Miles Rowland: Why Every Portfolio Company Needs an AI Engineering Team

Speaker A: Hello. Welcome back to the AI Pathfinder for Private Equity podcast. AR Pathfinder helps private equity firms make sense of AI and make it work. It's an expert network built on insight, experience and connection, bringing the right people and ideas together to turn AI from something firms are curious about into something that delivers real results. If you'd like to attend one of my regular AI strategy briefings in London and New York, please check out the show notes for details. This is the fifth episode in uh, our AI Operator series. Each conversation has revealed how differently firms are structuring the role. Different titles, different mandates, different tools, different backgrounds. Today's guest adds another important dimension. The Transatlantic Mid Market Fund and a firm where he was the very first AI and data hire. Building the function from scratch My guest is Miles Rowland, Head of Data and AI at Searchlight Capital. Searchlight is a mid market private equity firm with a workforce spread across Europe and the us Miles has a distinctive career arc for this space. He's built data analytics function from scratch not just once, but three times. Starting his career as a strategy consultant, he went on to establish the data analytics practice at Leke Consulting, then established and ran the business analytics team at SHL before joining Searchlight in 2022. Miles, welcome to the podcast.

Speaker B: Hi Steve, thank you for having me.

Speaker A: Looking forward to getting into the conversation. Miles, before we get into Searchlight, can you take us through the path that brought you here? You built the data analytics functions from scratch three times. Now what drew you towards this kind of build from zero work? And how did private equity become the next chap?

Speaker B: Yeah, so I suppose I started my career in management consulting and I was um, straight out of university, thrown into all of the experiences that you get from working with a variety of clients across different sectors. But one thing I suppose that we observed, and this was back in the early 2010s was so many companies had lots of data that they just weren't making use of day to day to drive decision making in the companies. And that leaves a lot of space for consultants to come in and you know, do work on companies data and generate novel insights that are helpful for taking strategic decisions. I think what we found rather than lek, was that the technology was emerging such, you know, I'm talking about tableau and databases and Alteryx tools like this which allowed us to now uh, work with much bigger data sets than we ever could before and deliver better insights faster for clients. So that for me was you know, an interesting opportunity back then. And uh, I think the pattern that I've Seen since then has just been so many companies still, even today have that same challenge of having lots of data in lots of systems, struggling to unify it, struggling to bring it together to drive decision making day to day. So for me personally, I've always enjoyed greenfield opportunities and building something where there was nothing before. And that's honestly been the pattern across a lot of the companies that Searchlights invested in as well. A lot of companies we invest in don't have great data maturity when we first invest, so there's always been a lot of opportunities to do that. So yeah, we spent a lot of time over the last few years helping our companies to get data warehouses and dashboards and operational KPIs set up and then to use them, like I say, to drive decision making every day in the business.

Speaker A: Okay, thank you. And if we move to, to Searchlight. So that's the potential application, isn't it? But if you're walking in with no predecessor, uh, no team, no established remit, where do you even start?

Speaker B: So I think the natural starting point for a lot of this activity is the value creation planning process. So lots of funds now will go through an exercise with companies either during the deal process or very, very early in the whole period to map out the biggest opportunities for growth during the whole period. And oftentimes I think what we see at Searchlight is that a majority of those value creation initiatives are underpinned or enhanced by having access to data. So you think about initiatives like increasing customer acquisition, think about uh, pricing optimization, you think about elements of operational optimization around whether it's routing or rostering or staffing or customer retention initiatives. Doing those well is much, much more feasible if you can actually see the metrics and see the data and see what's really happening in the company every day. So I think one natural path into this is looking value creation opportunities for a business and working backwards from uh, okay, if over the next, let's uh, say three years we want to be able to achieve these commercial outcomes, what data sets do we need management to be looking at and have access to, to deliver those really, really well? And I think a lot flows from there. So I think a lot of the time I'm working backwards from those outcomes, backwards into what teams do we need to build in our companies, what technologies do we need to deploy? Working groups, we need to pull together those sorts of things.

