076: AI rollups: all VC sizzle and no stake? Thoughts from a recent Fortune article.

Fortune article: http://fortune.com/2025/06/27/ai-rollup-investment-strategy/

Nikola reflects on his recent Fortune commentary (above) into the transformative potential of AI rollup strategies in Business Process Outsourcing (BPO). He and Damien discuss the concept of AI rollups (where private equity firms consolidate service companies to create more efficient, tech-enabled businesses), as well as  the unique challenges and opportunities associated with integrating AI into existing BPOs, the role of Global Systems Integrators (GSIs), and the evolving landscape of customer service automation. Hear why the world won't be rewritten by AI overnight but will transform progressively" in this week's Deep Learning!

• Follow PolyAI on LinkedIn
• Watch this and other episodes of the Deep Learning pod on YouTube

Transcript

Nikola 0:00

the world won't be rewritten with AI tomorrow. It will be rewritten bit by bit for many years to come. The GSIs have a huge role to play. I think that BPOs shifting towards doing more of that work themselves is BPOs doing a great thing for their shareholders and their, the future of their business. I think some have done really good jobs. Others are trying to,

Damien 0:29

hi everyone, and welcome to another episode of Deep Learning with Poly ai, the only podcast. That's it. We're just the only podcast. News is out. Nicola. All other podcasts have faded from existence, so congratulations. We made it.

Nikola 0:42

We made it All right. Cool.

Damien 0:45

You were in the news recently with a early investor and pal yours, Nathan Bei. Is that how you say his last name or did I butcher it?

Nikola 0:54

Yeah. I only discovered I think his name's Bena. I've been butchering at St. Bena my, for the first few years of knowing him,

Damien 1:00

i'm sure he is heard worse. Anyway, you all had a nice little commentary in the tech section for Fortune recently talking about AI Rollups, and I want to pick your brain about some of the concepts that we're leading out in there. First and foremost, what is the concept of an AI rollup strategy?

Nikola 1:15

Yeah a rollup in general is basically a private equity firm that buys companies that typically do services, right? And kinda buys them one after the other, puts them together in a bigger company, realizing some economies of scale, making it a better business, right? So that's not an AI roll, but's just a rollup in general, right? Maybe you're buying, say, car washes or you're buying any kind of services, businesses, laundry services or whatever kind of like business you can think of. And you're creating a bigger business, right? You're buying them because, and they're all run a bit maybe inefficiently and at a greater scale you get to negotiate better rates with your suppliers and build a better, more efficient company then you. Increase profitability. Maybe you make it a more efficient business. If you can turn it into a bit more of a tech company or like a tech enabled services company, maybe you get to trade at a higher multiple. Contact centers typically trade at five to seven times their ebitda, right? And they're net profit margin is like five to 10, maybe 15% if they're really good. Now, that means you don't think a hundred million in revenue, 10 million in profit. You can buy the business for 50 million right now. If you could automate a lot of it with ai, right? And you make that a hundred million in revenue, let's say you keep it fixed, right? But you increase the profit margin to, or really, you do more with the existing people, which is the concept that we were interested in. And you automate, say, half of the work. So the same number of people now does the work of twice as many, and maybe you charge, 170 million for it, and maybe you're no longer a 10% profit margin. Maybe you're at 25, 30% right now. Hopefully at that point you trade a bit more like a tech company, right? So you're not trading at five times, ebitda, maybe you're trading at, 10 times revenue, right? At that point, that company's worth 1.7 billion and you only bought it for. 50 million, right? So if you could marry the whole AI model, inject the ai, fairy dust into a services business, if you can make the whole thing really work and you have a bit more of a tech company or even something in between you'd make killer returns, right? And then if you can prove the model, you do what private equity always does, you lever it up, then you raise, say, another 200 million. You borrow 600 for 800 million. You could buy a really big BPO, you might even be able to buy most of them, right? Or most of the large ones individually. And at that point you're dealing with billions of revenue and you can do it again. You've just found a shortcut to creating like the biggest AI company in the world. That is at least the theory that I fell in love with in 2019. I distinctly remember pacing around one of the meeting rooms we had in White Bear Yard in Clark Andwell back in 2019. And speaking with Sri Ch Shaker who is our longest standing board member, I think maybe we could even look to do a podcast episode with SRE and Nathan. Sure. Let's, okay. S

Damien 3:59

she on.

