Full episode transcript: AI & PE: The Future of Value Creation

Guest: Miles Rowland, Data & AI Initiatives, Searchlight Capital Value Creation Team | Host: Kyle Roemer

Miles Rowland: I think this natural language interface to your data is going to be huge. There are so many questions people want to ask — any time you’re heading into a board meeting and you get a question you didn’t know the answer to, just being able to fire it off and get an answer from your data will be, I think, transformative to how people use data in businesses.

Kyle Roemer: Welcome to AI & PE: The Future of Value Creation — the show where we cut through the noise to hear directly from experts, practitioners, and members of the community on how AI is transforming portfolio companies, accelerating growth, and rewriting the playbook of private equity. I’m your host, Kyle Roemer, Head of US Data and AI here at Accordion. Today we’re joined by Miles Rowland, who leads data and AI initiatives at Searchlight Capital’s value creation team.

In this episode, Miles breaks down his clear, practical framework for AI maturity — from basic chat tools, to agents that take action, and ultimately to ambient AI running quietly in the background. From there, Miles shares how AI is reshaping the data stack, and why natural language interfaces to data may be the biggest unlock for mid-market businesses. Now, my conversation with Miles Rowland.

Miles, welcome to the show.

Miles Rowland: Thank you very much for having me.

Kyle Roemer: You come to AI from a data and analytics background — what drew you to the AI space?

Miles Rowland: For me, AI is just the latest step on the journey of helping businesses use their data and technology to make better decisions. My career started at EY, coming out of university — I started as a generalist management consultant and ended up establishing the data analytics team there. We built tools that went beyond Excel, to analyze the larger datasets our clients had.

That’s been the common theme through a lot of my work over the last few years — so many businesses have tons of data they struggle to use to support decision-making day to day. There’s a lot of potential, particularly in the mid-market, to help management teams make better decisions using data. AI is the next step — not just how do we help management teams make better decisions, but how do we make better decisions across companies all the time, whether that’s computers making some of those decisions or humans. There are just a lot of decisions to be made constantly in most companies, and we can do that really effectively if we use our data properly.

Kyle Roemer: I love that, and agree there are plenty of decisions to make across these businesses. Before we get into the topic, I’d love to hear more about Searchlight and your role there.

Miles Rowland: Searchlight is a mid-market private equity fund — about $17 billion in assets under management across North America and Europe, investing across telecoms and media, industrial services, financial services, healthcare, and education. The common theme is that a lot of the companies we invest in have really solid business fundamentals, but a lot of potential to use their data and technology in a more advanced way to drive value.

I work on the value creation team at Searchlight, a small group focused on working with our companies post-investment, and I specifically focus on using data and AI to drive that value creation.

Kyle Roemer: There’s a lot out there right now around AI, interfaces, automations — everything from ChatGPT-like interfaces to agents taking action to big engineering projects. Can you unpack the paradigms you’re seeing right now, either in the portfolio or in the market more broadly?

Miles Rowland: Yeah, absolutely. I think there’s a progression through three or four layers of sophistication that we’re all going through right now with AI — it’s helpful to have a vision of the end state, to know where to go next. Who knows, this might change in the next six months, but here’s my sense today.

Level one is: you get ChatGPT, or Claude, or Gemini, and you’re asking it a question — purely out-of-the-box, consumer-grade stuff. That’s often the starting point for a lot of us, including internally at Searchlight, and just getting good adoption at that basic level is really important.

The next layer is connecting that AI to tools and data within your business — your email, your calendar, and critically, your data warehouse. Some of that’s become possible over the last year or so; there are connectors to Outlook and SharePoint now for most AI platforms. But connectivity to your data warehouse is still something we’re figuring out.

Level three, I think, is that plus the ability for it to take action — you ask your AI to do something on your behalf through your general-purpose AI assistant. Then the question that feels important to me right now is: what’s level four?

