Why PE Firms Are Asking the Wrong Question About AI

Speaker A: We are calling it AI, but at the end of the day it's not really AI, it's an LLM. And a large language model by definition just echoes back what it seems in the past. If an apple will fall on the head of the server, uh, of an LLM, it will not discover gravity. We have that and we need to maintain that. It's easy to click the button, enable copilot, but it doesn't do much right, you get the tool, it's a great tool, um, it not being incorporated to how the company operates. If you don't go that deep you will just fall behind. There is no other option.

Speaker B: Behind every value creation plan there's a data problem nobody wants to talk about, fragmented systems, metrics nobody trusts and uh, decisions made on gut feel dressed up as analysis. Welcome to the PE Data Guy. Each week host Graham Crawford talks to the operating partners, advisors and practitioners who are doing the work inside portfolio companies. If you care about what actually drives returns in the market, then you're in the right place. The PE Data Guy starts now. Elon Sarfatti is the founder of Salfati Group, a Zurich based AI consultancy that works with PE backed businesses and enterprise operators to decode their expert knowledge into governed AI systems. Replacing manual people dependent processes with operating infrastructure that scales without adding headcount. Elon's background is in cybersecurity, scalable systems and applied AI research and he has advised the UK House of Lords on AI policy alongside leaders from Microsoft and Palantir. Elon has a contrarian view on what most companies are getting wrong about AI. Not do you have an AI initiative, but what changes in your cost structure and revenue capacity when AI runs the operational layer. And most of them will not like the honest answer. Today we're going to talk about the gap between AI initiatives that generate board slides and AI initiatives that actually change how our business runs, why that gap exists and what PE backed companies can learn from the ones that have done it.

Speaker A: Right.

Speaker B: So welcome to the show Elon. It's fantastic to have you on.

Speaker A: It's a pleasure to meet you as well. I'm uh, super excited to be here today.

Speaker B: Great esteemed company. I mean how different is this to talking uh, to the House of Lords for example?

Speaker A: You know, slightly more fun, it's informal but I think that we can make a lot of impact in our conversations in the House of Lord. And um, you know, at the end of the day, AI is everywhere right now. Um, so we really get a unique opportunity to change how AI consumption look

Speaker B: like, yeah, it's a technology that's going to shape humanity's uh, future, um, one way or another. And you know, there are a lot of different opinions out there at the minute. And as always with media coverage, you get the. Either end of the spectrum is the thing that gets the engagement. So either the absolute doomongering of like all white collar jobs will be gone in 300 days, um, all the way through to, oh, I asked AI what my name is and it didn't know. Therefore the whole thing is a sham. So as usual, either end, uh, gets the coverage. But I think you and I, as you know, practitioners and people who've seen how AI impacts businesses know that the truth is somewhere in between. And we're just kind of working out where that in between is and, and um, and how best to use it. So what, uh, what led you down this path? You were in cyber before. What, what led you to, to jump from cyber to, to AI? What did that transition look like?

Speaker A: Yeah, so it's funny, but I don't think I've uh, went too far away from cybersecurity.

Speaker B: No close cousins, right?

Speaker A: Yeah, exactly. Uh, but I think a lot of the conversations that we have right now, uh, around m. How should we leverage AI? And I think a lot of organizations today probably jump somewhere in between, uh, going all in. We will let AI, uh, do what it do and then we'll figure out everything else later on. And on the other side of the spectrum where you have the organizations that say, you know, it's the security risks are too high or I'm not going to do anything until someone else's video goes out, I, um, don't think you can be in either camp these days. Um, you cannot just ignore it. The company will die too fast. And I don't think that you can just simply go all in without taking anything into consideration. And what I found is I'm actually doing my PhD in AI CT right now at Imperial. And uh, what I found is that a lot of the work is around, um, how do we build AI in a way that can make it secure and how do we build cultures in the organization that make the usage of AI secure? And less about how do we just build more secure models. It's important nonetheless. But, um, this is where it kind of ended up going in that direction.