Speaker A: In this series, I spoke to a number of AI operators and a different split in terms of their remit by the vax. Looking internally at the firm and how they're using your EOI and obviously portfolio. And I get a sense that we're talking that a majority of your time is on the value creation end, is that correct?

Speaker B: That's right, yeah. I probably spend 80% of my time working with our portfolio and maybe 20% on internal Searchlight initiatives. I mean there's so much that we could be doing across both the portfolio and internally in the fund. But certainly my bias is towards our portfolio where we have a portfolio of about 25 private equity investments right now and I'm the only person in the fund that's leading on AI and data right now. So there's plenty, plenty of work to go around just with the portfolio.

Speaker A: Well, uh, that was a question I was going to ask, are you building a team? Are you leaning on external partners or something in between?

Speaker B: Yeah. Part of our philosophy from very early on when we were setting up the value creation team at Searchlight was that we would intentionally keep the team quite lean and lean a lot on third party resources, really try and connect the best resources into our companies through a combination of both consultants and contractors and permanent hires into our companies. And I think there's a number of benefits to that in terms of just how we execute with our companies.

Speaker A: We might come back, back round to talking about capability in a minute, but just be good, good to just dig into it a little bit more where AI has had an impact and obviously I'm not expecting you to provide anything commercially sensitive, but an example of where AI has really seen something change for the good, where we can see real world impact and that can be obviously a searchlight or wannabe the portfolio be great to get example there.

Speaker B: I mean, look, we're still very early on in the AI journey overall in Searchlight in the portfolio, but also just in the economy generally, I think. So a lot of the technology is still maturing, but one of the early use cases that I think we've seen a lot of success with, it's a simple one, it's hooking up AI assistants like Claude or ChatGPT who structured data, so data warehouses in our companies. So if I think about it, over the last four or five years we've had, I think we've made a lot of progress building out these data warehouses and KPIs in a lot of our companies and that's been great. But there has always been a challenge with these BI initiatives that is like the friction involved in opening up a dashboard or, or wanting to ask a new question that isn't quite answered by an existing piece of analysis. And you know, traditionally if you're an executive or you know, a manager in a business, you typically would have to go and ask your new question to an analyst, a specialist, who then go off and they'll run the analysis, they'll build the dashboard, it'll probably have to be prioritized in a queue against other people's requests. And so it's not uncommon to have to wait some number of days or weeks or maybe months to get that piece of analysis done. Or, you know, some companies have hired hundreds of analysts, um, built very large teams, but that's typically not the path that our sort of profile of companies would go down. And so what we found is with AI, of course, AI is very good fundamentally at writing SQL, that's a big unlock. So the models are trained on lots of programming languages, of which SQL would be one. And so at the most basic level, we can have AI now run queries against our data warehouses to answer questions for people all across the business who would previously have had to have either waited and relied on somebody else to answer that question for them, maybe just not asked it at all because it was going to be too hard. And so what we find now is we've got folks up and down businesses who are non technical actually running their own analysis against these data warehouses. And so when you go from having to wait days or weeks to have a question answered to literally a minute or two and be able to do it yourself at any time of day, whenever the idea comes to you, we're seeing that as being a huge unlock. So I've got examples of COOs and CFOs and CIOs who are all now running their own analysis rather than having to go and outsource that to somebody else in the team. But we're still very early days on that in the sense that a lot of folks are still working out how we make that really reliable and robust and governed. There's lots of challenges with that, but the early signs from that are very, very positive, I think.

Speaker A: So democratizing the access to insights rather than data is the outcome there, isn't it? Well, just thinking how you prioritise both in terms of the number of portfolio companies you've got, but then also workflows within those portfolio companies. Is there a priority there? Is it the imagine it's potentially commercial, but is there an area that you focus more on first?