Over the last couple of years, when someone wanted to build an AI agent, they usually ended up doing a lot of engineering work. I’ve always been biased toward avoiding heavy engineering projects, because our companies typically don’t have large development teams — often zero software development expertise, and there are always questions about maintainability. What I think we’ll see happen in 2026 is that the tooling has become sophisticated enough that you can just give an agent tools and instructions, and it can choose how to execute against a given task. With a few hundred lines of code, you could create an agent using, say, the Claude Agent SDK, connect it to the relevant MCP servers to read and write data, and write instructions in markdown files — just human-readable documents. That could be enough to create your agent — that’s the level-four thing.

So to me, level four means you don’t have to prompt the AI to do something — it’s running in the background, on a scheduled or triggered basis, and taking action when it sees something in your email, a customer support ticket coming in, or an inventory stock-out. For it to do that, you have to get out of the AI assistant platform entirely — out of ChatGPT, out of Claude.ai. What I’m really excited about is that building those agents is becoming much, much simpler.

Kyle Roemer: I’d agree — I love this term that’s been floating around, “ambient AI,” which is kind of where we’re going: it’s operating in the background, but that requires a variety of things happening in concert. We see something similar — the amount of code you need to write is becoming less and less. There are more tools, APIs, things you can call for the more complex workflows, and while it still requires a decent amount of engineering effort, you see that getting abstracted over time as agents take on more of what you’d otherwise have to code for, because they have tools accessible to them.

Miles Rowland: Personally, I’m not the best at, say, my Python syntax — I haven’t spent a ton of hours writing dbt code myself. But I can read what it’s doing and understand the architecture of how it’s building things. I think of it as combining the brainpower of you as a data expert with the brainpower of the agent — those two things together are really effective. Just having one on its own isn’t nearly as good.

Kyle Roemer: You’re in a really interesting role as an operating partner focused on data and AI. Having seen that role at other firms, there are a lot of different operating models — and one downside is you can get pulled into a lot of conversations about AI that don’t always lead to strategy or value. How do you integrate with deal teams, and how do you prioritize where you and your team focus?

Miles Rowland: The key for us — through a bit of trial and error — is that the easiest way to integrate with deal teams and management teams is to be heavily involved at the back end of diligence and the deal process, and into the value creation planning process, through the first year of the investment.

Over the last five years at Searchlight, we’ve set up a robust, consistent value creation planning process that deal teams drive, and that the value creation team engages heavily with. That process ultimately identifies a handful of key value creation levers we want to pull in the first couple of years of an investment. The best way for us to engage is to pick one, two, or three of those levers and figure out how to use data, AI, and technology to identify the actual initiative. It’s one thing to say “let’s improve customer retention” — it’s another to identify the actual drivers of churn, fix those, and then monitor the results.

That’s where a diagnostic project comes in, and increasingly, what I’m looking to do in 2026 is slot AI automations in there too. Take the customer retention example — there’s a diagnostic piece upfront to figure out why customers are churning, then an initiative design piece, some of which will be pure business initiatives. But some might be a churn prediction model that identifies which customers to intervene with, plus an AI automation that actually does the outreach or sends the promotion — so maybe 80% of those customers get automatically addressed in some way. There’s a natural pathway there from BI to data science to AI automation that feels pretty intuitive.

Kyle Roemer: Agreed — the churn example is applicable to most of the businesses you invest in. But it’s one thing to build the predictive model, and another to augment your customer success team with a recommendation, or to actually take the action for them, in a way that’s genuinely relevant to the customer — because you’re only as good as the context and history you have with that customer. If you can feed all that in and augment those individuals, you should get a higher success rate serving customers.

Miles Rowland: This is a really interesting thing I think we’ll start to observe as these AI projects mature — the change in agility of a business once you put these automations in. Historically, automations meant relatively heavy engineering or IT projects that themselves took time to execute. Going forward, I expect AI automation to be something you can set up in a matter of weeks or days for a given topic, like the customer churn example. Once it’s set up, updating and tweaking it might take hours or days — and it’ll be able to update its own instruction set fairly soon too; that’s already possible today for AI agents.

So something that might have taken you three months previously — say, changing your pricing structure and training the sales team on how to sell to customers — you could probably do in a day or two with a more automated system. That’s not to say we won’t need humans — of course we will. But if you can keep the size of the human team roughly the same, the agility will be much, much higher.

