Episode 345 The AI Bet: Huge Investment, Job Cuts, and Uncertain Returns
Today on The International Risk Podcast, we turn to the accelerating transformation of the global economy through artificial intelligence. Firms are making aggressive bets on future demand, and mid-market companies are grappling with rising costs, limited visibility, and mounting pressure to prove ROI.
To help us make sense of this, we’re joined by Craig Unsworth, a portfolio Chief Product Officer and Non-Executive Director working at the intersection of private equity, AI, and product transformation. With over two decades of experience and more than 60 transactions across SaaS, data, and B2B services, Craig works closely with private equity firms and their portfolio companies to drive growth and deliver high-impact, product-led transformations.
For more of Craig’s work, check out his Substack: http://chieflyproduct.substack.com/
The International Risk Podcast brings you conversations with global experts, frontline practitioners, and senior decision-makers who are shaping how we understand and respond to international risk. From geopolitical volatility and organised crime, to cybersecurity threats and hybrid warfare, each episode explores the forces transforming our world and what smart leaders must do to navigate them. Whether you’re a board member, policymaker, or risk professional, The International Risk Podcast delivers actionable insights, sharp analysis, and real-world stories that matter.
The International Risk Podcast is sponsored by Conducttr, a realistic crisis exercise platform. Conducttr offers crisis exercising software for corporates, consultants, humanitarian, and defence & security clients. Visit Conducttr to learn more.
Dominic Bowen is the host of The International Risk Podcast and Europe’s leading expert on international risk and crisis management. As Head of Strategic Advisory and Partner at one of Europe’s leading risk management consulting firms, Dominic advises CEOs, boards, and senior executives across the continent on how to prepare for uncertainty and act with intent. He has spent decades working in war zones, advising multinational companies, and supporting Europe’s business leaders.
Related Episodes
Transcript
I think certainly with the enterprise tech. Brands that you talk about, we, we really are still in investment mode. They are writing almost blank checks to build almost spec list data centers and, and really, really throwing the, the entire kitchen sink at it. And they’re making a bet on tomorrow that they’re, they’re not, they’re not serving today or, or retrospectively looking at yesterday. They’re making a bet on tomorrow. With most of them taking an exponential view that, that this will will not just be iteratively, uh, growth, growth in the next few months, but, but huge, huge transformation. Where I see a bigger difference is in those kind of mid-market companies just now who are on the journey with the, the growth, and they’re on the journey with the investment, but they’re really starting to struggle with visibility of investment and visibility of spend. A lot of companies I see and I work with companies from kind of, um, 10 million, uh, revenue, right up to a few hundred million of revenue. It’s that kind of range. They’re seeing their AWS bills, their GCP bills, et cetera, jumping 30% month, month on month because they’re building, they’re experimenting and they’re building new tools and applications which are being used, which is great. Currently, most of the unexpected spend is something that is got away with because. You can have a 20% extra spend on AWS, but if it’s replacing many times that in FTE equivalents investment with people doing jobs, you, you let it go.
Just now. We will reach a tipping point soon, know where CFOs turn around and say, hang on a minute. I need to be able to budget for this. I think it’s all of the above, which is not a cop out of an answer. I think it really has to be all of the above you. You have lots of options of how to deploy capital. You have lots of options of how to, to report and, and strengthen margin, but it all starts with, with adding to, to, to revenue, uh, top line. It all then continues due to making your operation more efficient. I think all of these things lead to, uh, key metrics, which every CFO will be, will be watching obsessively just now. I think one of the, the most interesting things is, is the direct impact on people. And one of the metrics that I’m using now is FTEE, so the full-time equivalent. Equivalent. So literally taking a a, a list of, of people, human people and saying, this is what this number of people used to do. Now this is what people plus agents, people plus tooling, et cetera, can do instead, the delta being the FTEE and that as for as long as that number is larger than the additional spend on hyperscalers, et cetera, then we’re in a good place. But, but there will be a tipping point. I think again, it does come down to the size of company in the, in the mid-market space. I’m not seeing. Those layoffs and redundancies are being attributed to AI artificially. I, I’m genuinely seeing processes, teams, functions, being augmented with new tooling, which is allowing people to do more with less. Uh, and we started with a, a brief of do more with what you have. It, it is now very much in the do more with less, which means account reduction as well. I, I’ve got a few different examples from across the portfolio of clients I work with.