Speaker B: The answer is probably no, in a sense. As I think about AI opportunities across companies generally, I kind of put them into two buckets. At the moment there's a sort of personal productivity opportunity by which I mean giving people Claude or ChatGPT or Gemini and having them enabling everyone to utilize the power of AI for general kind of personal productivity tasks. And by that I mean it's this mode of I'm going to ask a question or I'm going to specify a piece of work and the AI is going to help me answer that question or deliver that piece of work. The use cases for that are very broad and very general. But having those platforms rolled out in all of our companies is I think very important. And we're seeing a lot of traction with that at the moment. That's going to become table stakes of course very quickly. The second category I think of as being agents that are doing work. So this is where you take a, uh, task that would previously have been led by a human and you now package it up in a way that you can actually delegate it to an agent to do and have like primary responsibility for. Humans may well still be reviewing and checking the work periodically, but the core like uh, day to day delivery is being led by an agent. Those sorts of use cases for those, I wouldn't come in necessarily with a prescribed view of where the biggest opportunities are because I think it really depends on the nature of the business and the sort of state and maturity of the business and the data and the teams and processes within that company. But what I would say is we've for sure anything that drives EBITDA or cash flow is very important for us. And so we've seen, we're seeing really important use cases around things like quoting and pricing, for example, in businesses that have kind of high transaction volumes or perhaps bid writing for companies that have like large tenders that they're bidding on. And then on the more operational side of things, things like scheduling and routing for think of like field services, businesses that have thousands of people out in the field every day. There's a lot of opportunity to optimize cost and drive more revenue from optimizing how people are utilized or how assets are utilized day to day. Uh, and so those are perhaps some of the more tangible agent use cases that I would think of as complementary. And in addition to the kind of personal productivity ones like being able to query data warehouses or execute customer research through AI chat interfaces, there's a growing

Speaker A: frustration around the amount of investments going in, whether it's tokens, whether that's their uh, expertise and not enough in terms of clear roi. Is that the same with the C suite in your organizations, with the management within Searchlight, Is that frustration? Are you seeing that as well? And how well can we start to offer articulate roi?

Speaker B: I think that's right, yeah. I think that's one of the biggest blockers right now for not doing more. And I think, I think there's a lot of enthusiasm now for doing a lot with AI across all of our companies. And I think that manifests itself as like, these personal productivity use cases are relatively easy to get started with. There's not a huge amount of upfront investment to get them working so they, we've got traction on those fast. I think some of these agent build use cases do require a bit more focused effort, but perhaps not so in terms of technology capex necessarily, but in terms of team time and attention, I think it's a very legitimate uh, question for a CEO to raise right now is like, yeah, sure, show me the ROI on doing these things if I'm going to get my team to spend time on it and not do other initiatives that we could be doing. So I think, like I say, we're very early in the overall AI journey generally in society. I think we'll look back in five, 10 years and think, wow, okay, we're still 20, 26. We're still right at the beginning of what was possible with the technology. And so I think we are seeing early signs of ROI on some of these early projects. But there aren't enough demonstrations of that just yet. But I'm very positive on it. I think that if you were doing an AI agent two years ago or even a year ago, you were building something that was intrinsically quite different to what we can build today. Today we can build agents that are taking action across multiple systems and tools and interacting with lots of people in a company over multiple channels. So we can really build, build things which are tangibly valuable to companies today. That was a real struggle a year ago and probably wasn't possible at all certainly two years ago. And so I think the ROI is definitely there. There just aren't enough demonstrable use cases of that just yet.

Speaker A: And do you find part your role then is to build that confidence to continue the journey, continue the investment?

Speaker B: Yeah, I think so. I think for me, one of the things that I'm focused on right now is developing a library of examples of agents. I think lots of us sit here talking about agents in the abstract, but very few management teams have actually seen an agent working. Everyone's seen and interacted with email and search engines and Word documents and whatever. But how many people have at this point in 2026, actually interacted with and relied upon an agent as doing work? I think not many still. So bringing that to life for people with working demos in a believable sort of environment I think is very valuable. I think of it a little bit like every product team that I speak to who's pitching a product to me turns up with a product demo. Every consulting partner that I speak to turns up with a slide deck. I think bridging those worlds, bringing like a working demo of something which may ultimately be delivered through something that looks more like a consulting project or a, or an internal project is very valuable.