Kyle Roemer: I love that word, “agility” — I think that’s a major theme starting to permeate, and we’ll see a lot more of it in the coming years.

Miles Rowland: That’s where I think private equity funds can bring a lot of value — in the pattern recognition and experience gained from having a portfolio of companies where we can see what’s working and what’s not. We’ve got an active role to play in hand-picking the right skill sets to recommend and inject into our companies. But getting the right people around the table is still the number one thing a private equity fund can do to accelerate value creation in its companies.

Kyle Roemer: Despite the headlines, people still matter quite a lot — it’s not done.

Miles Rowland: The returns to expertise are only going to increase, I think, as a result of these automations — there’s less distance between someone having an idea and actually making it happen in a business. Fewer steps, less time, going forward with AI. So expertise is, if anything, more important than it was before.

Kyle Roemer: Translate that a bit — if you’re a CEO, CFO, or investment professional listening, why should they care?

Miles Rowland: Traditionally, it’s been hard to hire good data analytics people — too much demand relative to supply. We’ve been through a set period, 5 to 10 years, of increasing “engineering-ification” of data analytics, which meant we drew in more technical expertise, but if anything, diluted analytical expertise. That’s led to projects that have been, at times, frustrating for CEOs and CFOs — asking for something that seems simple to a business user, and it generates a lot of man-hours, a lot of technical work, and longer timelines than expected.

AI is helping invert that. With a few highly skilled data analytics specialists, you can get a lot done — smaller team sizes, probably lower overall team cost, and dramatically faster time to insight. Where you might have needed a team of ten people doing data analytics in a mature mid-market organization previously, you could probably get the same result now with two or three — but they have to be genuinely skilled people. You need deep expertise. If anything, it means we can focus a bit more on the analytical skill set relative to the technical skill set than we might have in the past.

Kyle Roemer: It’s a really interesting insight — data has become more engineering-heavy over the last several years, and somewhere along the way we lost some of the core analytical depth. That’s getting easier now. If you can get your deeper analytical, data-science-minded folks doing some of the engineering work too, because it’s AI-enabled now, that’s really powerful.

Miles Rowland: There are people in businesses with a genuinely analytical mindset who haven’t had that technical depth — they’ve lived in Excel, maybe dabbled in Power BI. For those people, it’s becoming much, much easier to get up to speed. The learning curve on the technical side — SQL, dbt, whatever the technology is — is much faster now. In the past, I’ve spent hours in Stack Overflow trying to figure out how to write some piece of code in a given language; that now becomes a 30-second job when you ask an LLM. So I’m hopeful we bring more analytical, less purely technical, people into the data analytics space over the next few years, as a result of how the day-to-day job is changing for data analytics specialists.

Kyle Roemer: That’s a good hope, and hopefully we see it as a market trend, because there’s been real scarcity for certain skill sets. As you invest in new mid-market companies and think through your early plan — meeting your thesis, your set of hypotheses — a year or two ago it might have taken 6 to 12 months to get the foundation right. Today you can move a lot faster. As you go further downstream — transforming data, getting it into the right data models, making it available for analysis and reporting — what’s been an interesting use case you’ve seen in the portfolio so far?

Miles Rowland: Sometimes I’ll just ask Cursor to build a pipeline, go do something else for five minutes, and come back to find it’s iterated its way to the right answer — incredibly helpful. The other angle we haven’t touched on is the front end. Data pipelines are one piece of the puzzle, but building dashboards and charts is another key piece. Dashboards have historically been the delivery mechanism for analytics into businesses, and I think we’re seeing that get disrupted by tooling — there’s so much activity in this space right now, with pretty much every BI vendor trying to add an AI agent into their platform at the same time.

I think in 2026 we’re going to see a lot of progress there, where the number of clicks it takes to spin up a dashboard drops by a factor of 10 or 100. The speed to build not just your pipelines but your analysis comes down massively too. Together, those two things mean time-to-insight — which, from my career, is probably the single most critical predictor of success for an analytical project — is coming down fast.

