with it, what they ask, what they miss, he’s creating a feedback loop that makes both the humans and the agents more effective. This is a practical look at augmenting, not replacing, expertise, exactly the kind of measured adoption most firms need now.
TRANSCRIPT
Greg Alexander: Hey everybody, this is Greg Alexander. You’re listening to the Pro Serv Podcast, brought to you by Collective 54. If you’re new to this show, this show is dedicated exclusively to founders of boutique pro-serve firms. So if you market, sell, and deliver expertise for a living, this is for you. On this show, we aim to help you make more money, make scaling easier, and make an exit achievable. On today’s show, we are going to talk about… How we can blend Humans and AI agents in professional services To deliver more value to clients, and to generate more profits for firms. Now, there’s a lot to cover here. So, we decided to skip all the theoretical stuff. And ask a member to be on the call. And actually share… have him share with you how he’s got this in production right now. So that’s what we’re gonna do. So joining me on the show is John, and John, I’m not gonna try to pronounce your last name again. We’ve known each other for years, and every time I do it, I screw it up. So, I’m gonna skip the last name and have you introduce yourself. Please, introduce yourself and your firm to the community.
John Ikosipentarhos: Yeah, John Ikosipentarhos. Ikos for short, so maybe next time we do a podcast together, you can just go with the four-letter, acronym with just IKOS. … But I’m one of the co-founders of Zeroed In Consulting. We are an accounting and finance consulting firm, that specializes in outsourcing technical accounting, and I will spare the boring tidbits of that, but just know that we love numbers, and we love back office work, so I’ll… that’ll be my intro today.
Greg Alexander: Sounds good, John Ikos. Alright, I’ve got some categories here. So they are the knowledge base build, particularly with ClickUp. The human-AI collaboration, the continuous improvement loop, The modernization of the workflows. overcoming internal resistance, etc. So before I jump into all of that, maybe you could lay the groundwork, because I know what you’re up to, but most don’t. Could you kind of explain, you know. What happened, and where you are right now?
John Ikosipentarhos: Yeah, and I think to give you some context, it’s important to understand, like, what we do. So, we provide… services at, like… so, we’re… we have two pillars of the business currently. We have an outsourced accounting side of the business and a technical accounting side of the business. So, to understand what technical accounting means. Technical accounting is basically the nerd’s nerds of accounting, where we go in and you have a revenue stream, and you need to account for it a certain way under U.S. GAAP compliance. Historically, the way that that work stream works is you’d have someone go and do a bunch of research on this company that basically tries to peg how they do their work stream with the guidance. Now, that would take a lot of manual effort, because as you can imagine, there’s a lot of different companies out there that do a lot of different revenue streams, and the way that they do it, there’s a lot of nuances within those revenue streams. So what we’ve tried to do is we’ve tried to automate that, and actually we have automated that, with ChatGPT and agents to essentially now, instead of spending 5 hours doing research on, hey, this is what the company does, and this is the path that we should go down, we’re now basically arriving at that answer in probably 5 to 10 minutes, because we’ve built out these really, really, really good prompts that we’ve refined over time to essentially hyper-focus on getting us to those answers.
Greg Alexander: Okay, that is a great, context for us, and I’m glad you brought that up, because we would have lost people without that in the backdrop. So, you know, this, this… I guess, knowledge base, for lack of a better term. … You built it up through ClickUp, is that correct?
John Ikosipentarhos: Yeah, so ClickUp is our kind of brain of our organization, so ClickUp is a project management tool. It has a document layer, so you can basically store information in docs within ClickUp, but alternatively, it also has the ability to query SharePoint or Google Drive, and so we’re trying to transition everything into ClickUp, but we do have historical documentation that sits in SharePoint. And so, originally, the knowledge base that we had built out pre-AI was all built in OneNote, and so if someone wanted to go and understand, hey, how do you go recognize revenue the zeroed-in way, you’d have to go into OneNote, go into the notebook that says Revenue Recognition, go down and look at what are processes, and as you can imagine, that probably never happened.
Greg Alexander: Yeah, yeah. I mean, you know, one note is, we used to say that’s where documents go to die. Okay, so… So, I understand that now, okay? … So what happens now, I guess? Contrast the after with the before.
John Ikosipentarhos: Yeah, so now what we do, let’s say we get a new client, and they’re looking to get, you know, again, we’ll go with this revenue, example, and like, hey, we need you guys to come in and do a revenue recognition implementation, look at our revenue streams, and then essentially the end result, what they want out of this, is a memo that accurately describes all the revenue streams and how to treat it under US GAAP, and then if there’s any modifications that we need to do in the accounting. Side of things, we give them the journal entries, and then they go and book them, but let’s not worry about that piece. So now, when we start receiving those contracts from our customers, what we end up doing is we throw those contracts in with our pre-built prompt, and we basically have AI take the first pass at it. Now, the output of that, so we kind of break it up into 3 separate, I guess. bots, if you will. One we call, like, the data extraction bot. The second one is… it takes that extraction, so let me maybe take a step back. That first bot takes the contract, extracts key terms, and then lays them out for the human to basically say, are all of these key terms correct? Once we validated that, we throw it into the second bot. The second bot then goes and does the accounting analysis, again, with a prompt that we’ve built out there, to essentially come, and the output of that is, hey, this is how I think you should be accounting for this revenue contract. And that’s the output. And then that gets taken and thrown into our third bot, which we’re actually actively building, which is the memo generation bot. And then that would effectively write our ASC606 memos for us.