Speaker A: Yeah, it's certainly something that, as I've grown this network has been really valuable, is bridging the sort of knowledge gap and actually rolling up sleeves and going, right, we're going to prompt today, we're going to do something here, we're going to vibe code, you know, something that gets people out their comfort zone. And it's not the expectation that this is something they do day to day is just for them to sort of take that leap, I suppose, to what's possible. And it brings me on to the next point. So we've been running the AI benchmark for a year now, so every quarter we just get a check in to see how progress we're making. And last time out in June, we were asking around mobilization and where the main gaps are actually where they felt strongest and where they felt there was a weakness and strongest. The top one was prioritization. So there's feeling that we're getting better at pointing the resource, the investment. But the biggest gap was people and change. 30% said, said people and change was. Till 50% said, uh, it was people and change for the portfolio. So there's a real sort of, I suppose, concern, but more work to be done in the portfolio around people and change. Is there anything else that you're doing right now that's tackling that?

Speaker B: I mean, I totally agree. The people and change challenge is going to be massive. I think, uh, maybe there's a specific subset of that which is like process redesign, which is quite important to me. So I see examples where we start thinking about how we could build agents that are taking on work. We quite quickly find that that probably spans today. Different teams with handoffs between them. There's an example of this in one of our companies we're just looking at last week. So what might have taken five different people to execute before, and now you want to bring that together into a single agent and the agent can look across the whole thing. And so there's an element of kind of process mining and thinking through how you'd redesign a process to make it suitable for or to maximize the value from having an agent do it. But I'm also like, I've got a slight degree of skepticism also myself around pointing at the people change challenge as the main blocker at the moment, because like I say, the technology is still so nascent and we've still got so few examples in the world of agents doing work in real companies. So I also don't underestimate just the technology challenge at the moment of building and deploying things into companies in a seamless and cost effective way. But for sure the two things need to go together. I'd say the other thing that's obviously critical as we go through this is these are not IT projects. That's one of the things that's become very clear to us. Often it is, is like it's a CIO or a technology leader who is perhaps on point to drive the AI agenda in a company, which is fine, but the actual execution of a project has to be like so closely paired or led by a functional leader like uh, a CRO or a COO or a cfo. If you don't get that right, then all of the conversations end up being about how, you know, what new features should we build, how should we change the platform rather than about how should we change the way the team is operating. How do we really almost kind of mandate or really strongly encourage people to change their ways of working versus what they're used to? Because maybe that's where the change management M part really shows up. You've got to have that strong leadership from a functional leader in the business with the right authority to change the way people are working.

Speaker A: You, uh, mentioned when we spoke previously around also a lack of experience and capability as well. Could you just touch on that too, please?

Speaker B: There's maybe a question around. Do companies have all of the right skills in house to execute on these opportunities right now? And the answer is almost certainly in most cases not right. Because so much of the technology is so new. And I think about our portfolio of companies is I think of them as real world companies. They're not technology platforms, they're not SaaS platforms. They are building bits of aircraft fuselage or laying fiber optic broadband cables, or educating nurses or doing in home healthcare. And so in those companies, technology has traditionally been a relatively lean function, typically. And so you have it teams, you have various aspects around that m, but typically not large teams of developers. And I think what I'm seeing is we can, we can talk about obviously the relative shift now from um, productized SaaS solutions to in house more bespoke solutions. I'd say, um, I've got a balance to you on that, but I'm seeing a lot of opportunity around building bespoke solutions in companies which just fit the workflows and the data sets of that company so much better. But if you are building, then you do need some level of software engineering expertise which a lot of our companies haven't historically had. So that's one area where I think we're going to need to build out a lot more what I would probably call AI engineering teams within our uh, companies. And I think of it as equivalent to what we've done with data teams over the last 10 years. If you wind back 10, 15 years, very few companies had uh, dedicated data analytics teams. And now pretty much every one of our companies has a dedicated data analytics team and they're driving huge progress and value in those companies. I would not be at all surprised if we see the same pattern for AI engineering and maybe even the role of data analytics teams merges into AI engineering in the future for all we know. But having dedicated specialist resources to build and maintain some of the core infrastructure and better practices and govern some of what we're building is I think going to be very important.