Speaker B: I remember I'm showing my age here. I remember when the Internet was booming and uh, all businesses were migrating towards the Internet and the impossible ask was out there then I was doing technology consultancy at that point. And um, you know, we want to be online, but we want to be 100% protected from uh, cyber attacks. And I, um, think as a wise mentor once told me, the only way to be sure you're 100% protected from online cyber attacks is to reach, uh, for the LAN cable or even better the power cord and pull it out of the wall. Um, because, you know, everything is always evolving. And um, I'm sure the same is true of uh, AI as well. And I think what you just positioned is another example of when we went in the opening of the extremes are always what gets the engage and the clicks. Uh, it's an unfortunate side effect of our current media environment. Um, but let's dive into that middle, um, and back to a framing that you use that I think I'd love for my listeners to hear. Um, and you say that rather than coming up with the wrong answers, most companies are asking the wrong question. So they're saying we want to be AI or we're going to buy our AI domain or we're going to add AI onto our company name, um, and asking what can we use AI for? When you instead think the right question and a more nuanced question is what is our operating model when AI orchestrates it? Walk me through that distinction and the nuance, um, and guide us towards the truth in the middle.

Speaker A: Yes, I see a lot of companies today, um, where the drive to use AI comes usually from the board. And when the board pushes it down it tends to be we need to do AI and it doesn't tend to be how should we do AI? And at the end of the day AI is a tool, it's a very powerful tool, um, but it's a tool and it needs to serve a purpose. And what we see when it comes from the perspective of we have to use AI is that we get a lot of point solutions. We will adopt an AISTR SaaS solution, we will adopt a custom ChatGPT, we will adopt a, uh, whatever cloud projects. Those are all valid tools, but those are being brought on top of the mocking process. And just adding more and more of those won't make the operations of the companies faster and more efficient. It would just make it more chaotic because now everything gets disconnected, things move in a different speed. And the most, um, call it successful projects that we have seen come from the other way they come from. We need to look at the processes that we have and what is it that we truly believe as an organization to be the tools, what is it that we want and Then how can we get to that, uh, end state? Through AI? And it can be those point solutions, but it tends to commonly be an orchestration layer on top of it that merges all these tools to become a much more proactive company.

Speaker B: Right, yeah, that makes sense. I mean, I used uh, a similar line in the context of data. Right. Often people come and say, we want to do better data, or we want to move our data onto, um, new infrastructure and think incorrectly that my team is going to show up with some, And I'm sure you get this with AI, some secret menu of tools of like, aha, this is the secret menu of tools, the holy grail, the hidden document. Um, if you put all these things together, then suddenly your business is going to explode. Um, data like tech, like AI, is not a business case in its own right. It's a tool, as you say, and a very impressive and rapidly improving, uh, tool, but a tool nevertheless. And it has to serve a purpose. So if I take it back in the context of private equity, back to that initial board request, like we want you to implement. AI boards usually want one of three things, and it's, we want you to get more revenue, we want you to become more profitable, or we want you to mitigate some risk, um, that you have in the, that you have in the business. And I think it's the, in terms of the operating partners and the um, executive management layer that's the tricky bit because it falls to them to say, okay, within those very broad strata of making a company more valuable, what exactly do I do? And I think the difficult first step is we have to admit, um, that our data or our processes or something is broken in the first place in order for us to accept that, we have to fix it, um, before AI would work. And, um, I've seen that play out politically in very interesting ways, um, internally, um, especially companies with longer tenure because, you know, folks who have built careers or promotions or reputations on the back of certain systems don't love, um, someone coming in and saying this is inefficient or this isn't working. Like, um. How do you navigate around those nuances of sometimes, like data, sometimes AI reveals things that people would rather, um, went unheard.

Speaker A: Yeah, it's a great point. I think that, um, we are working closely, uh, with, um, the person who used to be The CMO of PepsiCo, Blackhawk News Corp. Uh, she's an amazing, uh, brand leader and narrative builder. And we tend to work really well together, mostly because we understand that, uh, the technological side of it, while important and critical to get right, is just half the picture. And it never works just by itself.

Speaker B: Maybe even less than half, I would say.

Speaker A: Oh yeah, that's so true. Um, and really, to get any of these projects done well in any of this transformation, we need to build a narrative both internally and externally for the organization because a. Exactly. Like you said, executives don't really like to hear that things aren't just working well. And I think we have in the AI space some leverage because AI is a new technology and it's not that their processes don't work well. It's just there are new tools in

Speaker B: the market or unsuitable for use with this technology that you could never have understood would exist when you made the process in the first place.

Speaker A: Exactly. And then I think on top of it, you know, there are other criteria. Right. Because there is like the executive, uh, buying, but there is also a lot of like internal buying. No one. If you go to AI transformation with the assumptions that you are just going to fire everyone, you don't get buying from the team. It doesn't work.

Speaker B: Yeah.

Speaker A: And it's also not the goal.