One of them has, has taken a 15% head camp cut because of just adding new tooling. They’ve reduced all of the, the manual, the manual effort. It’s a, a legal tech business and that that’s what they’re doing. That’s what they’re doing for customers. So they’re kind of drinking the Kool-Aid internally as well. Uh, I’ve also seen other companies say, well, we would’ve hired 35 new people. We haven’t, uh, we, we’ve, we’ve maintained our base at 150 heads. We should have hired more, and we haven’t. We’ve, we’ve put agency and we’ve built tooling, uh, and, and changed our workflows instead. I think that’s, that’s more tangible. It’s more viable. It’s more believable and it’s more visible. You can see it happening. I think if I were running a business of a hundred thousand heads and quite fancy, getting rid of 15,000 just now, like we’re seeing with large, large tech. AI would be a really good excuse. I don’t know how much of it is wholly authentic. I think it is happening, but I think that we are also seeing a, a natural leaning, um, of, of business models. I think it’s, it’s worth jumping in at the very beginning of that because I still, I still judge quite harshly the investors who ever were letting the vague spending go, they, they got caught up in the hype cycle and they should not have done that. Uh, so there is a change. I’m seeing more of a change in. Investor attitude than I am in company operations.
Actually, uh, most of the companies that I see at that level are still spending strategically, they’re still building that tactical defense and, and they’re still trying to build and, and deepen a moat and defensibility. But the, the investors have changed their mind on what is, what is vague and what is not vague. Where the, where the. From just spending on AI experimentation, uh, and shifting from innovation into more operationalization and more productization of ai. So this time two years ago, people were saying, uh, we’ll, we’ll experiment with Gen AI and see if we can replace some, some, some bodies, some heads. Now it’s very much a no, we’re building an agent workflow.
We’re, we’re replatforming our operational team and we’re going to have 20 cluster agents, and we’re investing in, in agent number 12 just now. It has a very defined spec, a very defined brief. We know what success looks like, and this is what the ROI has to be for us to succeed and move on to agent 13 in that, in that workflow, I’m seeing much more of that, and I think that’s being driven by. The, the narrative externally, especially with large tech, that, that there is bubble like behavior.
I don’t think it is a bubble, but I think it’s bubble like behavior in some, some areas. Um, you’re seeing individual engineers be hired with half a billion, billion dollar package is insane. There is absolutely nothing that can sustain or defend that. Um, so I think we’re starting to see some of those. Early warning signs of bubble like behavior trickle down. And the people who, who take heed the most are the ones who don’t have the war chests to, to, to put up with that. So again, it is in the smaller companies and, you know, still very much medium, large employers, hundreds, thousands of people.
But it’s, it’s those who, who are not sitting on an unlimited ability to, to write money. I think there are a few interesting parts of that, so, so one is. One is really how it has changed recently, and I think a lot of that is legislation driven. We’re seeing the us, the uk, the eu, and Asia, all going quite, quite dramatically different directions when it comes to regulation, whether it’s, um, how to extend GDPR for an AI world, whether it’s looking at, um, individual data, data privacy, whether it’s thinking about ai, uh, defense, et cetera. And I mean, uh, the defense of the structures of AI not used for defense. Then we get to the, the world situation that we’re in just now where defense is also the capital D in front of mind. So I think there’s a lot of things going on just now.
One of the, the things I, I’m really curious to see is, is whether or not we, we hold our nerve and continue developing AI with the original vision, which for some people, if you take the Altman’s of the world, et cetera, is this, this general intelligence position. Historically, we’ve been really bad as humans sort getting halfway down a road with a new technology and then using it for something else. We’ve militarized it, we’ve, we’ve weaponized it. Um, socially we’ve turned it into something that actually is about increasing wealth and revenue for, for a very specific niche, and we’re quite bad at that. I wrote a Substack article about this recently, which was, uh, quite plainly titled, you know, humans might not be smart enough for a GI. And I, I do think we have a track record of, of getting halfway down an innovation cycle and then completely messing it up because we act like humans. It’ll be very interesting to see if we act a little bit more like machines and what that, uh, does directionally here.