Nikola 4:00

And I remember SRE saying, in his very distinct tone why don't we just buy a contact center? And I'm like why not? That's group thing right there where we're like, okay, like how can we do it right? And I spent six months around, north of England and in Scotland, a lot of time in, in Glasgow, in Edinburgh, and fascinating businesses, right? I think that when you get like a glimpse and then you spend months looking at the processes and stuff, it really, you build an appreciation for what it is that we're trying to do here with ai, with the software technology, right? And you realize that it's not just about AI being this or a bit better or a bit better. Like I mentioned, I think in our last episode, it really is CapEx versus opex. Like how much work is that business, the enterprise that's paying that BPO for that labor for opex, right? How much are they willing to invest in their own transformation so that they can now do this with ai? And I think the equation has changed since then. More can be done with AI and the ways of transformation have maybe changed, but less so the equation still stands very much to this day. Our conclusion was that it had, it was threefold, right? The first one and the most important one is that you have to align the incentives, right? The reason that existing BPOs, the likes of tele performance, or well many others, almost any one of them we'll find it hard to do this, is that they're there to provide human labor, right? So if they automate that labor, they are eating one size of their revenue to grow another, and the other may trade at higher multiple, but it's gonna take the market a lot of time to appreciate the mix and reward them for that. And it's a hard thing to do for a business that's so operationally heavy, right? So what you would need in that case is just whole outcome based model of A BPO. So say you have your business, you get, I dunno, 10 million calls a year. And I say, Damien, I'm gonna do that for 5 million a year. And I promise you that your Cs a's gonna be at least, 4.3 out of five, you go okay, cool. If it's not what we figure out, some penalties, we shake hands and then I realized that you didn't tell me everything. Maybe there's double the calls and I didn't negotiate right? And I'm in a real pickle, right? And then I started losing money. I like Damien, man, look, I can go chapter 11 right now, or we can renegotiate and you're like. What's the equation between this person going bankrupt and then like you switching to someone else versus giving me a slightly better contract? Then you realize that it's really like the daily reality of a lot of this business, right? It's a lot for a startup to wrap their head around because the DNA of A BPO is so strong, so labor intensive and so difficult to the business. I really, I have so much appreciation for what they do and like the Scotsman that I met like in Glasgow working on this. They're phenomenal. I really enjoyed spending time with them. They're a lot of fun. Glasgow has pubs open till 3:00 AM and it's a phenomenal place. I had a lot of fun. It's really hard to rewire the, that business into what that company, it won't just happen overnight. So did you,

Damien 6:54

did you think that was the case back in 2019? Because obviously as you mentioned, part of the history of our company is that you said. No, to purchasing a contact center to purchasing a small BPO.