Greg Alexander: Okay, very good. Now, that is a great example of… Humans plus tech. Automating workflows, improving the quality of the work for the client. Reducing the time it takes for employees to do it. … Give me the ex… like, what type of improvement have you seen in team efficiency?
John Ikosipentarhos: Yeah, so we actually… it’s funny, this came at a very timely, portion of this journey.
John Ikosipentarhos: We had a team member actually go and do a retrospective. So we had done a bunch of business acquisition work for a client, and we had done it manually. And then, so she came in and basically redid all of her work using our AI prompt. And she went in there and she highlighted, here are the 15 different things that we would need to look for under a business acquisition accounting memo. And AI hit, like, 12 out of the 15. And the three that it didn’t hit, it was actually incredibly insightful, because it ended up pointing out the same thing across all three contracts that she had reviewed. So what we did as a takeaway was, we clearly have a gap in our prompt. We need to go in there and update our prompt to essentially add in, hey, you need to look for employee agreements within business acquisition. Like, those are very important. And so now, taking in that feedback loop, that prompt is going to be updated, and that future iterations are going to now have that included as part of it.
Greg Alexander: On this recent example, and it’s a very… Fantastic example, and congratulations to you on doing this. … This might be an unfair question, but I’m gonna ask it anyways. You know, if you think about the job to deliver that service for the client, end-to-end, how much of the job, expressed as a percentage, is now completed by the tech, and how much is completed by the person?
John Ikosipentarhos: It’s funny you ask that question, because I actually prepared a slide on that for the presentation. So, I will say that for… I use leasing as an example. It’s 87% faster to do it the way that we’re doing it now, as opposed to the way that we were doing it before. I don’t have the exact metrics for the exact example that I had, but I can tell you that she was able to go in there and do a full review of the three contracts line item by line item. She said it took her 2-3 hours to do. And I can guarantee you, those memos used to take 5 to 10 hours per memo, minimum. I’m of the mind that, you know, to get that last 13% is gonna take an army, and I think 87% is an incredibly amazing marginal gain, and I’m happy with that. And so what I want to actually try and focus on is how can I get the most out of the employees that I currently have, to where we’re not really hiring as fast as we thought we were going to be. The revenue keeps increasing, however, we’re not really going out there and getting more and more employees. I think that’s an indicator to me that things are working as we would expect. Now, it has been brought up, just to be fully transparent, the team has raised their hand and said, what’s the expectation here? Like, are we going to be managing, like, 20 clients under this model? And honestly, I didn’t know how to answer that, because I can understand from their point of view, wow, 20 clients is a lot, context switching, we’re human, that is going to be pretty hard. So I don’t actually have a good answer for you on that. However, it is something that we need to consider. Like, at what point do humans start to become inefficient in managing all these different projects, right? So, yeah, I don’t know.
Greg Alexander: You know, I just want to summarize something for the listeners. You know, today’s topic is about human plus AI. And combining those two assets to grow your firm. The big reality of the situation is that people are not losing their jobs, at least within our community, because of AI. However, the founders are making a lot more money, because as revenue grows, they don’t have to add payroll costs. And payroll costs are somewhere between 80% and 90%, depending on your firm. So if you can keep headcount flat and throw more revenue on top of it, profit margins expand dramatically, therefore more dollars in the front pocket of your jeans if you’re a founder of the firm, which is awesome. Now, what will happen going forward? I don’t know. Where this point of diminishing returns is, and context switching becomes a problem for employees, I’m not 100% sure, but that’s the big learning. So the takeaway, if you’re listening to this, is before you okay the next job rec, ask yourself the question, am I maximizing the productivity of my existing employees before you add to the payroll cost? Speaking of internal people and questions that they’re asking you, John, we’re seeing across Collective 54 some resistance internally. Did you deal with that resistance, and was it strong? Was it mild? Was it non-existent, and how did you deal with it?