I think it’s, it’s really telling that the. Vast majority of who we will talk about today are American companies. We haven’t seen challenges from, from anywhere else. Uh, you know, there are, there are still no major AI leaders coming out of Europe. Um, nothing in the Middle East, nothing in in, in Southeast Asia, et cetera, really. And then we have China, which is in a different space. And I think what we will probably see is a split strategy, a little bit like China’s taken with the EV and the battery rates as well. He’s going after a different part of the market. So any of your, your listeners in the UK will, will be really aware just now that when you watch TV or get an ad on on your phone, it’s for an yet another un previously unknown EV manufacturer. You have to Google and say, never heard of these guys and the market’s being flooded. I think we’ll see more of that, uh, as well. And the, the uk, the eu US needs to need to make decisions around how much infiltration is going to be comfortable.
Uh, we saw previous U-turns with, um, things like Huawei in 5G, et cetera, in the uk. And, uh, the, the reversal of that decision is still why you can stand in London King’s Cross and get absolutely no phone signal, uh, because the infrastructure lent took years and years and years to catch up from, from what was a LEAP proposition. So it’ll be interesting to see what happens there.
I think a couple of things changed. Uh, and for me it was Christmas 2022. Uh, we were all sitting down for dinner and my, and my, my nan, um, said. What do you think about this generative AI stuff? And I thought, oh goodness. We really, really hit a kind of. A new, a new level of accessibility, um, and typical humans. You know, we took the, the lowest common denomination of intelligence for us to, to, to be hooked in. And it was, it was vanity, it was the AI photos on Facebook, et cetera. And it was a real consumer touchpoint for everyone to think about this, this new ai, without realizing that that AI is as old as my, my nan indeed. I think what what then happened was people started to, to, to do what we do as humans. We started to think, well, what is, what is the use case here? But let’s do it in a human way. What do I not want to do anymore? Can we make AI do it? And philanthropic recently released, you may have seen it in the last couple weeks, a really interesting chart about the difference between actual usage, um, of AI and observed capacity in terms of careers. It’s a double layer chart, um, mapping where we’re using AI and artificial intelligence, uh, systems today and where we theoretically could use them. It doesn’t look great if you are in a management role, uh, computer role. Uh, if you fancy doing something in architecture, you’ve got, you know, a really large potential displacement there. If you want to bake bread or be a gardener, great crack on. Uh, there, there’s a real, a real difference there and a lot of it. Points to, again, humans being human and thinking great. We’re going to own the machines a bit like the Jetsons, which I grew up watching and and loved and was desperate for the world where we’d all be kind of served by robots, et cetera, and we’ve taken that approach to it like we always have in any other industrial revolution. It’s always been a case of, well, what do I not want to do anymore? I’ll get the machines to do that. I’m fascinated that now we’re actually at a point where we, we are going to have to look at ourselves in the mirror and think, what am I actually not very good at? Rather than what do I not want to do? And that’s gonna be the huge change. And I think the west and the, well actually the global North is probably a better way to, to think about it, is going to take a different approach to that. And it’s going to, it’s going to take a longer time. For us to be honest with ourselves and say, actually there are things that machines can do better than us, rather than just things I don’t want to do anymore.
I think it is enough just now, I think. I think Joe Blogs on the street can. Can really develop enough of a skillset here with a little bit of investment. One of the things I, I get involved with is, is mentoring. And I’ve been really, really, um, I’ve been really developing a large number of relationships with underrepresented groups recently because I’m determined to try and play part in, in this, this new revolution, not not being something that widens inequality even further. One of the things I’ve been able to do is work with, with groups of people who usually wouldn’t think that they had access to, to things here. And what we’ve been able to do is put together a, a, a really basic curriculum, which is an hour a week, and I’ve been helping supply issues and tasks and challenges, et cetera, with training notes. And it’s an incredible how quickly people in that, in that group have been able to, to grasp the, the principles, the fundamentals, and then some really quite advanced things. I’ve got, I’ve got people who. Three, four months ago, I’d never logged on to chat GBT, who are now building agent workflows in, in NA 10. Uh, and they’re really basic is to help with life admin or it’s to help with, with school homework, et cetera, for the children. But it’s keeping them current and I think, I think investing an hour or two hours a week like you would if you were trying to learn a language or get good at sport or, or any other hobby. I think is a really important message to get through just now. I’d like to see, I’d like to see that that message coming from government and regulators about the individual responsibility we have to, to stay ahead of this, and that’s for everyone. Before you get into, uh, well, how, how else do you develop your, your t-shaped profile at work? How do you, how do you get career specific tools? Because if you are sitting there in a legal profession, uh, the, the old slightly cliched adage now of, you know, an AI won’t take your job, but someone using AI will completely applies. There’s also the chance that AI might take your job as well, but in terms of getting there to a first line of defense. Pick up some tools, experiment and, and start learning. Learning about the future.