Speaker B: What, what, what do you just on that as soon as it's hot on the news at the minute. Uh, I'm sure you saw the news this week that Matter has started tracking, uh, keystrokes, employee keystrokes, so they can understand more what the employee does, so they can teach AI. Like, what do you make of that move?

Speaker A: That is, I saw somewhere that, um, in France, uh, the name Claude is very common and. Oh yeah, of course, 99% of them believe that they are going to be replaced by AI. Significantly higher than other people, um, of

Speaker B: the people called Claude.

Speaker A: Yeah. Um, and I think it's the same. Right. The moment that you are becoming a tool that is being targeted to be replaced by AI, uh, you're not going to get a lot out of it. And I also don't think that this is the right way of managing. Mostly because at the end of the day, you don't hire the people who hands on keyboard, do the work just to do it. You don't hire an engineer to write code. You hire them to solve hard problems. You don't hire.

Speaker B: You have to work out what code needs to be written in the first place.

Speaker A: Right, exactly. The code piece was always just a tool. You don't hire a sales rep to just do calls or put data into the CRM. You hire them to be strategic and be creative on how they get people's attention.

Speaker B: Yeah. I think that in my experience, by the way, most of most of them don't put data in a CRM. But, but back to our point that you were making, like, this is a great use case because you could, you could be an amazing salesperson. Of course, you know, typing the data into the CRM is not the most important part of that role is building relationships, building trust. It's identifying problems that client. Potential clients have and finding ways to convince them that your help is going to be valu beyond the cost of whatever Azure product or service costs. But the, um, AI as a magnifier can maybe capture those calls and automatically update the CRM. And that makes the best use of the capabilities of both the human as well as the, uh, model.

Speaker A: Exactly, yeah. And we have done that internally, so it works really well. We are very proud of it. Uh, we're trying to shift how we operate for ourselves exactly what we are pitching. Um, and we're trying to flip the narrative and we want the company to activate us rather than us. Ah, activating the company.

Speaker B: Yeah, yeah, we want to do AI but, uh, for what? Right, yeah, for what purpose and, um, what reason? Um, one of the things that we talked about prior to this was this concept of, uh, transformation theater. And I see this in data. It happens in technology more broadly. I'm sure it happens in AI as well. Um, and this is the flow of, as you say, the board asks for AI, the company press release says they're going to do AI, but yet nothing actually changes, uh, within the company, with the exception maybe. Well, let's give people a pass because the extent of some AI strategies I've, uh, seen. Sorry. Have been. We activated Copilot for everyone in a Microsoft shop. We didn't tell them how to use it or where the boundaries were or what it works well for. We pressed the button in the, uh, admin console and turned it on for everyone. So why do you think that that's happening? Is it the, uh, do we need to, like, grow the capability of management teams to convert DO AI into this is that this is how we want to do it. Why do things stall between the, you know, the intent and the execution?

Speaker A: Yeah. So I think it tends to be a delve, a few forces pulling an executive trying to make that decision. And everyone wants to see value early on this whole 90 days POC saga. And on the other side, there is a lot of fear. I'm not going to pour, uh, 5, 10, 20 millions into an AI project and hope to get the best. Um, so this in between problem going back to our lvl conversation, I think it's the biggest challenge because it's easy to click the button, enable copilot, but it doesn't do much. Right. You get a tool, it's a great tool. Um, it not being incorporated to how the company operates. If you don't go that deep you will just fall behind. There is no other option. Um, and it's scary. I get that it's scary. It's going through a big transformation.

Speaker B: Yeah.

Speaker A: And I get that it's not, not a 90 day project but this is where we are, right?

Speaker B: Yeah.

Speaker A: It's coming either way.

Speaker B: And the other thing I've seen, particularly in my corporate career, um, 90 day proof of concepts or whatever they might be or pilots, um, are occasionally used as vehicles for executives to like, you know, satisfy eager employees who want to try something out without actually ever committing budget or resources or priority to something. So it's one of the things that I always advise companies. If you POC or a pilot, fine. But there has to be baked into that a decision point at the end of like if this result from POC pilot then we take this course of action. If this result then we take this other course of action. There has to be a decision baked into the end of it. And I think that's happening a lot of places as well. You get some m very AI forward self starting, proactive employees who want to do something and they're given um, permission to do a prototype and then it just goes and dies on the corner somewhere because it never really gets the, the backing. Um, I want to move from theory to practice and specifics. And you have a great uh, case study from uh, Kloop. And you and I work with um, private equity backed businesses that are facing the very same structural problems that you came across um, at Keloop. Can you walk me through that example of a successful transformation program and give us the indicators and the things and the conditions that made that a successful program where so many others fail.