John Ikosipentarhos: Yeah, I think if you ask my co-founder, I was beating my head up against the wall for a good 6 months. Like, once ChatGPT3 came out, I was like, we need to jump on this. And, you know, he was like, alright, cool, we’ve introduced a lot of technology very quickly, and this is unproven. You know, it was kind of like NFT, crypto, and now it’s AI. And you’re like, is this another fad, or is this actually going to be something real? And I was like, no, this is real, this is real, I’m shouting from the rooftops. But I was actually met with resistance at all levels. But I think I persisted in the sense of, alright, we’re not gonna just jam it down people’s throats, we’re gonna take the slow approach, continue to kind of talk about it, talk about it, talk about it. And we weren’t really getting anywhere, and I think what ended up pushing people over the edge, there were two instances that I think really helped our firm in particular. One, we meet up at least twice a year in person, the whole firm, and we do these kind of hackathons, where we do a case study together, and we work on it as a team to collaborate. And one of the things that we did was we used AI as a team. So instead of having it be this nebulous thing, I had been trained on it, and so I designed this hackathon to coach people on the power that you can get out of it. And so that was an aha moment. That was number one. But as I’m sure you’re aware, I’m sure Collective 54 members ask this probably on a weekly or daily basis: okay, that’s great, all this theoretical AI stuff, but how do I actually use it for me, right? I don’t care that this super intelligent AI firm is doing it… I need simple things to get started. And so what we did was we have a weekly AI call, where everyone has to submit an AI use case, and from that, it sparks not only a competition of how do we do it, but also opens people’s minds, because we’re all working on similar things. Like, oh, you can put in a contract and have it spit you back key terms. That’s crazy, I never thought about that. And that’s now the base layer. And then we’ve obviously leveled up as a firm where the things that maybe were base layer 6 months ago are now just the norm.
Greg Alexander: Oh, I love that. The hackathon idea is a great one. The weekly AI call featuring real individual-based use cases is a great one. A great way to overcome resistance. I might add two cents there. You know, when I hear this, I say to the employees of our member firms, I’m like, listen, you’re gonna fall into one of two camps. You’re either gonna resist, and therefore, you’re gonna get replaced.
Greg Alexander: But you’re not going to get replaced by the AI, you’re going to get replaced by a human who’s capable with AI, someone who doesn’t resist. So the previous expectation of output, however you want to measure that, you know, work product per hour, or something like that, those previous expectations are gone. Now, because we have the superintelligence, expectations have gone up, and you’re either gonna meet those expectations or you’re not. So that’s one group, okay? The other thing is, isn’t this an opportunity for all of us to redefine our roles? And if we can contribute that much more valuable to… that much more value to our clients and to our firms. Shouldn’t profits go up? And if they do, shouldn’t your individual comp go up because you’ll participate in that profit increase? So, I’m appealing to both the fear factor, you know, the loss side of the equation, and the greed factor, you know, the gain side of the equation. And I think that’s the way to do it. Sticking your head in the sand and just being a resistor here is ridiculous, because, I mean, you know, this is coming. It’s here, right now, and John’s a living proof of that. Alright, let’s conclude this interview with one last question, and then we’ll save the rest for the private member Q&A. If you were to tell a… your peer, an owner of a boutique pro-serve firm who maybe isn’t as far along as you are, maybe, I don’t know, 1, 2, 3 things or so, just to get started. What would you tell them?
John Ikosipentarhos: It’s funny, because I’ve actually had this conversation with multiple members of Collective 54, and so I will just give that same advices. Like, the first thing is you just gotta start somewhere, and I think it’s one of those things where you say, okay, what is a simple task that I don’t like doing? For example, I don’t like invoicing. Let that be an example. Okay, so how are you currently doing your invoicing? Is there a way that you can automate a piece of it or all of it using AI? And if you expect the solution to be, I’m going to type in one sentence, like I used to do in Google, and have that be the outcome, and if it gets it right, great, and if it doesn’t, well, see, I told you, AI sucks, then that is, like, the wrong mentality to have. What I usually tell people is, like, treat it like an intern, and if you can kind of, like, hey, this is what I want you to do. And it’s like, oh, did you mean this? And I’m like, no, I actually didn’t mean that. I actually want this, and kind of, like, walk it through and maybe spend some time, you’ll see that there’s a lot of gains to be had. I think the number one mistake people make is that they don’t actually know how to explain things very well. And so they just think that, oh, you… they just, like, understand my words. I’m like, well, it actually took your words very literally, and maybe you have to break it down a little bit more. So I’d say just be very descriptive, spend more time on the front end building out a really good prompt, whether it be 5, 10 minutes, not 5 or 10 seconds, and you’d actually be amazed at how good it can be.
Greg Alexander: Okay, fantastic advice. We’re going to leave it there, because again, I want to give the members a chance to ask their questions directly. So, John, on behalf of the community, you’re always a great contributor. Thanks for coming and sharing your wisdom with us today.
John Ikosipentarhos: Yeah, it’s my pleasure. Thank you.
Greg Alexander: Alright, a couple calls to action for listeners, so if you’re not a member of Collective 54 and you want to become one after listening to this, go to Collective54.com and fill out an application, and we’ll get in contact with you. If you are a member, and you’ve got a ton of questions for John, look for the invitation for the private member Q&A. You can ask your questions to him directly at that point. But until next time, thank you for listening today, and I wish you all the best of luck as you try to grow, scale, and someday exit your firm.
Note: This transcript was generated by Zoom.