How TWG AI Is Turning Enterprise AI Into Real Business Outcomes

[00:00:00] And if you are running a business right now, you may have noticed there's a quiet shift happening. One that most people are still underestimating. And that is your company doesn't live inside your network anymore. It lives inside the browser. That's where your SaaS apps sit. That's where your data moves. And increasingly, that's where attackers are focusing their attention. So NordLayer has just launched its new business browser.

[00:00:27] And it's designed specifically for small and medium-sized companies that need visibility and control without the overhead of enterprise security tools. What I like here is the balance. You get advanced protection, better compliance, and full visibility into how your team is working online. But without slowing anyone down or forcing them to learn anything new. It feels like a practical step forward rather than another security layer that adds friction.

[00:00:56] So if you want to see more about how it works, please head over to nordlayer.com slash browser and check it out. And let me know your thoughts. But now, on with today's show. What does it take to move AI from just another promising demo to a real operational advantage? Well, today on Business Tech Perspectives, I'm going to be joined by Milan Cooper, Head of Product at TWGAI.

[00:01:25] And they are a company working with organizations including Cadillac, Formula One, Guggenheim Investment, and many more to build AI systems inside live business environments. And he brings with him a rare mix of enterprise experience, having previously led AI product and global technology strategy at JP Morgan, as well as more than a decade advising major companies at Accenture.

[00:01:52] So in this conversation today, we're going to talk about the AI gap and why it is often structural rather than technical. And understand why AI needs CEO level ownership and what's separating companies counting use cases from those measuring real business value and business outcomes.

[00:02:12] And we'll also discuss everything from model choice, governance, concentration risk, Palantir, the future of SaaS, and what happens when AI becomes part of your mission critical workflows in everything from finance, insurance, and even Formula One. In short, we've got a lot to get through today. But enough for me. Let me introduce you to my guest right now.

[00:02:38] So thank you for joining me on the podcast today, especially because you've done a string of flights. So you must be pretty tired. But thank you for joining me. Can you tell everyone listening a little about who you are and what you do? Yeah, sure thing. Thanks, Neil. So I am Milan Cooper. I lead the go-to-market and product team at TWGAI. And we do enterprise AI. So we deploy and productionize enterprise AI for primarily three industries.

[00:03:07] So financial services, insurance, and sports. We partner closely with a company called Palantir. And we have a joint venture with them. And the overall mission of the joint venture and TWGAI is on the premise of the fact that we have the best models in the world. But the U.S. and U.S. companies are way behind in adoption of AI. I mean, we're like way down the list. We're like 24th, 25th, something like that.

[00:03:35] And so our mission is pretty simple. It is, can we find companies and leaders that have the vision and conviction to be disruptive and go all in on this trend and work with us to take their legacy business down to the studs and rebuild them back up as AI native? And so we, and there are leaders and visionaries out there. And the ones that want to go down that path, they're ready to go to work. And it's not such a grim story.

[00:04:04] I know there's so much going on in this space at the moment around enterprise tech and enterprise AI tech, should I say. And especially if you look at what's happening in the labs with open AI, Anthropic, et cetera. What are you seeing here? What excites you about the kind of movement that you're seeing at the moment? Well, the models are phenomenal. We use all the models. So our products and what we build uses the best available.

[00:04:29] And so across all the primary labs, across open source, whatever is the best fit for that particular task from a performance perspective, from a cost perspective, whatever it is. But I think, you know, the structural problems that some of these companies are having is like, number one, who do you use? And how do you avoid locking?

[00:04:54] So like the concentration risk now is way different from concentration risk when you were looking at cloud, which is, hey, you know, you went all in on one cloud. Regulator comes in and says you need to be on another cloud just in case that CSP blows up for some reason. And you actually never had to test that. When I was in financial services at JPMorgan Chase, like you never had to like completely fail over your cloud into a different environment. You just had to prove that you could do it if that like once in a lifetime event ever happened.

[00:05:23] But that's like an infrastructure problem. I was a network engineer. You know, you move traffic from A and you send it to B. And as long as you don't disrupt the applications, like everyone's happy. But concentration risk now is these are intelligence layers. And so like to flip from one model provider, you know, in the event of some catastrophic thing to another, that's like brain surgery.

[00:05:48] And so how do you build up your environment and your use cases and all of that stuff such that you don't have to perform brain surgery if you need to flip to another model provider? And that could be a catastrophic event. One of these companies goes away for whatever reason. Or another provider comes out with a better model, which basically happens every six months. And so how do you avoid those situations and stay competitive?