I think you’re right. I think the LLMs are, are grabbing the headlines. Uh, even, even LLMs are not really LLMs. So some people bundle perplexity in there, which of course is not an l lm and it’s on Right. But it, but it, it’s doing a similar job. For me, the, the glue that connects everything together has to be your workflow. And, and right now, you know, I run a, an advisory business working with multiple clients, all of whom are private equity firms or, or their assets. Uh, and I, I’m able to punch way above my weight, um, and what I would be able to years ago ’cause I’ve leaned into agent workflows. Little things like note takers, you know, having every single meeting that I’m in. The note taker is there, making sure that I’m, I’m capturing every single action. I can look back and, and absolutely be able to be sure what was said. I know how to, to bring quotes into documents, et cetera. All these things that would be so time consuming. Otherwise, diary booking, capacity planning. I optimize my, my diary every, every week I’m looking ahead and thinking, well, how do I almost. Like Defragment my hard drive, uh, and I kind of defragment my calendar and I use AI to assist on that. And these, these are all tools that are there. You know, Gemini is incredible at, at helping me with calendar management. I’m Claude Chat boutique. Great at helping me write up thoughts and, and notes, et cetera. I love. Playing them off against each other as well and saying, well chat GBT. Give me this summary of these notes. Claude, what do you think? I’m getting a second opinion, and suddenly, before you know it, I’ve got, I’ve got a virtual ea, I’ve got a virtual researcher, I’ve got a virtual archivist, I’ve got a virtual. Kind of classic 200 year old stereotypical secretary writing up every single, uh, minute of every single call and meeting. I have all of these things come together to make a, a workflow that, that really ends up providing me with three or four FTEs that we chatted about before that kind of full-time equivalent. Equivalent that if I’d had to hire, would make a rather bloated team for the work I do. I think the ones who are making the most headway just now got in early. They, they really did get in early, and I wrote an article recently about, about the, the, the impact and the cost of waiting. With ai, it’s no longer a, you miss one thing, you fall one step behind it. It’s really, really changed and, and that compounding effect of missing an opportunity is something that I’ve, I’ve seen a lot of in the companies that. I both work with and just observe in the bus, in the segment.
Generally, the idea of of getting started seems to cripple everybody. Where do you begin? How do you do it? And there are several people out there now that you would be able to reach out to and bring in as an AI enabler in your business. People who know your sector, who are technology leaning and are able to actually assemble experiments and hypotheses, et cetera. So you can get started, but. All the basics are there that we’ve seen before. You, we’ve been, we’ve done this, we’ve been through the shall we Go Digital period of time, which was insane that people took so long to think about. There’s still something that hits me every time someone sends me a contract that I have to physically print out, signs, scan, and send back. Uh, knowing that the e-sign has been with us for what, 15 years. I don’t think we’re gonna have such a delay in adoption here, but I think the, the flip side of that is we’re gonna have things, people, businesses, processes, products that just die because they’re being completely replaced by something that did lean in and did embrace, uh, this new technology sooner.