Speaker A: Yeah. So I think a successful program really starts with a key question which we use as a test for any projects that we take on. And we tend to structure it in this way so we say um, if this specific AI effort works, whatever the effort will be, this specific metric changes by this amount of, by this whatever amount, ah, during this period. And I think that when you try to put it to such a short sentence it forces a lot of organization to think about what that uh, product or effort looks like. And exactly like you said it Puts a target decision at the end of a poc. Um, and I think this is where a lot of good pilot program and good transformation start. Um, and we have done the same with Keyloop in partnership with other firms. Um, and the idea was, can we get um, churn and retention churn reduced so we can increase retention? Um, it also meant that we completely rethought the process. We drew up everything, we started to map out what are the human moments, what are the conversations and the meeting and the in person touch points that have to happen, the strategic creative points we want to protect those, uh, what are the judgment moments? What are the points that we want a human to come in and say, well yeah, I was wrong, this is not what we intended. And what are the hands on keyboard putting data into stuff, uh, creating reports, all of those things that we don't really need a human to do anymore. This allowed us to completely re change what the process looks like. And the pilot came hand in hand with a change management effort. It by itself won't work.

Speaker B: And um, that's fabulous. Are you able to share some specifics or even in general terms of what sort of outputs were you looking for? I mean it's a dangerous thing to measure perhaps on decrease of human interaction. Is it number of sales calls, is it qualification rate? This was um, a revenue engine. Right. So what metrics did you go after and succeed, um, in pursuing?

Speaker A: Yeah, so, um, the key metric that drives the entire program was churn. And we were lucky enough to work with the uh, CRO to kind of define what this churn looks like and where it comes from. And we actually identified that a lot of the churn come from the duration that it takes them to engage with customer requests. Um, so we were able to completely change how the engagement looked like. And the key metric was that can we reduce this uh, churn number by a certain percentage year over year? And so far we are really to the program and we see really good success. We also see pushback, but really good success as well.

Speaker B: And that's because you identified that time to respond to a customer request was one of the key factors and have implemented AI to reduce that time presumably.

Speaker A: Exactly.

Speaker B: Um, so you've got your leading indicator of time has reduced and I guess if you're a year in and year on year, you're starting to see early results and things look promising.

Speaker A: Exactly. Yeah. But we also see a lot of challenges. Right. So for example, we see that ah, um, the technology moves so fast these days. Like a week is a Lifetime. And corporate decisions don't happen in a week.

Speaker B: No, normally.

Speaker A: So you have these really interesting forces structuring the program. Right. You want to have a decision but then by the time you made the decision the proposal is no longer relevant. There's a new technology and a new AI model and new opportunity.

Speaker B: So again I think the M mistake I see this when people are looking at data and analysis as well as human beings we're very prone to look at what does this human do or what does this human role do. And you know, for example within a marketing team like you might find someone who's very good at uh, graphic design but they don't know about technical SEO. Right. And vice versa. But you know an AI doesn't have those limitations necessarily. So instead you have to, I've found you have to think about data and then subsequently AI in the context of jobs to be done as part of the workflow. So sort of doing that breakdown of um, workflow is super important. And I think if you have that core understanding then you become a bit more um, resistant or mitigated at least to suddenly there's a new AI model that comes out of nowhere that's suddenly the best thing ever. And I think as long as you understand fundamentally your business and you're not leaning on the tool to be your competitive advantage then I've seen people flip, including my business as well. Like I've, I'm one of many that's flipped my kind of backend operations from GPT to Claude over the last three um, to four months. As you know, anthropic came flying out of the gate in 2026 with you know, paradigm shifting capabilities. Um, the keylook that we talked about is a slightly bigger company. If you imagine for example, a PE backed company that's you know, smaller, more in the kind of 10 to 20 million revenue range. They don't necessarily have the resources. They might not even have a, you know, a fully fleshed out technology team in a company uh, of that size, you know, which makes two year transformation programs very hard to resource. Right. Understandably. So say you take your keylook example and then try and apply that in a smaller company. Like what are some of the critical must still be on the table principles that you'd have people apply.

