#113 - Young Lee & Asheesh Gupta: Audax AI Value Creation

19:44Asheesh Gupta:We don't remember them. But human beings have biases. And I would say that on balance, if I'm comparing a human bias or a Gen AI bias, there's a way to code and dial up and down the Gen AI bias to be as neutral as possible versus the human. And so we find it actually is less of an amplifier. It's more of a dampener function to the conversation where people can go down paths that we may want to be cautious about.

20:08Young Lee:It's a great point. And it is a very valid concern. For us, it's something that we have to be constantly vigilant and proactive about training people, just natural human behavior of wanting to hear what they want to hear, using AI as a way to support why they want to do an investment or to rationalize what happened with a bad deal or a decision that didn't go well. For us, the way we like to use AI, my favorite tool, as Ashish was mentioning. We have to date about 60 custom chat GPT portals, and each of these portals address a discrete part of the work stream as we kind of think about the overall workflow from deal sourcing all the way through exit.

20:55Young Lee:We identified over a hundred discrete pieces of workflow where we thought AI could help us be better. To date, we've developed 60 of these proprietary chat GPT portals that are in use. As an example, an information memorandum comes in and it's used to summarize based on all of the information that we have in our proprietary database set. It's an exercise that used to take an analyst or associate three plus days to provide an output. It takes us 10 to 15 minutes. Of those 60 chat GPT portals, my favorite is what we call the devil's advocate. And what the devil's advocate portal does is, as Ashish was saying, is it asks the uncomfortable questions, the questions you don't want to hear very clinically based on the mistakes that we've made or pointing out where things have gone bad with the benefit of all of the data and our experience that over the past 25 years.

21:53Asheesh Gupta:Alex, it's funny, like when you think about, you know, biases, there's biases we have, but then biases we express. And I always think about some of this, you know, if you think about eight or 10 people making a really productive decision, their personalities and history and styles. And when you're doing the devil's advocate, there's no emotion. There's no, I'm going to hurt your feelings, right? And it's sort of a, it's a pretty clinical response. So it's funny, like when we send the investment committee memo, almost all the partners drop it into a devil's advocate, then they interrogate it back and forth on the devil's advocate.

22:24Asheesh Gupta:And you can see it because the questions that happened in that room are slightly different. But sometimes people actually joke, oh yeah, devil's advocate said X, Y, Z. It's like, no, no, no. you wanted to say it, but you're worried about sort of the interplay of the humans in the conversation, which is natural, right? Alex, you work with people for a long time. You get those kinds of dynamics that you have to keep in context.

22:44Young Lee:And I'm sure you've experienced this, but when it comes out of the devil's advocate, the person receiving the feedback probably has a very different reaction than if it came out of a human's mouth and if it was the exact same words, because you can't get offended by a computer pointing things out like you would versus a human.

23:01Asheesh Gupta:That's been our experience. It's been much easier to have that. And we say devil's advocate, that GPT is not written just for an IC memo. I could say, I want to hire this person. Here's what I want to pay them, et cetera. And it's going to say, hey, but what about this, this, this? You know, you drop the position spec. So you can apply it to all manner of things. And we've written it such that you drop a document and all you do is hit proceed. And it then says, this document is X. My assumption is you want me to take the opposite side. Please confirm. Yes. And at the end of The Devil's Advocate, it goes through the whole interrogation.

23:31Asheesh Gupta:You can also, as you know, share the link with somebody else if you want to have them continue the conversation. Or it asks you, hey, do you want me now to affirm why this is a good decision? And so you can actually play both sides of that conversation because it doesn't matter if you're saying, I shouldn't hire somebody telling me why I should or I should hire somebody to tell me why I shouldn't. We're trying to figure out how to make sure we have, you know, Young uses the term sparring partner, right? We want a sparring partner, somebody who can go back and forth based on the situation.

23:54Young Lee:But not the type of sparring partner that's going to go and knock you out. That's right. And the tone of what I'm hearing is this endless seeking of what the truth is without any biases or historical reference points. You're just looking for what is accurate and what is the truth. And so you want to have a balanced perspective. And using AI can be extremely helpful because of the biases and emotional attachments that humans may have. That's exactly right. We've talked about the need for AI fluency at the GP leader level for it to flow through the organization. How do you think about fluency in practice?

24:37So Alex, if I think about fluency applied to a different domain,

24:42Asheesh Gupta:if I were to take a simple example of at Audax, we do 100 add-on acquisitions a year. And if you were to ask us, do our senior leaders at Audax have fluency on the complexity of M &A integration, that's born out of many years of doing the work, seeing it work, seeing it not work, etc. And when I distill that down, I end up coming to some, I guess, the same ideas again and again. One, are you asking the right set of questions, but not advancing some objective? Meaning, are you saying, hey, we're trying to build a better business. That's why we do add-ons. We're trying to build a business faster.