A big yes to everything you just said. Uh, I’ve, I’ve been looking at universal basic income for about 10 years and wondering what that could look like. The reason that you and I are sitting here at, at three 30 on a, on a Tuesday and not not tending the field, is because an industrial revolution happened, and that was, that was net positive. It allowed us to be out there doing other things. This, this time, the net positive is, is far less clear, and I think we will have. Mass employment change. I’m not quite saying mass unemployment yet, but I think it’s definitely realistic that every single person’s job will fundamentally change in some way, what that looks like. We need to, we need to be mapping out, we need to be looking at it and, and you know, if I, if I was in a treasury role or a revenue role, I’d be looking at what that does to, to my tax profile because we’ve always assumed that AI would come for the low paying jobs, low paying, uh, tax contributors to the impact being lower. At the point where you’re taking 30% of your, um, legal, um, workforce out, that’s a different impact on, on tax take at that point. We need to be modeling that out, and I, I hope somewhere someone much smarter than, than I am and much more connected to central government is looking at this and absolutely modeling out scenario A versus B versus C. Because I, I, I hear very, very little about, about how we fund this, what this, what this opens up, what this enables, what the impact could be positively, but specifically also, how do we make sure this isn’t a, uh, a, a poverty driver? How do we make sure this isn’t, isn’t eradicating the middle class, for example? Um, these are all things that, that, that would have significant political impact, uh, significant societal impact and, uh, significant individual impact. There’s gonna be a large number of people here who are in what they have always thought were very safe professions. You know, you go, you go to university, you get your undergrad, you get your, your masters at business school. You go into consulting. Eventually you make a partner, you retire, you play golf there, that, that’s been defined and uninterrupted as a career. Until now, you can replicate that across investment banking. You can replicate it across legal, lots of these very safe, um, safe jobs or previously safe jobs. I’ve got a a, a friend who. He is in his early sixties. Uh, he’s got three kids who’d just gone to university. Um, he started late. Um, and, uh, they, they all changed their minds at the very last minute about what they were doing. Um, and it was really interesting to think about, about what, what jobs they were going to do before. So one was going to do straight medicine, one was going to do, uh, a business degree, and one was going to do a finance degree. They’ve all changed to, to go to different universities, lesser universities in terms of profile, but with more specific and more applicable technology. So one is now doing a medicine with technology degree, which I hadn’t heard of. It sounds fantastic and it’s absolutely about using. AI using ML to look at diagnostics, et cetera. It’s a fascinating, uh, deviation. But without that, she would otherwise be joining a, a, a quite a long list of, of professions that we now have a question mark on to say, how will that change?
So I, I love binary black and white situations and um, and so my brain would really like to stick with just those two options, and you can navigate between the two. You. Can build a preparation plan for them both. The reality is, I think it will end up being somewhere gray in the middle. And, uh, and I think what, what we probably will have to ask ourselves at one point is how do we change, uh, taxation? So what, what would the normalized profitability be? What is a a, an un-normal, uh, what does a variated uh, profit look like with. The example you mentioned of everything moving to AgTech and having 100 unemployment, the profitability is changing significantly there. How is taxation shifting from income tax, which in, in your example, has been lost entirely to the equivalent of, uh, corporation tax. Uh, for example, that’s gonna be a really uncomfortable con conversation to have because universal tax, bandings tax, gradings tax policy, et cetera. It is a one size fits all policy and what, what, what you describe and, and my rebuttal to it would require an individual company by company tax policy, basically just got every single politician zero votes. So, uh, so I think it’s, it’s a really complicated, um, uh, time to, to be thinking about how you balance technology policy, how you look at fiscal planning, how you look at societal society and welfare. All of that needs to come together. In not just un not just unison within one party or within one governing structure, or within one country or within one block of countries, but globally, and I dunno if you’ve noticed, but we don’t seem to be able to agree on anything globally just now at all. So I, I think that is, that’s a real, real pending challenge for us is how we do that. Because first mover disadvantage would apply, you know, if the, if the UK suddenly changes tax policy to be. Harder hitting on AI profitability, for example. Guess what? Every single AI startup, every single AI business will just decamp to Ireland or, or a different, a different country that hasn’t enacted that policy yet. So I think we, we have a, a, a lot to do there. I’m not sure how positive or optimistic do you feel about the people who, who are in charge in a, in a space to regulate and legislate, but, but I would definitely like to see a few more of them leaning in a bit more heavily here with some experts to guide them as well. So I think, I think we will see a couple of big changes this year. So one is the, the relatively recent ability for companies to build their own foundation models and their own frontier models.