Kyle Roemer: Yeah, I think this is as fascinating as the upstream data pipeline work, honestly. For a long time you had the early advent of Tableau and Power BI, and some more self-service-oriented tools — ThoughtSpot, for instance, which leaned a bit more into a semantic layer, ask-a-question model — but it never quite got there technologically. Now it is there. People want to talk to their data, which was always the selling point five or ten years ago. Now you see platforms like Hex — if you have a data platform and you can expose an agent to all the context within it, you can get answers really quickly without clicking around a bunch of dashboards.

Miles Rowland: This is the part that’s most exciting to me right now, something we’re actively working on — how do you expose your data through a natural language interface to all of your business users? If you think about the work data teams have done historically, there’s been a lot of dashboard building, and dashboards grow as you try to preempt more of the questions business users might ask. But there are inevitably more follow-up questions, and the dashboard just expands and becomes unwieldy — one of our companies ended up with, or had until recently, 1,500 Tableau dashboards. You end up in that kind of chaos because a dashboard can’t really answer every follow-up question someone wants to ask.

The promise of natural language interfaces is that you can streamline your dashboards — there’s still a really key role for dashboards to play as the source of truth for KPI metrics — but all those follow-up questions can be handled through natural language instead. One observation from the last few months: the LLMs themselves are very good at writing queries and interrogating data. Through 2025, we’ve seen LLMs get better at instruction-following and retaining context over longer conversations, which means it’s not clear to me whether you need a dedicated tool to serve as the natural language interface for data, or whether you can just plug your AI assistant of choice directly into your warehouse.

The caveat is that your warehouse needs to be well structured, and you need great documentation to provide context. I think a key challenge in 2026 is exactly that kind of context engineering — how do you inject context at the right moment? But tangibly, we’ve had a lot of success recently just plugging Claude into Snowflake, injecting the right context through an MCP server. It does a really good job answering difficult questions in a really efficient way.

Kyle Roemer: To your point — bringing your favorite agent to your data platform with the right context, structured data, and so on. Bringing it up a level — how do you think this changes the role of the analyst?

Miles Rowland: Significantly, I think. Data analytics specialists — data engineers, visualization experts, data architects — have historically spent the majority of their time building the machine that provides the answer. There’s a piece upfront around requirements gathering, and a piece at the end around user acceptance and embedding tools in businesses, but the bulk of the work has been the building phase. That’s changing — the building phase is becoming shorter and faster as AI helps build things, and we’re going to open up so many more questions users can ask through natural language interfaces, which won’t require as much building, but will require scaffolding and measurement. I think we’re going to see a lot more work required around evals — does a user get the right answer out, when they ask a given question through their AI interface?

Kyle Roemer: All right, Miles, we like to end these with a rapid-fire segment — quick reactions. When it comes to AI and your tech stack, what’s one piece of advice you’d give the C-suite at a PE-backed business?

Miles Rowland: Keep it simple. Don’t overcomplicate it. Start with the things that are relatively quick and easy to achieve, and go from there. There’s a lot of value in just starting on AI.

Kyle Roemer: Outside of work, personally, what’s your favorite use case?

Miles Rowland: I hate having to think about what to cook each week. I’ve got a little agent that throws recipe ideas at me based on my historical preferences — it saves me a lot of time when I’m doing the weekly shop.

Kyle Roemer: I love that. When you’re cooking dinner regularly, you run out of ideas fast — having an agent for that would be genuinely helpful.

Miles Rowland: That’s right. There’s only so many times you can eat the same five meals.

Kyle Roemer: What’s your prediction for where AI creates the most value in PE-backed businesses over the next 12 months?

Miles Rowland: I think this natural language interface to your data is going to be huge — because there are so many questions people want to ask. Any time you’re heading into a board meeting and you’ve got a question you don’t know the answer to, or you’re on the train on your way into work, being able to fire off a question and get the right answer from your data 95% of the time will be, I think, transformative to how people use data in businesses.

Kyle Roemer: That’s fantastic. You heard it here first — so when it comes true in 2026, you’ll know who to reach out to. Miles, thank you so much — this was awesome. There’s nothing like talking data with someone who’s operating at a high level in private equity. It’s a genuinely exciting time to be in this industry. Miles, thank you, and I’m looking forward to continuing the conversation.

Miles Rowland: Nice — I’m super excited too. Until then, thanks for having me.