25:15Asheesh Gupta:That's why we do add-ons. We don't just do add-ons for the sake of add-ons. And I share that in the age of AI, because two things I think are true. One, And if we're fluent at the GP, then the conversations, the richness of the conversations we have with our management teams is higher, right? We can then have a conversation about, here's how we use it. Hey, how do you think about it? This is not how we tell our management teams, go use it. And we are saying, no, no, it doesn't apply to us. Because I think that conversation just within the history of our firm is to say, how do you think about this?

25:41Asheesh Gupta:Here's how we think about it. How do you think about it? And sharing that experience is very important. When you think about the word fluency, Alex, I think over the last couple of years, it's gone from, I'll say, proficiency to fluency. And the fluency often is in asking, can leaders identify high value use cases? In the past, it was like, hey, we have a bunch of tools, let's push tools. And it was like a tools forward view versus it is a use case back view. And I think highly fluent leaders ask the question, why can't we do this? You guys go figure out the tools. Let's not debate, is it tool X or tool Y and get enamored with all the tools?

26:16Asheesh Gupta:Let's talk about, here's the use case. How do we think about it differently? And what we find is in Gen AI in particular, you have to have this catwalk view of an organization very differently than you did previously, because you have to understand all the interlinkage between one team maybe generating data that is going to inform better decision-making for another team. And that's how we think about fluency at the GP is the senior leaders asking these questions in a more proficient way, but trying to be very careful that fluency is not conflated with tool proficiency. Because that's what I worry about is people like, oh yeah, I use it all the time.

26:49Asheesh Gupta:I don't know, so you use it all the time. Are you proficient in the tool or are you fluent in the use of generative AI? That mindset is what we're trying to advance is the fluency of this as a capability that cuts across many different use cases and not being, hey, I got a tool, let's go deploy the tool and the end of five years we're going to tell the next buyer, oh, everybody in the firm uses the tool. There's plenty of examples in our portfolio that I've learned where that pushing a tool doesn't actually lead to proficiency or fluency. One of the other topics we've discussed in the past is the different stages of AI adoption

27:22Young Lee:as you progress through that journey. In the first stage, AI often shows up as productivity gains. What's the highest quality productivity gain that you've seen?

27:32Asheesh Gupta:I'd say, Alex, today you see productivity gains in those use cases that are highly repetitive. So at Audax, we see over a thousand confidential information memorandum a year. Reviewing and processing those, as Young mentioned earlier, is a three-day, used to be a three-day effort. Very systematic, like output was very clear what the process was would take three days. Today, when you can do that in 10 to 15 minutes, just to produce that output, but then ask all manner of follow-up questions. If you apply the same three days, the density of what you can do is just totally different. So that's where people have run towards productivity.

28:05Asheesh Gupta:The way I describe it is in the persona of I've got somebody helping me carry the load, productivity being a proxy for carrying the load. That's where people run towards. When we actually value stream mapped our entire deal lifecycle and identified at the time 28 categories and over 100 use cases, we basically said to our early career group, what are the things that you are outputting, producing output on that you think is repetitive enough and is frequent enough that you will see a productivity lift? And so in stage one, that's what we see. And we see that at the GP, Alex, and there are plenty of examples.

28:38Asheesh Gupta:We see that in the portfolio, simple things, ticketing systems, one-way inquiries to HR, simple, you know, order processing, you know, thousands of orders that need to be hand-keyed or transmitted. That's all productivity uplift. Those use cases are so easily bounded that you can actually see the benefit very, very quickly. And I say within three weeks, you can have a POC up and running, and then obviously the strength of your governance drives how broadly you can deploy that within your organization. But that's what I would say in stage one is very, very easy to see. And we see that many, many, many, many areas of our business.

29:11Young Lee:And I guess it's also easier to implement because you just, it's almost like using a tool and you can see the immediate impact.

29:18Asheesh Gupta:Yes, Alec, there's always the people who are the most curious and the most frustrated that we love to partner with because they're curious. So they want to try and figure out how to do it better, right? They have this, I want to, they're always aspiring and they're frustrated with the status quo because they think there is a better way. And so when you show them a handful of tools, man, magic happens, right? If you give them a tooling, what you started with and what you end with. I remember when we built our first custom GPTs, you know, they were two pages, three pages, Alex. And then you get this like short cycle feedback.

29:45Asheesh Gupta:And the feedback was, I ran it. It doesn't do this, this, this, not as a criticism, but as an energy for getting better. And so then the team that was building them, now it's like eight or nine pages. The prompt is eight or nine pages with all manner of permutations. And as you know, Alex, as the technology evolves, you can then stitch them together and do all manner of things. But that's the people that we always seek out is the highly curious ones and the ones that are slightly frustrated because they bring a good energy to trying to drive improvement.