I think being able to, to say, we now. Have harnessed the power of all of the LLMs. Plus we’ve got all of our proprietary data, plus we built it into our own proprietary workflow. That suddenly changes everything because it, it really puts us in the position of potentially replacing humans on mass. We, we will see that happen gradually, I think because of the political, societal, um, sensitivities there. And then I think there’ll be a, a relative tsunami effect where once one person moves, the rest will will follow really quickly. Loads of risks. Uh, and I guess we should talk risks because, uh, that is the, the subject here that there are, so, there are so many risks. Uh, my risk register, I look at this the other day, but the average risk register I work on has gone from about 40 to 50 lines to around about 200 lines in the last 18 months. The, the risks I’m looking at primarily just now are, are moat and defensibility, and how replicable is, is your, your business without a significant proprietary data set. Which doesn’t mean all of your data needs to be proprietary, but you need to be processing in a proprietary way or aggregating in a proprietary way. Your entire business could be replicated. Overnight. You know, I, I used to sit in, in board meetings and investment committees and talk about moats being 12, 18, 24 months. Um, I don’t know many companies that have a moat of more than 12, 18, 24 weeks, uh, just now. So I think that there’s a, a lot to be done there. The other risk is, of course, that very. Very well tooled and very well equipped Humans in your employee base, um, will be the ones who are a most critical to, to you as a business. You wanna keep them for the longest, but b, they’ll be the ones that could very easily leave and go into something else as well. So I think we, we tread a bit of a talent tight rope moving forward as well. And then I guess the third one that, that I think about a lot just now is just the change on accessibility, uh, as well. So the opportunities that that used to exist for very small companies or only for very large companies, how much of that is being rewritten just now? How quickly can you legitimately compete with a business that is now a thousand times your size? I think there’s going to be examples of where it’s now completely possible to do that. That creates opportunity, but there’s also going to be new examples of you have no chance, uh, you might have had a, a tough journey to compete with, you know, the David versus Goliath before now game’s over. So I think there’s, there’s a large, a large risk there before we even think about hallucination, inaccuracy. And then the one that I think people are going to be most reluctant to talk about is the addiction aspect of this as well. We’re finally seeing after nearly 20 years of social media, a grownup conversation starting around social media addiction. I think the, the, the, the AI assistant addiction, dependency, et cetera, confusing the fine line between what is machine, what’s human, is going to become a societal, uh, challenge for us also.
Defense comes to mind first, I think, uh, humans making decisions in a defense setting have led to some pretty disastrous outcomes in the past. There is hope and potential for machine aided or machine control decisioning to produce better outcomes, which could mean fewer people dying, which would be a, an incredibly positive thing. There, of course, is the, the risk that a bad human plus a bad machine can create an exponentially worse outcome as well. So I think that there’s a real issue there. Protectionism. Uh, and, and trade protectionism, I think are, are also significantly interesting. Who is going to fund the chip, um, shortage? Uh, who is going to fuel the water shortage? Who is going to take away resources from humans to give to machines? There’s a, there’s a lot of really big questions coming. The data centers that we talked about, uh, we haven’t, haven’t, uh, taken an overlay of where they are, but when you do take an overlay of where they are and then you, uh, you look at that mapping of where flood risk is, drought risk is, et cetera, they’re all in horribly inconvenient places. Uh, where actually flood risk is really high or drought risk is really high. Electricity consumption, you know, would, would. We’re nowhere near getting on top of clean energy yet as a, as a, as a whole world, I’d say that able to be semi sunk. I suppose coming from Scotland where we produce more energy than we use the global market doesn’t, doesn’t reflect that in, in prices, availability, infrastructure, reliability, et cetera. Um, add in 20 data centers, multiple football pitch data centers, and that that antagonizes that situation. So I think that there’s a lot of stuff coming and again. I can’t remember a time where we’ve had so many risks that would all come back down into one very narrow area of governance, either within a political setting or a, or an enterprise setting. Uh, the, the job of the CIO in, in some of these large enterprises that we’re talking about just now is 10 20 30 x what it was just five, 10 years ago in terms of responsibility. Thank great to, great to be here and, um, yeah, I look forward to, to seeing all the rest of your content on YouTube. I hadn’t seen your YouTube. Uh, I, I looked earlier before I came on and it’s an incredible wealth of, of, uh, content, so I look forward to joining that and exploring the rest of it as.