[00:06:12] Because if you're running your entire AI stack on a model that's now six months behind a competitor and you can't flip because you've got 12 months of work to do that brain transplant, you know, you're kind of in a bad situation. And so, you know, there's a lot happening. The models are progressing very quickly and it's super exciting. But we think you need to have a strategy and a delivery that allows you to access all the models.

[00:06:39] And then you also have a strategy that doesn't just give chatbots to your employees, but actually takes down value streams and takes down full processes. And that's how you reimagine the business over time, not by not chatbot by chatbot. Yeah.

[00:06:53] And I think when it comes to increasing AI adoption and helping enterprises become AI native, as you said there, I think it's important to highlight you've worked with organizations in so many different industries from Cadillac Formula One, Guggenheim Investments and Group 1001 Insurance, where all these decisions that are being made, they carry very real financial and competitive consequences.

[00:07:18] So for what you're seeing here and the work you're doing, what changes when AI moves from those experimentation and pilots and into environments where getting it wrong actually costs real money or performance? Because I suspect for some people listening, that will be very, very top of mind. So tell me a little bit more about some of the work you've been doing there. Yeah. Look, it's the last mile of AI delivery is by far the hardest. Yeah.

[00:07:42] And I'm actually fascinated to see how the labs handle this because going from, I mean, again, these products are phenomenal, but they're consumer products moving into enterprise is a whole different world. And it goes way beyond the technology. It is, is the program set up correctly? So meaning is the person that has the budget actually incented to solve the business problem? And then you know where I'm going here.

[00:08:10] If it's technology that has the budget, then they will deliver technology. If the business has the budget, then they're more aligned to what outcomes everyone wants, which is like, you know, faster, cheaper, better for the business. And so, you know, there's structural things within companies of how these programs get established that could cause problems. The other thing is governance. Again, not sexy stuff here, but like, you know, if you're doing one or two use cases, it's very easy.

[00:08:39] You can govern, you know, just by counting on your fingers. But if you're doing hundreds or thousands or tens of thousands of quote unquote use cases, so units of AI, then how do you govern that? How do you make sure that when a regulator comes in and says, how did you make this decision at this point in time two years ago? So you're able to run a report, you're able to talk to it, you're able to, you know, prove that it matched your AI policy that you wrote of how you do these things.

[00:09:07] And so you need a whole program and a whole infrastructure to be able to do that. And then lastly, which I think is where kind of the labs are getting this right. And I think, you know, Palantir, and I actually think consulting invented the forward deployed model, by the way. But I guess Palantir popularized it. But you need to sit with the business.

[00:09:26] So like, if you're not sitting with the users daily and grabbing time with them to show them what you built yesterday, getting immediate feedback and then turning that around overnight and then showing the speed of iteration to them to get their buy-in, then you're not going to be able to, you know, build something that's ultimately effective.

[00:09:45] And I think this is where the SaaS model breaks down significantly because the one size fits all go to market for software companies is kind of broken. So everyone wants customization because they can get customization because software is new right now. And so the expectation is that like your application, your AI product that you build for me is for me and my team. Exactly what I need, no more, no less.

[00:10:13] And so you get that by the human interaction of sitting with people, sitting with operators and figuring stuff out. And I've got to ask, where do you stand on the so-called SaaSpocalypse that we keep hearing about? And in particular, making investors fairly nervous in this area. Which side of the fence do you sit on that? I think I probably sit on both sides depending on the company and I wouldn't name any.

[00:10:37] But like we've built apps and products internally for ourselves in a week because we couldn't find anything on the market that was exactly what we needed. Like we could go and buy a CRM, but we don't have the kind of business where we need to run marketing campaigns. And so it's like, hey, you just turn that module off. But why is it there? Why do I need to see it? Why is it bloat within the software? And so we just built, you know, we built various products that we run our business on that are just like precisely what we need.

[00:11:07] And if we discover we need something else, then we just add it. So there's like the low hanging fruit of things. So there will be companies that will just disappear, you know, due to that model. But there are companies that are going to continue to provide value either through the fact that they have such institutional knowledge on some of these processes. Like our CRM is relatively simple. Some others are not right. Like the cycle for some companies is extremely complicated.