Speaker A: If we go back to something we spoke about earlier, which is um, leaning on external expertise and I suppose, uh, PE firm has the opportunity to deploy that where needed across the portfolio and this is something that we're hearing more and more about is forward deployed engineering, you know, solving the last mile problem of AI. Are we talking that sort of area?

Speaker B: Yeah, I think. And by the way, the Ford deployed engineer term I think is kind of funny because it implies we're going into a war zone. We're not deploying AI engineers into like a battlefield here. Like we're in a friendly environment working in our companies. So I think the FTE term, um, I think of as mostly associated with product vendors who are trying to then deploy their software into a, uh, business, which is fine. That's great. That's definitely helpful if you're buying a product. I don't know that we are going to be buying products for lots of these things. I don't know.

Speaker A: Right.

Speaker B: But I think about all the different tasks and workflows that our companies are doing today and there's probably hundreds of different tasks and workflows that we'd want to automate. And a lot of them are quite specific to that industry and that company, their customers and their products and services. So I suspect that a lot of uh, that won't be covered well by more horizontal SaaS solutions or even vertical SaaS. So the FDE role, if we think of that as being tied to like a SaaS product, that is great, but probably still only addressing a minority of the AI use cases that we're going to have in our companies, I suspect. But the idea of having folks who have like a technology spike, who have depth, they can write code, they can architect systems, obviously the writing of code is becoming less and less critical, but the, the architecting and design of these things is more important than ever. People that have that spike and are closely coupled into the business will be very important. And this is the same thing we've done with data teams as well. I've always been a strong advocate for having data as close and integrated to the business functions as possible as opposed to carving it into like a, uh, very technology focused sort of separate role. I think it'd be the same with AI, whether we call them FTEs or not. We'll want to have these folks with technical expertise but really plugged in to the business users where they're going to be most effective.

Speaker A: Same theme. If we're just looking ahead to the next 12 months and taking your particular role within private equity, what do you think that, how the focus will change? Where will the attention be, do you think?

Speaker B: I think we're going to see a lot of attention on unlocking these true agentic loops, as I call it. So maybe there's stages of maturity that we're all going to go through as we figure out how this technology is best deployed. But Almost like phase one is, we get ChatGPT or Claude, we hook it up to a few of our internal systems so, so it can be grounded in our company's context. There's a lot to do there that we're still overall pretty early on that, but think of that as maybe phase one of like effectively ask a question or delegate a piece of work and have a response come back. The second phase is obviously building these agents that we've talked about, but within phase two, maybe I'd call this like phase three is turning those into true reinforcing loops. So there's all kinds of subtleties in this, but I think one of the lessons of the last few years has been just how powerful kind of raw compute can be when applied to the sorts of problems we've seen that from the scaling of the LLMs themselves, I suspect we're going to start seeing it in the application to business problems as well. By which I mean if we think about say a scheduling agent, we can build a scheduling agent today and encode it with our knowledge of how the world works today and what we want the agent to do and the rules we think it should follow. And then as you set that agent loose and it starts doing work, it can start identifying patterns right around oh, okay, this job was, you know, it's meant to take 2 hours, but actually on average it takes 2.6 hours. Or this customer over here tends to actually be late, like not available at the start of the appointment or something. And therefore I need to budget some flexibility around that particular type of customer. When we can kind of close the loop and have the AI generate learnings from doing its work and that feeds them back into its next run, then I think we're going to unlock another level of improvements of these agents. So I think we'll start to see that becoming a priority in the next 12 months or so. Once companies have got their first agents running and really delivering work, they.