Nikola 7:05

So as I was with Nathan in New York like what, three weeks ago. And we're talking about this and I was like look at the board deck. He was talking about writing this article and I was like this is probably the most thoughtful piece of work I've ever done. So number one, BPOs have offered this kind of pricing and then failed. Like they run each other into the ground and then move back from it. It's not very common to see outcome based pricing. The second bit is where it gets harder, right? It integrations, right? People assume that if you buy, and, there was a, I think the CEO and the CPO one of the companies doing this right now, right? Backed by one of the leading venture capital firms. Commenting that, you then just buy it as an accelerant. To accelerate what,'cause their claim was that you need to change the, that it's hard to change the way the people work at A BPO, but if you buy one, then you have more control. That's great. But it's also incremental because the real issue was this, you go into automate and then you basically have to convince the enterprise that you're working for as A BPO that this AI system you're gonna put in front will do the same work that humans will. And to do that work, it need, that it needs access to their backend systems. Now that access either exists or it doesn't, but they need to work hand in hand with the IT team of that company to get that done right? Sure. And to date, as we work on digital transformation of companies using ai, we are really just often stuck against days and weeks and years of being able to. Figure out what we can do when and how, and then we go through those changes. And we have the luxury as a tech company of filtering to those companies that have opted to work with us. Many do. Most don't. It's a sales funnel and we're filtering for the most innovative people who have the most to gain. If you're a BPO, you buy a BPO, you're not, that filter does not apply. You just have the general public, right? You then have to sell inside that, which means that basically most of the revenue you buy will be that.'cause they won't want to move unless what you're doing is not full on automation and the vision that we have of 90% of customer service being automated. Instead, you may be talking about the different. The other category of AI products for the contact center, which is agent assist, make humans more efficient so that you can pay them even less and they can churn and you bring in new ones and they're okay from day one. That's the pitch of a lot of the agent assist platforms, and it's one that we don't really believe in because we think that you have to do the hard thing. To benefit the most as a business, it means that we work with businesses that have the most to gain in terms of revenue generation or in terms of the uplift of customer service, which has collapsed'cause they just can't afford to have as many people. If it's incremental, if it's about saving 10% of the customer, making the service just 5% better. That's not that interesting. And that is something that you can do at scale with technology because it means that you're not actually replacing the human. The human is there clicking through and it's like an autopilot for them, where their average handling time may not be six minutes. They might be able to be done in four and a half minutes, right?'cause you bring up the right answers, you tell them the right thing. Maybe you can get them to be better at converting certain leads and this or that. But the more knowledge, heavy and difficult a call type is. Or transaction type, the less likely it is to be outsourced as well. Because you hear of people outsourcing customer service, you don't hear of many people outsourcing sales, right? So those more valuable things where automation can be really good for some of those call types, that's not something that's a small lift on top. That's like really a transformation that to do it, you have to get'em to buy in and to get'em to buy in, you have to get'em to buy in, right? You don't get to buy that business by buying the right. To be the one provisioning the labor. And that's at the end where I looked at it and very, with a heavy heart. I was like, okay, I guess we're not gonna have, a case study named after me and ri our board member who decided to bag the players and we walked back right from the whole thing. And okay.

Damien 11:07

Yeah. So that, but that was, okay. So that was 2019. And like you said, there were some of these traps that, I think at the time. We're highly visible that the pricing trap the self cannibalization, the, that illusion of control the integration cost and expenditure that would come into staffing. Then do you think that some of those have pretty much weakened as generative AI technologies have gotten stronger? Do you think it's easier to go in and plug and play some sort of let's talk in the framework of AI type rollups, right? No. Where AI is essential? You don't think so?

Nikola 11:40

No, not at all. I think that there's one that's changed a little bit, but not fully changed. The pricing stuff hasn't changed at all. The first piece. Yeah, sure. The second piece, the integration piece, there are technologies on the horizon, things like opening eyes operator, where instead of building backend hooks into the systems and figuring out how to do that may not exist right now. And then maybe that CapEx that I'm talking about, you just have. A tool that if you let it loose on the agent screen, it can do all that. Like at the moment, it's neither reliable enough nor fast enough, but it will get there. Right now, it will get, but that increases further the risk associated with letting an autonomous AI system run amok and like just do that work on your behalf. Now being the, AI fanatic that I am, I think it'll work eventually. But by that point, like the IT threshold to kind. And what is that

Damien 12:34

point? When you say eventually, what do you mean?

Nikola 12:36

It's really hard to make these predictions right now. The barrier right now I'll be ever as optimistic I think in two years time at most. I. We'll have systems that are fast enough to like do the whole scan of the agent screen and be like, okay, I've seen how humans do it. I'm gonna click here. Then I'm gonna click al tab, and then I'm gonna go into this screen. I'm gonna paste this thing there the way that a human would tick, and I'll do it faster than a human. If the system will allow, gimme some waiting screen or something, or whatever. I'll do it as fast as a human could. That'll come right? But then like you'll again have a bit of an extra period of enterprises accepting that someone can just do that with AI on their behalf.'cause the first time that the thing accidentally, transfers money to the wrong account, which it will do. Sure. Then you'll get oh, we have level four, but we can't have level five. And then again, like the onus of doing all that. AI transformation work with a human still being there to supervise, it's not gonna be all that appealing. So we're already like growing super fast and achieving transformative outcomes when we just rebuild it from scratch.'cause that's the whole like age old thing. Will existing incumbents transform with technology or will new entrance come in and completely plan them because they don't have those dependencies?