30:10Young Lee:So if we move to the second stage, the impact shifts toward better decision making. Would you talk through that?

30:18Asheesh Gupta:Sure. Yeah. At the end of the day, when we think about AI, our view is like, you got to drive better decisions faster. It's not just faster decisions. It's better decisions faster and quality will not be subordinated to speed, right? Quality is always the dominant, is the prime directive, so to speak. So the way we think about it is, if you think that example that I just mentioned, you know, three days now are down to 15 minutes. If that means you go off and don't do anything for the next two days, that's not a productive use of time. But, you know, we got asked this question, how does the early career group, you know, analyst associates, senior associates, how does their job change?

30:50Asheesh Gupta:And the answer is they ask a lot more questions And they ask questions that they typically would have had to hand off to the principal or the vice president to go answer. But now an analyst can say, hey, I did the summarization, the analytic work. Here's the 10 questions that I'd love to answer. Can I go off and think about those? That body of questions, Alex, is where we think the better decision making comes. Because you don't say, hey, I'm resource constrained or time constrained, so I can only answer three questions. No, no. You can actually now ask 10 or 12 questions. They have to be the right questions, naturally, but your ability to make better decisions is a function of the questions you ask and the data you're able to go and sift through to get to that, again, that same comment from earlier, the signal versus the noise.

31:30Asheesh Gupta:So we see that early in the portfolio where we find a lot of values. And there's many examples that I'd call low profile jobs, right, Alex, those repetitive jobs. It's where we're seeing the impact is in hiring. So in the past, if you had three candidates, and I'm talking about C-suite candidates, if we're hiring a vice president of financial planning and analysis, and there's three candidates with different profiles, in the past, you would do a lot of interviewing. You would record. Now we record the interviews. We do Hogan assessments. All of that Hogan assessment, the recording of the interviews, the resume, the key accountability scorecard, you can actually package that up to produce a remarkably detailed person-by-person question.

32:10Asheesh Gupta:Meaning, if I'm interviewing an FP &A person, it tells me, hey, Ashish, you should ask these five questions of this person, different questions of this person, different questions of this person, because it identifies gaps in experience or demonstrate experience. Alex, if you think about that, in the past, you have to be a very expert interviewer to be able to do that. So the quality of a conversation that a vice president is having or a principal is having with a senior leader that they're interviewing in the portfolio, totally different conversation, Alex. You'll hear executives come out often, either I go first or I go last.

32:40Asheesh Gupta:Somebody say, wow, man, that was a full day. You're like, what do you mean? So, man, the questions you guys ask, you guys ask some intense questions. And you just see that intensity, the thoroughness of that conversation. It's totally different than the tell me a time when X, Y, Z, right? You just get to that higher fidelity. So many other examples I can share, but those are ones that I thought might be helpful to elaborate, driving better decisions about making a better underwrite and getting a better team and getting a team that matches the mission, we spent a lot of time on that in stage two.

33:10Young Lee:Okay, so we completed freshman and sophomore year. Now we're on to the third stage. And this is more about the transformation of the operating model, particularly in how companies interact with customers. Would you walk us through a few examples of that?

33:25Asheesh Gupta:Yes, so Alex, this is often what we chatted about a little bit here today is this question of, do I have enough data to drive a good decision? And so what we find is in that operating model evolution, you ask questions that are, if I had X, I could do Y. And then this question of what is the friction to get X? If I had more data about the customer, I could target them a bit more effectively. Great. What information do you want? And so, again, as context, we're partnering with middle market companies where their sophistication on how they get fidelity of their customers is not as high as it would be for a billion dollar enterprise.

34:01Asheesh Gupta:And so those companies often, I don't know enough about my customer. And Alex, you probably know this. There's dozens of tools now that can actually go scrape and build customer profiles very quickly. Right. So we sort of joke every time we have a meeting, we're not starting with a Google search. We're starting with a custom GPT search of that person building an entire profile of that person. So a portfolio company can now go do that for customers such that they can be more explicit in their targeting. This isn't a dial by volume type of construct. It's here's a customer, here's their needs, here's their context.

34:33Asheesh Gupta:And then the last one, Alex, that's really interesting is how do we find connections? So one of the things that we've done internally at Audax is we take our entire talent database, which today has almost 50 ,000 executives, and all of our connections to them, meaning all of the interactions we've had over the last dozen years. And so if somebody says, find me an executive in the pest control industry, today we can go find that executive and how we're connected to that person. That's what we're deploying into the portfolio as well. So portfolio companies say, hey, I'm trying to sell to this company.

35:02Asheesh Gupta:What is the collective wisdom of our network that can be brought to bear? And I share that as a difference in operating model, because in the past we were constrained by our perimeter and we didn't ask the question, hey, what if I had this data? What could I do differently today? You can get that data with very low friction. So I share it in the customer area, Alex, because that's the most exciting, right? That's the most impactful. We're going to see uplift in revenue and organic growth. But there's all manner of different areas in which the operating model will shift in terms of cost on the cost side as well.