[00:11:32] And what those companies that provide those SaaS products are able to do is are built upon decades of institutional knowledge. Data moats as well. And what they're able to do with that and the inertia of change, although that is a less positive one. Like I wish that wasn't the case because I think people and companies should have control over their data and not be locked into, you know, a technology stack just because that's where the gravity is in that business. But yeah, I think it's very company dependent.

[00:11:59] And when you're looking at that side of things and you're creating your own software, do you think enterprises will eventually struggle with the management of that and the roadmaps and updates, et cetera? Because there is responsibilities with creating your own things, isn't there, too? When you previously just paid somebody else to do it for you. Yeah, I think you're right. I also think that, you know, arguably the workforce for doing that is increasing as well.

[00:12:23] So previously, you know, let's say 20% of your workforce and you had to write code, you know, probably is going to trend to 100 and we're probably going to get to like 60, 80 pretty quickly. And so, yeah, there is an overhead that you may be bringing in-house, but you arguably, you know, increase the supply side as well on having more capacity to be able to make those adjustments. But yeah, you know, that's on a small scale, like large scale and large systems. Like you've got to have the right setup.

[00:12:51] And before you join me on the podcast today, I was doing a little research. You've said some great things out there. The first one was the AI gap that we're seeing is mostly structural rather than technical. And you also make a very strong case that AI has to be CEO-led. So what does that look like in an organization and how does the conversation change when AI is tied directly to revenue, cost structure, and competitive advantage? It seems like a sensible move, but what does that look like in enterprises that you've been working with?

[00:13:20] Yeah, so I think probably, you know, what we're kind of betting on is that there's this emerging category that is going to exist in the AI landscape, which is context and delivery. Yeah. So like, do you, who owns the context layer and who has the technical and delivery experience to be able to put these things into production? All the non-sexy stuff that I mentioned.

[00:13:45] And so you can kind of see by the shape of our company, we've built it around that premise. And so you still need, you know, extremely high level AI capabilities and data science capabilities, by the way, not just generative AI, but also like traditional ML models. And so we have, there's a public partnership that we have with Polymarket, where we're building the market integrity layer for prediction markets for Polymarket in sports. That's an anomaly detection model.

[00:14:15] There may be some generative stuff in there, but like the core of that is fraud detection, anomaly detection. So there's still a traditional ML out there and anyone in banking and, you know, risk management and people who are looking for fraud on payments rails, like they'll tell you exactly the same thing. There are classical models that are still working very well. And so we built a company around the premise that you need multitude of AI disciplines. You need to have the experience of how to deliver those into production.

[00:14:42] And we have a compliance team. We have teams that know how to engage with regulators. We have, you know, folks on data use governance on to make sure that, you know, all the data that you use to train these models and serve these models is tracked and governed in the correct way and approved. And so all of these kind of like pillars that you need.

[00:15:03] And again, like this is built up over years and years of institutional knowledge that we've kind of accumulated within our company from JP Morgan and financial services and the military through Duke of course background as well in the Pentagon. But then you also need a platform to be able to build all this on top of. And that's where, you know, our partnership with Palantir comes in. And I guess the open question is, you know, where are those other context layer platforms? That's why I said it's an emerging category.

[00:15:33] Like I think there will be other players in the market. Open question on whether the labs will, you know, how the labs will handle context. You know, will enterprises ever get comfortable sending their most proprietary data to these relatively new companies in the grand scheme of things?

[00:15:51] That's kind of one of the other problems is, you know, what we've seen with one of our core products, which is when we got compliance to approve the use of M&PI, so material non-public information, the most sensitive information that you have as a bang on deal information and what's happening in the market and private conversations. When they approved use of that in our product because of the security standards that we hold, usage went parabolic, slowly increasing curve.

[00:16:20] And then literally the Friday that happened, the Monday usage just went through the roof. And so if you as a company don't allow your employees to just freely use these things and make sure that the security is there and the layered protection is there on the back end, then you're going to have a pretty low ceiling on the productivity and the usage of these tools. And so it begs the question, like, who do you trust to give your data to? And that's what we're trying to solve.

[00:16:48] We're trying to be the most trustworthy partner to these companies such that we can get to the point where they approve, you know, the most sensitive information in our platforms and they can get the most out of our products. Big thank you to Denodo for supporting the Tech Talks network and making these conversations possible. Because when your lake house stores the data, the real challenge is getting that data where it needs to go and faster.