Damien 13:44

Right.

Nikola 13:45

I bet On the latter thing, I think that there is the mirage of multiple expansion. Which is just too hard to resist for all these Wall Street Buccaneers and VCs alike, not necessarily the real VCs of like small few hundred million fund that invests looking for like standout returns, relatively early stage. But venture capital firms have changed, right? They're managing more and more money, and they're really converging with private equity at this point, right? So for a lot of those guys. Why not? And maybe they get really lucky and get like a standout, like multiple expansion story where the thing is a fund returner for a$2 billion fund. Oh, why not? And good luck to them. Look, I think that anyone working on this to create a world in which people believe in these outcomes, even more we can all do a lot more good work. Like good luck. I hope someone manages to do it'cause it's gonna be a great story and it'll gimme something to kick myself above for, the rest of my life to make it work.'cause it will be a, a shortcut to mushrooms. But it is it is unclear to me. I would still not bet on it.

Damien 14:48

And what would you say is the top reason that you wouldn't bet on it right now?

Nikola 14:52

Because they don't have free reign again, like you buy BPO. And you go, okay, like I now license to provide all the human agents to a telco or someone else. And then I go in and I say, Hey I'm gonna put AI in front. They're like, cool. You know what happens when you put in some technology as A BPO, a really hard core procurement person the most? Evil evolved form of the procurement person that exists is brought into the room and the lights are turned off and they say how much do you expect to make from this thing? Oh, I think the savings are a hundred million. Cool. Really nice. Nice. You can have 10. You're like, wait, what? I'm doing all the work. I'm taking all the risk. Yeah. And you can make 10 million. And you're like I don't get it. It's just culturally not acceptable for a BPO to make out like a bandit with Yeah, big outcome right? Then, you might, if you're, if you bring in an equally, sturdy negotiator, you'll walk out with 20, right? And then in the meantime, you're taking all the risk, you're doing all the work. God knows what it's gonna take to move them through, and they're not ready to commit. And at any point they're,'cause the real thing that I found almost offensive, if I'm honest with you, was just like, the attitude of you or BP are gonna sell me ai, come on. Move on. I need 10 more agents. In week three and another 10 in week four, and there was just like this really wide chasm between what the BPOs themselves wanted to do to reinvent themselves and what the clients were willing to look at them and kinda buy from. And I realized that in each of the conversations in the early ones where we went in and partnered, they would very quickly just contact us directly and be like, Hey, let's talk'cause they knew that whatever they pay us, they're gonna have our undivided attention.'cause a few million dollars for an AI vendor. In the early stages, there's a lot. But for A BPO, it's it's not all that much, right? Because they're charging tens of millions just at a very low margin. So it's difficult, right? And I think that it's just, you're, you, it was such an easy, good idea. You'd hear more about it. So I think that the BPOs were, you start,

Damien 16:50

you're starting to hear a little bit about it. I would say there's a mindset, especially among some of the larger BPOs out there. One with big, or ones with big contact center plays. That are investing in ta. Tata's, CEO announced last year that they were putting most of their r and t into AI services and converting to a service organization as opposed to a BA traditional BPO because they know that's where the future's going.