Speaker A: Yeah, so we are actually also working with small LP firms. Um, so we have concrete examples uh, that we can talk about. But um, at the end of the day the key principle is that it needs to restructure how the company operates. It cannot just be another custom GPT or another chat interface that, that doesn't really do the job. What we want to get to is a new way of thinking, right. We want to empower everyone on the team to be creative and strategic. And the way to do that is by not by just fixing the key problems or just putting data, uh, somewhere, any of those. It has to be the backbone, it has to be the thing that operates the company. This is how you get meaningful change. And it doesn't have to be a 20 million dollar project. It can easily be something that you see early results very early on. We just need to find the right partner. Right. And I think a lot of the business had changed today.

Speaker B: Yeah, um, it's just further proof stacked behind the point. I think we're both violently agreeing on here that, you know, if I'm, if I'm in a company and my role is like, I wait for an email report to come through and I paste it into a spreadsheet, I do some analysis and I paste that over somewhere else. Uh, in theory, you learn that you put cloud cowork on that, learn the workflow and then replace it and then call it Graham GPT. Right? Maybe you get some efficiency, maybe you add some risk. But what you've done by taking a human centered view of it is blow away the best part of the opportunity to take that risk. Jobs to be done, workflow centric view. And I think the, you know the phrase you just, you just mentioned, um, it can be very emotionally loaded. The phrase was, um, we need to reimagine the whole way that the company works. I don't think there are many employees at companies, big or small, who don't hear that as like I'm going to get fired or I'm going to get, I'm going to get, I'm going to get laid off. Um, so, you know, how do you. But that's not the intention behind it. You and I know that. But it's important that you consider the paradigm shift to be so large in order to be successful. So how do you, how do you cross that bridge with management teams, with employees that you're talking to as part of these, um, implementations? Because inherently that's a scary thing to hear, right?

Speaker A: Yeah, and I get why it's scary, I really do. Um, but I think a lot of companies or a lot of individuals, right, they, the more they adopt it, I actually think they are more protected rather than less. Um, the people who will eventually go away are the people who refuse to adopt and halt the organization. And this is usually when layoffs happen, but really when you start to adopt it and when you start to go into it, you understand. And sometimes, uh, you need to remind yourself that the value that you bring into the organization is not the hand on cue ball.

Speaker B: No, no. It's not the pushing the button. There's a very famous episode of the Simpsons. I don't know if you've seen this one where, uh, Homer replaces himself with a chicken. This keyboard that is just like every so often it pecks down and presses a thing and then. And in the end, uh, even Homer is not replaceable by the chicken. AI because the nuclear reactor blows up. Um, but yeah, I fully agree. And m. You know, becoming a single point of failure or a single point of success in any system can feel good because you can feel needed in the short term, but actually you're, you're holding back the success of the company. And it's certainly something that we always look to. It's a prison that we always look to release people from because often their potential is largely unrealized if they're in that situation as well as the company's potential, uh, in turn. So, uh, next, I want to come on to one of your favorite phrases. Decision sovereignty. Sounds better with an American intonation, um, than my British version. But you use that as a core principle of how you build AI systems. Um, it sounds important. Decision sovereignty. It sounds like something that should be on a due diligence report, on a risk report that goes to an investment committee. Can you explain what you actually mean by decision, uh, sovereignty and why it would matter to a PE backed company or a PE firm?

Speaker A: Um, yeah, that's the one close to my heart. Um, I see two types of organizations, um, one of which is a company that just shifts the entire decision making process into an AI. And then you can see executives get super excited by the fact that they were able to produce a strategic, uh, marketing plan for 2027 with a click of a button. What they don't get is that by definition, LLMs are echoing back what they have seen in the past. You were not able to produce a new plan. You basically got a mix of five

Speaker B: other plans, maybe 500 other plans, but still a mix of pre existing.

Speaker A: Yeah, exactly. And that just directly means that if 1,000 companies would request for the same plan, half of them will get the same. You're not creating differentiation by bringing a strategic plan through an AI. This is just a hidden killer. And it's so hard to see.