35:31Young Lee:And so if we shift to senior year in the fourth stage, looking ahead to agentic workflows, what do you think is the most underestimated non-technical barrier to making that real?

35:42Asheesh Gupta:The connection across the boundaries of an enterprise, Alex. When we think about what's inside your enterprise, that we all protect and know. It's how you authenticate across the boundary of that enterprise. It could be even simply with your channel partners. It could be across your customer barrier, across your supplier barrier. Those are the areas where I think there will be a lot of security and privacy dynamics that will need to get played out over time. That's been our gating function is how do you actually manage those security information barriers? But I don't think it's a function of the technology not moving rapidly.

36:17Asheesh Gupta:In three years, the question you just asked, I think it'll be a very different answer of how more seamlessly connected we are to a variety of our counterparties. Would you provide some real world examples of perhaps what this looks like in terms of improving outcomes?

36:32Young Lee:Sure.

36:33Asheesh Gupta:Yeah. The way we think about it is I think there's a use case that I touched on earlier. There's a use case level of ROI. There's a the way we describe it as a functional. And in our taxonomy, we think about it as office all. So the officer, the CEO, officer, CFO, officer, chief, revenue officer, and then there's at the full enterprise level, right? And you think about that as like a digital transformation type of construct. What happened to revenue per employee, the top per employee at that level? On the use cases, you know, there's lots of use cases. We have a tracker that across our portfolio of 60 plus companies where we track every month what we describe as a library of anecdotes.

37:09Asheesh Gupta:So somebody says, hey, I built this capability, this use case, what's the benefit? And so two simple examples. One output is in safety monitoring. We have lots of portfolio companies that have large field-based employees. And safety is a complex topic. And it's one of the things I describe as the data that was being thrown off by those cameras and by those use cases was not being processed as effectively. And today you can process that very effectively. So a safety manager can actually look across hundreds of camera feeds to figure out, is there some systematic issue in a facility or in operations that needs to be addressed?

37:46Asheesh Gupta:I think of that as, you know, if it's safety for your employees and making sure you build a better environment as a remarkably valuable construct. When that same category, we have use cases in accounts payable. And those are like highly repetitive, right? Accounts payable, customer support, where today we're seeing all manner of ticketing systems get much more automated than they were previously. And so if you had a human in the loop to do ticketing or to do data entry, that's all being summarized and synthesized. So instead of doing, let's just take 100 data entry transactions per time period, you are now doing two to four because the system is doing 90 plus percent of them.

38:28Asheesh Gupta:And you're only doing the exceptions where the errors are. That's the piece that's really exciting, Alex. We're at the early stages of seeing those ticketing systems shift into what we call conversations, right? So it's not a one-way ticket, it's a conversation. So accounts payable, HR, sales, again, those highly transactional, either user or functional use cases is where we see the biggest uplift. As you know, as I mentioned earlier, we want to drive better decisions. We're starting to see a bit of that better decision-making happening. If it's in pricing, if it's in contracting, if it's in terms, the value will accrue, the outcomes will take probably a little bit longer for us to see play through in our portfolio at the organizational level.

39:06Young Lee:Yeah, we're at Calibrate, where we are in the stage with our portfolio companies is we're helping them implement AI to a point where they're getting better and faster answers, but they're not yet at a point where we can use AI to help them get better and faster decisions. Let me zoom out for a second and ask you a couple of questions. How transformational do you believe AI already is today? And how transformational do you expect it to become as the technology improves and importantly, the adoption broadens?

39:40Asheesh Gupta:Yeah, I would describe it as transformational in terms of if I think about, let me take a very simple analogy. When I first came out of undergrad, I used to go to the Hertz counter, print a map to get to the client site I was driving to. Alex today, I don't know when the last time I printed a map, right? Like I use Google Maps, I use Waze, like I'm using that all the time. And I would share that word of transformation, I wish there was like this very bright line that we knew happened. It happens slowly. The number of times I'm talking to Chad GPT when I'm driving in the car, on the drive-in or the drive-back, that happens all the time.

40:18Asheesh Gupta:And so when I say it's transformational, it has become part of our ecosystem at our firm to a point where I'd have to think about what would we lose by taking it away. So the word transformation scares me sometimes because it feels like it's like, you went from freshman to PhD overnight. These things happen much more in evolutionary steps. But I would say today, the impact, the benefit is material. There's no question about that.

40:44Young Lee:If you think about driving a car, it's not that it's going to be a new mode of transport, at least the way I can think about AI for the near to medium term. But we're going from manual to automatic transmission, from automatic to cruise control to adaptive cruise control. So it's incrementally leading to a better driving experience, a much better, a significantly better and safer driving environment. But I don't think the mode of transport will be different. And you'll still have a driver that's in there. That's the human being. And is your general sense that we're in the early innings of this, you know, quote unquote transformation?