Nikola 17:16

So there's more than them. And Tata, has always been a big player in like general, it like services provision space, not just in like BPO for contact center. Sure. They're all talking a lot about it. Like a cynic might say that it's like sprinkling ai fairy dust on themselves, hoping that it does something to their unbelievably depressed multiples. An optimist. An optimist might say that they're genuinely pivoting and some of them really are like, I think that's a fair thing to say about Tata, about Concentrics, about a few others that have really moved on to like being much more of a consulting company where it's yeah, it Transformation services development, co-owning, like Tech Stacks, owning outcomes over tech stack transformations. Look, that's fair. That's look at the transformation of their business. But make no mistake, those are now like consulting companies, right? They're more like a Capgemini or an Accenture than they are. Their previous shape. And that's okay. But they're also doing

Damien 18:10

the, but they're also doing the implementations, right? So if that's the case, aren't they acting more like professional services in a software company if they're coming in? And it's not just process change and it's not

Nikola 18:19

just No, but that is what Accenture does. That is what Deloitte does. That is what Capgemini does. So I think, the more, you see a lot of these previously more like BPO outsourcing. Style companies pivot towards this space. You're just gonna see more crowded space. And listen, AI is hungry for for deployed things for professional services inside these companies, right?'cause you need to map out the world. You need to codify it and then allow AI to run over and do the work that humans did previously. This is gonna be a huge space. Like I think that someone told me recently, the company just made the most money out of AI so far. Beyond the shadow of doubt is Accenture, not even like a Google or Microsoft. But Accenture, because, they are over Nvidia. Okay. Fair. Okay. No, fair. But like I'm probably not more than I, I don't know. We should look at the numbers. But

Damien 19:10

it would it wouldn't be anything to take a look at those numbers because, but also it's funny to see that. Those consulting and service organizations are making a ton out of this evolution because, and this is going back to something that was quoted from the Fortune piece about service businesses. They quote, aren't inefficient by accident. They're inefficient by design. The inefficiency is the product in that. You're basically baking in that level of customization that can't be attributed by an internal team using a widely offered product or a widely offered service. Yep. And trying to drill down to their specifics. So what is with the recursion, it seems that you're getting a tech accelerated. I don't wanna say step back in process because it, you would think that. Technologies would better enable individual organizations to optimize for their own purposes. But instead, it's enabling service organizations to come in and still offer that initial value prompt, which is, a throat to choke if all goes wrong, or someone to ignore if all goes right. But it's better enabling them for transit, for transmission. Yeah. If it goes right.

Nikola 20:19

Listen, I think there's also just my co-founder Sean, will tell you that it's the. Absence of a widely scaled enough citizen developer inside these organizations that is in your way. Oh, interesting. Which is why they have to go after these PS orgs like Accenture that really just have people who, you know, and like we have we do a lot of work with companies like that. Without naming anyone. And, they're making forward investments into training up their teams to use platforms like Poly AI to do this work, right? Yeah. So I think they're a master conduit for this technology finding its way into the market. And they can do that because they work with those companies on their tech stack, across all departments, across customer service, sales, it, hr, so they know these systems. So if we are to be successful, our technology is to be successful in transforming that business. There are few people better than a GSI firm that no delay of the land, the people, the processes, the roadblocks, who can go this is a good thing to go after this. Let's wait a little bit.'cause this project comes in two years time. We'll change all that and then the geography will be different. Then we can roll over that as well. So look, the world won't be rewritten with AI tomorrow. It will be rewritten bit by bit for many years to come. The GSIs have a huge role to play. I think that BPOs shifting towards doing more of that work themselves is BPOs doing a great thing for their shareholders and their, the future of their business. I think some have done really good jobs. Others are trying to there's a bit of that AI fairy dust attempts with those that are very heal. Labor intensive. Some of it I wish them luck. All forms of AI adoption is good for us and we're allied with all of it. But there is really the whole are you a disruptor changing it fully and rebuilding the whole thing, or are you trying to like, take an already and efficient human heavy cranking thing and keep yourself dependent on it by making it a bit more efficient so that you don't have to do the other thing.

Damien 22:12

I think that makes a ton of sense. Awesome. We're running up on time today. For those of you who listened or watched, we'll make sure to link that Fortune article so you can take a look at what we were referencing for today's episode. Really thought provoking piece. Good stuff in there. And Nicola, pleasure as always. We'll catch you next week.

Nikola 22:28

Likewise. Thank you Damien. You later.