Speaker B: I've, I I've seen it. And you know the, we had this, we had this phase maybe a year ago so, maybe 18 months now. It's hard. Time flies when you're having fun, right? But where prompt engineering was a thing and companies were literally hiring prompt engineers and already the technology seems to have blown past that. Um, but that doesn't mean that context isn't important. I'd say actually it means context is more important than ever. And what I always tell people, and this is true of data as well, which is why I love talking to AI folks because there's so much crossover and we're even closer cousins than uh, AI and cybersecurity I would argue. But um, if you just use AI in a very simplistic way, out of the box and give it all the decision sovereignty, then you can't possibly claim it as a competitive advantage because anybody else can do exactly the same thing. Right? It's how can you engineer proprietary context into it? How can you engineer unique insights about your customer and your market that you've learned over the years serving these customers? How can you bake in your proprietary industry expertise that you've learned from years of building products in this particular industry that the AIs don't have? Right, the AIs don't have. And I say um, even to folks who are supplying, you know, software engineering resource, for example. Right. Ah, like junior software engineers or offshore software engineers. Um, you could argue, and I've seen it argued very clear um, very clearly in some places that, oh, it's easy to replace that with, with um, you know, AI, even for a non technical person. I've seen Technical people vs non technical people vibe coding completely different results because there's. And software engineer really distills this nicely because software engineers still think like software engineers because they've been trained to do so. They think in terms of, you know, in an object oriented way. Right. So they think of the classes they need their functions, they need the variables they'll need someone who just retired from the marketing career, has never touched a line of code, doesn't think like that and therefore their vibe coded uh, platform in my experience is that doesn't touch what a software engineer can do. So I still think um, beyond context, there's a lot of value in, even if you're using the tool off the shelf, you can be a lot better at using it than the next person. And there's still differentiation for the humans there.

Speaker A: Absolutely, yeah. And I think that it's also about the execution and the way that you think. And it's also a lot about your ability to evaluate something. And I actually see exactly the software engineering context where it bites. A lot of companies where they don't see it, um, I've seen a lot of executives go, they vibe code something and they say hey, it took me like two hours. I sat down with Claude and here is a product. Why does it take the engineering team a month to do but they just don't have the capabilities to evaluate it? And this is where it kind of goes very long. And this is where you need the expertise.

Speaker B: Yeah. Uh, to draw equivalence between software engineering and the data world that I live in. The same thing people think data engineers um, and data product managers spend uh, the majority of their time pulling together nice looking visualizations when in fact that's the prize at the end of the 98% of the hard work that goes before it to get the data in the right place and the right position, properly defined and tagged, stored, aggregated, all the things that need to be true. Um, and yeah that's the danger of the vercel front end um, or base 44 whatever it is people are using, um, front end that it literally is just a front end. And you and I both know the back end of the integration work is actually where um, the hard work lives. And that front end by the way is probably a purple themed sensory font square button looking thing that's the same as everything to your point about marketing plans. The same as a vibe coded front end that everyone else is vibe coded as well. So there goes your competitive advantage again. It's um, yeah it's a fun time. Um, there's a um, lot of people talking about this future of a human with an army of agents and that's where the job market's going up. Like you're not just hiring a person, you're hiring a person plus their, plus uh, their agent or their agent swarm M or their agent army, wherever you want to call it. Are we a flipping point on, on that term? Is this the root of the problem that stalls most AI initiatives? Um, and how would you explain that to again and like the non technical people are super important. There are many complementary tasks and skills that go to making something. So a partner at a PE firm, like how do you, how do you help them grasp the gravity and the importance of this term M a human with an army of agents in terms of how the employment market is going to move.

Speaker A: Yeah. So I think we are at a big flipping point around that um for me A human with an army of agents. And then you can really like we, for ourselves, we create some sort of like a heat map of organizations. Are you afraid of AI? Are you kind of just playing along with ChatGPT and Claude? Are you going to this human with an army of agents or are you at the next tipping point, which for us is an army of agents with a human?

Speaker B: Ah, uh, right.

Speaker A: And I think the easiest explanation though is that we want to get to a point where the company is operating when you're asleep. And right now most organizations need to do something to trigger an automation for something to happen. But it's very human initiated. The human trigger initiation step. Yeah, triggering step. This initiation step doesn't have a lot of value to it. Right. Ah, you do it because you have to do it. You put something into cell, uh, because you have to put it. Um, but what we want to do is we want to flip it around. We want the company to reach out to us and say, this is the decision that we need from you. This is the creative judgment moment. This is the strategy that we want from you. Going back to the uh, decision sovereignty point. We want to get from you, the human, the capabilities to think, to be creative. And again, we are calling it AI, but at the end of the day it's not really AI, it's an LLM. And a large language model by definition just echoes back, ah, what it seems in the past. If an apple will fall on the head of the server of an LLM, it will not discover gravity. We have that. We need to maintain that.

Speaker B: That's a beautiful analogy. I love that one.

Speaker A: Um, so how can we get there? And for me it's just flipping it around, getting the company to reach out to us.