41:22Asheesh Gupta:I think the early innings, this is the analogy that I think we've all probably, you know, heard and tried to process this idea of we have companies born on steam moving to electricity. There'll be some companies born in electricity, right? So I'll give a much simpler example. When I joined the firm, we'd buy companies that were using some ERP system and over time move to a tool like NetSuite. Today, we're buying companies that are already on NetSuite. So we have to show up in a different way for the companies that aren't going to spend a first year putting in an ERP system because they already have it.

41:54Asheesh Gupta:Now, just to be clear, it's probably not deployed. It's not fully optimized, et cetera. So the motion has changed. And so I think that's what we're going to see is that where we invest in the middle market, lower middle market, the companies we're buying are either engaged in this topic. They're not proficient. They're not fluid. They're engaged in this topic. They weren't two years ago and we're helping them. So we're trying to match each company where they are. And I'd say that is still the early innings. Again, based on the nature of what we invest, we're buying businesses that are almost always getting to a point where they're like, hey, we want to partner with institutional capital or with a firm like Audax because they're going to bring all manner of capabilities to our firm.

42:29Asheesh Gupta:And how do they do that systematically? And are they going to tell us? Are they going to ask us? Are they going to show us? Are they going to help us? Like, what's the persona with which they show up? So, yeah, no doubt early innings. It's funny, Alec, if you ask that question of people, they've been saying early innings for 30 years. I don't know when you'll ever say we're in the late innings. Has anybody ever said we're in the late innings of that? It feels like 30 years people have been saying we're in the early innings on this.

42:51Young Lee:No, the challenge is the innings reference assumes there's a beginning and an end. And it's a constant evolution. Yes. Yeah. And interestingly enough, Anishish, you and I have had this discussion before about the concept of AI and the appreciation that we effectively, AI has been prevalent in what we do for a long time, not just in the recent past several years.

43:19Asheesh Gupta:Hey, Alex, just on a hopeful note, I hope we're always in this mindset that it's the early innings because the possibilities that that conveys is different, right? It's sort of this curiosity, this possibility. And I think that energy, my hope is that we always have that, oh, wow, like we're in the early innings. What else could we do? What else could we do, right? So we're striving to make our companies better over time.

43:43Young Lee:And it is a good point. As she said, not only are we, again, using that analogy in the early innings of AI, of what we think the solution set could be with what we know, but using, again, that same analogy of electricity, where we're even in the earlier innings of this, what are the products and services that electricity, or in this case, AI, that we can't even think of in terms of solution sets. You know, that's exciting. You can think of the same thing with the internet. The internet came out and there was an initial use case and what it actually turned into is far beyond most imagination. So if you look across the roughly 60 portfolio companies that you work with, how do you think about standardization versus customization?

44:30Asheesh Gupta:One mental model that has served us well is if I just think about a simple grid type of format where we say there are tools that are horizontal and tools that are vertical. And the way we think about horizontal is like a general purpose LLM. Across our portfolio, there's a tool set we use, and we think of that as horizontal, and how do we make that applicable? But then because we invest in a number of industries in different business models, I think of business model as really the variable here that matters, manufacturing distribution, blue-collar services, white-collar services, tech-enabled services in tech, when I think about those business models, those have very different use cases by function.

45:08Asheesh Gupta:So the finance function is going to be somewhat similar, but the customer success function is going to be different. The operations function is going to be different. So what we've tried to think about is how do we get the horizontal layer really solid? And on the vertical, we want to get proficiency, but those each company needs to build capabilities. So what we've tried to do is build a, I'll call it standard of saying, hey, welcome to the Atax portfolio, here's a starter kit. We call it an enablement kit, an Atax enablement kit, where we hand them today, it's 14 GPTs. We say, hey, here's 14 GPTs that you can use, that we have built, and you can go deploy them in all manner of places.

45:45Asheesh Gupta:So guess what? It includes a prompt writer GPT, a devil's advocate GPT, a candidate review GPT, all the things that any management team that we partner with, we think would benefit from, right? How do you compare ops review deck one to ops review deck two, et cetera, all of those things. And because we hand that to them, it's a little bit like we used to 10 years ago. Welcome to your tax portfolio. You're going to accelerate your growth through M &A. Here's a 1300 line M &A integration playbook. Now let's customize it to your business model, to your types of dynamics. It's that same idea, Alex, that start with a package and then customize to the situation on the horizontal layer.

46:24Asheesh Gupta:and we found that if you get that horizontal layer done right, then each leader, each functional leader says, hey, I'm doing this, but you know, it's not perfect. I think we should find a better solution. And then they go down their vertical path that is much more bespoke and much more relevant to their industry or their domain where it's different expertise. And so that balance between a horizontal thinking and a vertical thinking has served us well. It hasn't been this like, hey, just go do Gen AI. That's a little hard to unpack, Alex, But this deconstruct has helped us split between standardization and customization.