Speaker A: And for your role that you have, and I'm calling it a operator within private equity, how will that evolve over time? Do you see that building out to be a team or is it that capability will be embedded so there won't be a requirement. Where do you see the role going?

Speaker B: It's a good question. I don't know. I think one of the challenges that every fund has is sort of scale, right? I mean even the largest private equity portfolios are uh, 100 or 200 PE companies, these PE investments at uh, most. And so I think there is this opportunity for consultants and others to bring expertise from across, you know, from working with a much larger set of companies from the market. And then my point being that a lot of these use cases will have, I think a common thread to them. So scheduling agents and quoting agents or procurement agents or finance agents I suspect will need to be, will be most valuable generally when they're built specifically for the company. But a lot of the patterns of what those agents will be doing will be quite generalizable. And so for funds and third parties, having over time a bigger and bigger library of those templates or those experiences having deployed those agents in the past will be, I think increasingly important. And so there is a question of, you know, if you are a mid sized fund like Searchlight, how do you best develop that scale, I think is a question mark. And it's not necessarily by building a bigger team internally. There may be other ways of achieving that, uh, but I see that as becoming more and more important. So when we invest in a new company two or three years from now, I would imagine as a fund we should be able to say, ah, okay, for this sort of business, there's maybe five or 10 archetypes of agents that we've built before in the portfolio that would be very relevant for this company and we should be able to leverage our experience of building those before to just accelerate and improve the deployment into that new investment going forward. So I can imagine a large part of value creation in the future being about deploying your best practice IP in the form of agents do work inside of companies, which is fundamentally quite different from value creation today. Right, which is predominantly about planning and guidance and oversight and rhythm and structure of a program. We should actually have the opportunity, opportunity to deploy tools that do work into companies in the future for sure.

Speaker A: I think there's, there's obviously a first mover advantage here, isn't there? But then what becomes table stakes versus the uh, differentiation. Now we're coming very, very near to time. I just want to ask you one last question because there are firms out there that haven't, don't have someone like you in place. And so just a bit of advice for those firms really in terms of what should they do, what shouldn't they do in setting up someone to succeed?

Speaker B: Maybe I'm biased, but I think the data, uh, foundations here are more important than ever. So in all of this excitement around AI agents and everything we can do with AI, I think let's not lose sight of the fact that just having visibility of your operational metrics and easy access to your company's data is valuable not just for management teams, but also going to be very valuable for the agents we want to build as well, because a lot of the most valuable agents will need to draw on your company's context both in terms of structured data and also unstructured knowledge as well, which is a whole other topic which we're thinking about at the moment. So I would continue to prioritize data and analytics as a foundational capability as part of the overall kind of AI transformation program. Uh, and I think also having at least a person in the fund that is an expert in data and AI is very helpful for just translating between the ideas and excitement that CEOs and investment teams can generate and the delivery, how you actually deliver on that with technology leaders and data and AI teams in companies. So I'd advocate for having a person, I'd advocate for having that person having a background in some aspect of data analytics and technology. And then beyond that, I think being connected with as many other funds and third party advisors and individual experts as possible is just invaluable because everyone is learning in parallel here at the same time.

Speaker A: Yeah. Making sure they're not falling behind and in the pack. Uh, Miles, look, we've come to time. That's been a really, really interesting valu conversation for our listeners. Thanks so much for speaking with me today.

Speaker B: No, thanks for having me.

Speaker A: It's been great for our listeners who'd like to follow Miles work. We're going to put your LinkedIn in the uh, show notes as well as the uh, details for Searchlight Capital. That's it for episode five of our Operator series on AI Pathfinder for Private Equity. If you're in a similar role, would like to be part of a future conversation or join our growing AI operator network across London and New York, please do get in touch. Details in the show notes. Thanks for listening and we'll see you time next. Next time.