Speaker B: It's, you know, a lot of things have snuck into the AI umbrella, which is the other reason that doing AI is not really a good, um, a good description because the LLMs are their own class of AI, right. Obviously with the predictive modeling and the um, the way the prediction engine works, which still, by the way, most humans I talk to, um, don't grasp that AI isn't an encyclopedia of everything that's ever happened in the world. It's just a very good predictive model. It's good at predicting what humans want to. What, what the next word should be in the sentence. Um, the, um, Business process automation is not new, right. I, I continue to age myself in these, but I've been writing Cron Jobs in UNIX since the late 1900s. Right. And I think that the Reality that's catching up with people is it's become harder to deny that your business processes aren't automated. That's not the true unlock that's happened with LLMs and all the latest technology. You see it like synthesizing new proteins and all sorts of wonderful things. Like that's the real power of it. And I think still so much of the business world is still catching up and is actually using AI to do business process automation more easily. Like would you say that's. That would be fair. And there's two different classes of use case here. Like there's the automation of business process and then there's the power of LLMs as a separate bucket.

Speaker A: Absolutely. And I just think that it's magnified. Right. The problem is the same. Um, the opportunities are slightly different, the impact is slightly different, but it happens with every new tool and it's just a new tool.

Speaker B: Yeah. Where would you start? Like, I want to bring this, this like fully into frame for my audience. So say you're a general partner at a private equity firm and uh, you know, your LPs and other investors have come to you with the phrase of doom. We want you to do AI, um, to make this business better. Like, where would you, where would you recommend those people start?

Speaker A: Go as high as you can in the way that you think. Um, I would argue that for most traditional private equity firms, the game and the plan is not to do AI at the portfolio level, is to start to think at a higher perspective. So let's assume that this portfolio company has already went through the transformation. What does the acquisition play playbook look like? Right. And how do we take smaller companies and do like this, tuck in strategy and bring them into this ecosystem significantly more efficiently? Um, are there opportunities to. Um, we tend to work a lot with, uh, private equities in the services space. In the more like traditional spaces.

Speaker B: Oh, like home services and the like?

Speaker A: Yes, exactly. And I love this mostly because you can turn a lot of all those service companies, um, so even take like lead generation and all these marketplaces and you can transform them into a software company. Like you can completely change the way that you do, um, multiplications on the valuation. I would suggest most general partners to start to think at those terms.

Speaker B: What would the steps be? I mean it sounds a bit like uh, turning water into wine. Right. Turning a plumbing business into a software company. Like, what would the high level steps be along the way to do that?

Speaker A: Yeah. So let's take this example of plumber, um, which is definitely not our uh, commonplace, but I think it's an interesting one. Um, you could easily turn a plumber services company into an AI. First software that hires plumbers on itself where the AI ah, spins up a marketplace, hire the closest plumber next to your house, send out the order out and take a big margin on top of becomes a fully serviced company. And then any hacking that you do, any smaller acquisition that you do is just transforming the new company back into this backbone and you generate significantly more value.

Speaker B: I see, yeah, you become the place and you focus in the value of that software then becomes removing the friction from the I have to search for a plumber, then I have to call them, then I have to get them here. And it just takes care of that orchestration layer if you like.

Speaker A: Exactly, yeah.

Speaker B: Uh, that's interesting. And the idea is of course the more plumbers you can then tuck into the back end of that, the greater the reach, the more efficient you get, the higher the margin, the broader the reach. Higher customer lifetime value, lower customer access. Yeah, I can see how it all rolls downhill from there. Um, yeah, that is interesting indeed. Um, search plays a big, local search plays a big part in those business. I feel that local search has not changed as much as kind of national or global search. From a Google perspective, uh, are some of the same playbooks still working in there?

Speaker A: So that's uh, just a quite interesting question. I think that a lot of the work that we have done was kind of look more on the operation side rather than this, ah, outreach.

Speaker B: So yeah, rather than marketing side. Yeah, no, for sure that makes sense. So you're um, if we're thinking about summoning plumbers then through that um, orchestration layer is the AI like making phone calls on your behalf, like sending emails, booking jobs, managing your calendar, that sort of thing. Is that what we're talking about?

Speaker A: Yeah, I think that the moment you turn this company into a software company it means that the plumbing itself, which in AI cannot replace these days, um,

Speaker B: good job, good career, we should all go into plumbing.