46:57Young Lee:And I think that the philosophical overlay for us as a firm, as we think about our approach in partnering and collaborating with our portfolio companies and management teams is, it is what we call a pull, not a push model. And so to answer your question of, I don't think we would ever be in a situation where we're forcing a common set or one set of use cases or how they should be using AI. I think for us, the importance of that pull of wanting to have them ask and to pull in the use cases for how they want to use the tools, which is why it's so important for us in our journey, as Ashish mentioned, for us internally at the GP and within Audax as a firm to have gained that fluency within AI and that credibility so that when the measurement teams are asking, well, how are you using it?

47:49Young Lee:Are you using it? We can show them successful use cases. We can put them in touch with other portfolio companies and measurement teams who've had the successful case studies so they can make that decision and they feel that it's truly in the best interest of the company and the partnership versus having them feel like we're pushing something that they don't want.

48:08Asheesh Gupta:This is an important piece because I think part of our mental model has been how do we create a forum for that dialogue. And so one of the things that we have at Audax is what we call Audax Communities of Practice. We have three-letter acronyms. They're called ACPs. And so these communities of practice are where we bring leaders together. And so we have a generative AI ACP. It meets every month. And we invite portfolio company leaders to share both things they've done well and questions that they have. And it's in a forum of their peers. So we facilitate that. We seed it with some ideas. But that give and take that happens in the room, Alex, it's a little bit of calibrating across a 60 company portfolio where people can say, wow, you guys figured that out?

48:47Asheesh Gupta:How do you figure it out? What are the stumbling blocks you had? And you have that habit of every single month, let's jump on a quick call for the ACP. And we do this across many things. We do with CEOs and CFOs and CIOs. But generative AI has been one of those multidisciplinary areas where that cross-pollination, that sharing of best practices has really come alive over the last couple of years.

49:07Young Lee:And all of this goes back to where we start our conversation with the objective of enabling these companies to grow. And these are tools and philosophies and frameworks that you're helping incorporate within their businesses. Absolutely. So when you're deploying some of these AI tools across portfolio companies, have you noticed any adoption patterns that tend to predict long-term success?

49:31Asheesh Gupta:Yeah, Alex, maybe four things I'd mentioned. I think there's no substitute, as we know for a strong leadership sponsor. You know, when Young says he's using the devil's advocate, guess what? Everybody at the firm's using devil's advocate. Like it happens, right? We know this. If the senior leaders use it, it's actually just pulls the construct forward. So one is just executive sponsorship and demonstrated usage. Two, we find that if you can find the curious users, right? We call them power users, Alex, because that's a term that got coined. I wish there was a way we could say, let's calibrate curiosity.

50:03Asheesh Gupta:The curious users are the ones we want. In fact, they're a little bit frustrated with the status quo, but those curious users, you've got to go find them. You don't deputize them, you find them. And I say that very intentionally, Alex, because sometimes people are like, oh, yeah, hey, it's going to be this person in this department. No, no, no, no. Go find that person. You have to be intentional about finding the curious user to then make them a power user. Number three, have a weekly cadence. I think monthly calls, quarterly calls, interesting. A weekly cadence just becomes a heartbeat of, hey, what did we do last week?

50:29Asheesh Gupta:What are we going to do next week? What did we do last week? What are we going to do next week? It's very helpful. And the last one is in the age of everything is available at a couple of clicks in Google, actually having a human to human sharing. The idea that somebody says, hey, you know, this is what I built. What do you think? Hey, you have a problem? Hey, would this problem apply? And so that idea of creating those forums of sharing and people call it playbooks. I love playbooks. Playbooks are often a starting point, but then you have to call the plays. You have to modify the plays based on the situation.

50:57Asheesh Gupta:But finding those areas where you can actually have that forum for sharing. And that's why we've tried to think really heavily about the fourth one not being restricted to a portfolio company, but being across many, many portfolio companies to be part of the conversation. But you get the right leadership, the curious user base, a weekly cadence, the governance and weekly cadence I share because that's in all things that we do, right? It's things we do at the firm. So we're doing add-on integration. We have a weekly cadence. And then that best practice sharing, that idea of give and take, we found those four things.

51:28Asheesh Gupta:You get to some pretty remarkable outcomes very quickly.

51:31Young Lee:And you've talked about the devil's advocate GPT, but you also briefly mentioned the prompt writer GPT. Would you describe that and how it resonates with portfolio companies and with yourselves?

51:44Asheesh Gupta:Yeah, Alex, it's one of the things that if you think about the last three years, right, the titling has all changed. People say, oh, I need to hire prompt engineers, right? And people are like, oh my God, what is a prompt engineer? And one of the things we find is that you have to try to demystify some of these topics. So we wrote the prompt writer such that anybody who says, huh, I've never written a prompt before. How do I write a good prompt? Because when people say this hallucinates or doesn't hallucinate, we all know not a poor prompt and a good prompt will have differences in hallucination.

52:12and your, any individual's confidence is, how should I say, it's fragile.

52:20Asheesh Gupta:So if you build a prompt writer, you basically allow them to say, Hey, I've never done this before. Let me do something. Let me build confidence. And then surprise, surprise, you get momentum because they built a good prompt. As I mentioned earlier, they show the prompt. They say, Hey, I built this like, wow, that's a really interesting idea. Hey, let's evolve that. Let's develop it together. But that prompt writer takes away this like high hurdle. Think of it as like the catalyst or the activation energy to get people to start. That's how we've thought about that piece. And that's why we wrote the prompt writer and the devil's advocate very, very early on, because we found that senior leaders would engage differently.

52:51Asheesh Gupta:Senior leaders aren't going to write 300 GPTs. That's not their MO, but they're going to ask for 10 or 15 GPTs if they understand the lack of complexity in that, but you have to get started. And that's what we also spend a lot of time thinking about is enabling is one thing, but you got to build momentum. How do you build momentum? How do you build confidence? That's at least how we've thought about it.

53:09Young Lee:I guess one way to think about it is long ago, you needed to learn the language in order to get some of this output. Now it's reduced to common English, but you need, there's still a level left where you need to ask the right types of questions. You need to prompt it correctly. And this is a tool to, I guess, bridge that final gap.

53:30Asheesh Gupta:That's right. That's right. Because that feels like it is the, that's the friction that's left, right? Alex, if somebody says, hey, how do I get started? But also if somebody wants to build something and wants to have confidence to share it, we find it's a little bit like doing your work and making it available for the whole organization to see. You're a little bit hesitant about that. And the prompt writer gives you a little bit of that packaging. So the deliverable you share actually is at a good, healthy spot, right? It's not a overly raw document.

53:54Young Lee:What's the most counterintuitive thing you've learned about getting roughly 300 people internally to genuinely use AI as part of the firm's culture?

54:03Asheesh Gupta:It is so much more about the behavioral piece, Alex, than it is about the tools. I'll give you an example. Two years ago, on my iPhone, I replaced the spot in which I had my Google app. That's all I did, right? I just removed that from that spot. And I was so conditioned over so many years when I had a question, where did I go? And so I replaced that on my phone. Surprise, surprise, my use case library went from one set of tools to another set of tools. I got much more engagement. I had a way to save it, a way to share it. And I share that because I think as long as we keep in our minds that this is a tool that needs to be deployed in an organization, the organizational context matters.

54:46Asheesh Gupta:And so to keep that aware, and then obviously we're human beings inside an organization, we all have our own dynamics. that was the piece that I would say is more counterintuitive because it felt like build the tools, everybody's going to say, oh my God, this is fantastic. And they're going to come rushing towards it. But there is a little bit of subtlety to how you actually get people to engage. And I think that I wouldn't say counterintuitive perhaps, but it has been a little bit of that surprise of, wow, that's a very slight and seemingly easy adjustment that has a lot of benefit. And so guess what?

55:13Asheesh Gupta:A lot of our teams have made those kinds of changes where it becomes the default setting in a number of places and the use cases jump and it just feeds itself.

55:21Young Lee:Yeah, I think it's amazing how just the conversational element of it versus your standard internet web browser search engine. I think the users that were latest to the adoption were the ones that said, well, I don't need it because I'm not doing that complex analysis. I I don't need it for this complex task or it's too complicated or it's too clunky. And so what she said is what we've been doing selectively with some of the more, I'd say, persistent non-adopters is just subtly replacing their web browser on their phone with the chat GPT. And it is almost instantaneous when they start using it for the simple web searches.

56:09Young Lee:just given the conversational element and the interactive of it, how quickly they adopt and realize that light bulb moment. So it's been fascinating to see. It sounds like what you're describing is just incorporating it into the daily habit. And to break in isn't easy, but once it's in there, then it builds momentum.

56:28Asheesh Gupta:Organizational rituals are funny, Alex. They take a long time to establish. And just because you buy a bunch of tools, the organizational rituals don't suddenly change. And it doesn't matter if it's at the GP or in our portfolio, right? You have to sort of be cognizant of that in all things, not just generative AI. It's how they do pricing, how they build their sales teams. It's in all things. And I think that those first principles don't go away.

56:50Young Lee:Do you ever worry about AI becoming a crutch?