Speaker A: Um, we can get to a point where we can scale significantly higher because it means that we can have an AI put up a job. It becomes gig economy 2.0. Um, hire a human agent rather than hire an AI agent. Um, and then it becomes much more automated. You put out the deal, a planner, um, picks it up, puts it on their calendar, everything is pre baked

Speaker B: and then you can handle the financials, track the invoice, have a wonderful customer portal that not many plumbers have got. So you can go and download that later. And so it becomes true that Elon's army of agents with a human comes to pass. Right.

Speaker A: Really does.

Speaker B: That's how we get there. Do you think, uh, it's easier to see and imagine that in the home services industry? Do you think that moves into the world of white collar knowledge work as well over time?

Speaker A: I think so. Right. Because this is where you generally like the service industry, not just the home services, but the services industry is where you get a lot of relatively old ways of thinking in terms of operation, but you still really want the human component. Right. It doesn't matter what the service is. I, at the end of the day, will trust a human to take accountability for something over the machine.

Speaker B: Yeah. And it's going to be required as well, especially in particularly litigious, uh, countries like the United States, for example, where there has to be a lawsuit for everyone to file. If something goes wrong, then nobody's going to be able to sue an AI. And I'm sure, sure, um, Sam Altman isn't going to find, uh, himself in court because GPT recommended something. I'm sure he's well enough protected from that in the terms of conditions. Um, I want to finish Elon with a fun question. Folks like you and I who are, uh, kind of embedding AI in our, uh, work, often let it bleed into our personal lives as well. What's the most fun, uh, way you've used AI in your personal life, either for fun or for health or whatever it is?

Speaker A: Ah, that's such an interesting question. We have gone quite recently through the big change in the company, and we completely flipped how we operate. And I took that flip together with me to everything that I do in my life. Um, and we have like a fun small robot that we ordered recently, and we connected it to the AI of the company. So then it goes, you open the door, kind of look at you. It knows everything around the company, uh, which brought it closer physically to us.

Speaker B: Uh, and does it walk around the office?

Speaker A: No. So it's like this small, how do they call it, Richie? Mini form, hugging face? Um, I don't know if you've seen it around. So it's kind of sticking in the

Speaker B: desk, so it's static. But it's always watching.

Speaker A: But it's always watching. And it knows everything in the company, so it has the whole knowledge. So you can kind of go, hey, what are the agenda for today? And kind of looks at you, knows who you are, and then spits out everything that you need to know. Uh, which Wonderful and terrifying.

Speaker B: Wonderfully terrifying all at the same time. The technology's always been there. I mean, even my Amazon device, whatever it is, that sits in the kitchen, like, I can't walk past that without it springing its text on the screen, like, hello, Graham, how can I help you today? Uh, here's your calendar. And it puts the calendar. So it's just. It's just that taken to a. To an extra degree. And it's funny, isn't it, with humans how you take technology that exists and isn't actually that high tech, but then you personify it by putting in a vaguely humanoid form, and suddenly it becomes scary.

Speaker A: It's.

Speaker B: There's a weird thing in human psychology there.

Speaker A: I think it's the same magic that we got with LLMs.

Speaker B: Yeah, exactly.

Speaker A: It speaks our language. We understand it.

Speaker B: Exactly. You can communicate on our terms as if there's no other way to, uh, talk. Um, cool. Elon, this has been fascinating, and I'm sure folks who've been listening and hearing about the work that you're doing within portfolio companies and with PE firms, like, I'd be super excited to learn more. Like, where should they go on the Internet? Um, or what search terms should they put in ChatGPT to find you?

Speaker A: So probably if they would go to ChatGPT and search for any changes in, um, AI transformation in the PE ecosystem, they would land on us. Uh, we try to put a lot of content out there. Um, but if they go, they would search for me, they would search for Safati Group. They would find us, um, straight away.

Speaker B: Wonderful, Wonderful. Well, um, I'll put your contact details in the show notes below as well, for people who want an easier way to find you. Um, and then I'll add another backlink, uh, to your arsenal in the process, so that'll be good. Um, thanks so much for coming on and for a great conversation and, uh, look forward to, like, I think we're working in parallel channels to kind of make this technology accessible and valuable to as many people as possible. So good luck on your, uh, journey in that channel.

Speaker A: Thank you. And thank you for having me. It was a pleasant conversation.

Speaker B: Likewise. Take care. Uh, thanks for listening to the PE Data guy. The place where private equity meets data. Please forward this episode to your favorite private equity friend. Thanks for listening. See you next time.