56:53Asheesh Gupta:I think this is a concern that we have to be intentional about in how we train folks. And I sort of reflect back on, you may remember, so again, as I mentioned, I have two kids, right? 29, 23, we used to joke about the Google effect, right? Kids didn't have to remember anything because they just had to remember, oh, if I can find it on Google, I don't need to learn it, right? And I think that can be a challenge, but I think it comes back to our conversation about is AI your assistant? Is it doing all of your work or is it a collaborator or a sparring partner, as we've talked about? And I think when you put it in that persona, you can't escape the conversation that happens in a room if Jung asks a vice president or principal, what do you think about this deal, right?

57:35Asheesh Gupta:That's a conversation in his office. If somebody said, we're not going to do those conversations, then I would worry. Yes, you just have people producing output that they didn't understand. But because you have to have that interrogation moment back and forth, I think that's how we try to figure out how to mitigate the AI tool usage becoming a crutch. You know, Alex, the reality is also with any of these tools that leave a digital fingerprint or a footprint, you see the usage and you see the usage data and the adoption data. And you end up finding the correlation to performance with certain folks where they're high, high users, but they're struggling in the dialogue or they're not producing quality output.

58:13Asheesh Gupta:This is just another data point that our talent team can use to say, huh, high user, but not performing. What's the disconnect? Are they not understanding it? Are they not engaging in it? So is it a concern? Yes. and how we intentionally respond to that concern is one that we've thought about, you know, for again, two decades as different tools have come in and replaced the proficiency of the user.

58:37Young Lee:If you could leave listeners with one practical mindset shift about AI, what would that be? I think it's similar to what we were talking about before with Ashish, which is AI is not a tool. It's certainly not a crutch. The most important use case of it is treating it as a teammate or that sparring partner. And Alex, I think you had said it first, which is the most important thing as we think about AI, because AI ultimately, it's not going to replace judgment or context or within what we do in private equity, the human element of developing trust, chemistry, leadership, as we think about it. And even at a higher level, philosophically speaking, questions around reputational judgment, moral judgment, ethical judgment.

59:31Young Lee:AI is great with the quantitative repetitive, and it'll answer based on the objectives and the frameworks you give it. But there is no one set of framework around any of those philosophical areas that I don't think we could ever trust AI to be. At least at this point, I can't think of anything. but it really is. It's, you know, for us, AI as a tool to help us seek truth. Very well said. Well, young Ashish, I appreciate you joining us, sharing all your insights and all of your experience in this area. I've really enjoyed it. I've learned a lot and I hope our listeners did as well. Thank you. Great.

1:00:13Young Lee:Very much enjoyed. Thanks for having us, Alex. Thank you, Alex. Thanks for listening. We hope you enjoyed this episode. Please visit our website at insightfulinvestor.org to access past shows and learn more about our podcast. If you have questions, feel free to email us at info at insightfulinvestor.org. And if you enjoyed the discussion, please subscribe to this podcast to ensure you don't miss future episodes. And don't forget to forward today's conversation to others you think would enjoy listening. Important information. This podcast is provided for informational purposes only and should not be considered legal, tax, investment, or business advice.

1:00:54Young Lee:It is not a solicitation, recommendation, or endorsement. All opinions expressed by participants are their own and do not necessarily reflect the views of the Evoque Advisors Division of MAI Capital Management, LLC, or Evoque, its affiliates, or any companies mentioned. Information shared has not been independently verified by MAI or its affiliates. MAI Capital Management LLC, or MAI, is registered with the U.S. Securities and Exchange Commission, SEC, which does not imply any particular level of skill or training. Certain information contained herein has been obtained from third-party sources, and such information has not been independently verified.

1:01:30Young Lee:No representation, warranty, or undertaking expressed or implied is given to the accuracy or completeness of such information by any person. While such resources are believed to be reliable, Evoke does not assume any responsibility for the accuracy or completeness of such information. Evoke does not undertake any obligation to update the information contained herein as of any future date. The content is intended for a general audience and does not constitute a recommendation to buy or sell securities or adopt any investment strategy. Any examples or scenarios discussed are illustrative only, involve risks and uncertainties, and do not guarantee future results.

1:02:05Young Lee:Non-traditional assets carry significant risks and may not be suitable for all investors. Decisions should be based on individual objectives, risk tolerance, and circumstances. Statements herein are general and may not reflect an individual's or entity's specific circumstances or applicable laws, which vary by jurisdiction. Further, speakers' views are personal and may differ from evoke and MAI recommendations and are not specific investment advice, and do not consider client objectives, risk tolerance, and diversification. Guests may have current or past relationships with Evoke and MAI, its affiliates, or the host, including as clients, service providers, or business partners.

1:02:43Young Lee:Participation does not constitute an endorsement or testimonial. No compensation has been paid or received for guest participation unless disclosed. MAI and its affiliates may have business relationships with entities mentioned in this podcast, which could create potential conflicts of interest. These relationships may include advisory services, investment management, or other arrangements. MAI seeks to manage such conflicts consistent with its fiduciary obligations and policies.