[00:17:16] And your lake house stores the data, but Denodo helps deliver it faster. So with real-time access, built-in governance and a business-ready data marketplace, Denodo can help your teams unlock insights without costly duplication. And you can learn more by simply visiting denodo.com. There are many organisations out there that are still stuck in what many call AI theatre or we're still pilot purgatory.

[00:17:44] So from your experience, for people listening here that might be unaware that they're stuck in this phase, what are their early warning signs that their company is heading down that path? And how can they maybe course correct before wasting time, investment, etc.? Yeah, I think a classic, I mean, there's many, right? Yeah, yeah. It's talked about a lot, but a classic is counting use cases versus dollars. Yeah.

[00:18:13] And what I mean by that is like, hey, we've got 50 POCs or 80 POCs or whatever number of POCs or projects or use cases or whatever, in flight, in production, whatever. And the complication of actually calculating value on these things is extremely difficult. Because again, it's easy when you've got a handful, but when you've got hundreds of thousands in the organisation, then how do you have, you know, a common accounting mechanism, for want of a better phrase,

[00:18:41] to be able to say, hey, this delivered X dollars over here, and this delivered Y dollars over here, and we calculated X and Y in the same way, such that your CEO, and we did this at JP Morgan, is able to confidently say to investors on an annual basis and a quarterly basis during earnings calls, we have delivered this value this year from AI.

[00:19:05] Like for a CEO to be able to do that, work backwards into what you need to have in place for him to be allowed or comfortable or committed to be able to give in the street a number. Because as soon as you give the street a number, then they want to see that grow every single year or quarter. You have to have a robust kind of AI governance and accounting mechanism in place. And so that's hard. That's really hard, non-sexy work.

[00:19:33] You know a company hasn't gone down that path if they're still counting use cases versus dollars. Love that. Such a great point. And through your work with Palantir and TWG AI, you're embedding AI directly into core workflows. And I'd love to maybe bring that to life and everything we're talking about here to life a little bit more. Maybe inspire some of that adoption that we're both talking about. Are you able to walk me through a real-world example?

[00:20:01] You don't have to name any names, but of how an AI system can become part of a day-to-day operation rather than just sitting on the sidelines. And anything spring to mind there that would just help bring that to life? Yeah, I can talk more openly about this because it's one of our portfolio companies, which is Guggenheim Investments. Yeah. So Dina DiLorenzo is the president over there.

[00:20:25] And she is one of those leaders that has just great conviction and great vision on becoming an AI-native asset management company. And we have the remit to go all in and be super disruptive in her business. And she's fully supportive and sponsors all of our work. And so where we started were these kind of independent use cases. Like if you looked at them on a page, it would be like, that's just a list of things. And I don't know how they connect.

[00:20:53] And they just look like individual projects. And at the beginning, they were. But what started to emerge is when you map the value stream on top of all those products, on top of all those use cases, it all connects. And so the value stream of asset management is how much money can you pull into the company? Like that is the number one goal. So you've got an origination and you've got a sales team that's out there trying to attract investment and capital.

[00:21:20] Then you have a team that's portfolioizing this capital that's coming into the company. And so what do I buy? How do I securitize it? And how do I build a product that's sellable to the market? And then you've got to find people to buy and invest in these things. And then you have to manage it long term. And so when you map out the very simple high-level value stream of the company, like how do you make money as a business, as an asset manager?

[00:21:49] And then you map all these use cases onto that value stream, you start to see the emergence of an AI native company. So we have some over here in origination. We have some over here in securitization and how they review credit memos and these big complex documents to be able to decide whether to make an investment or not. All the sales processes over here and how they find clients, how they target, how they engage. And we weren't there yet. And we're still not there.

[00:22:19] Like we're one year into probably a three-year journey. But you're starting to see how these pieces are starting to connect together on a common platform. All of these independent AI use cases on a common platform, on a common intelligence layer is having transformative effects on how they do business. And one example of that is just one particular team in Guggenheim Investments and how they can now manage double or triple the number of deals being reviewed per week.

[00:22:48] Because there's no longer this bottleneck of the junior analysts have to review all these documents, pull out all these data points, put it in a spreadsheet, create a slide deck, and then present it to their leadership. And so we're starting to see more throughput of deals that are being reviewed. Well, of course, we're value stream based. So there is a bottleneck somewhere in the process. And the bottleneck at the moment is the investment committee.

[00:23:12] So at the end of the day, to approve these huge deals, someone has to take a look at the deals in detail and approve or reject them and discuss them. So you have this committee where everyone is exactly what you'd imagine, right? Everyone meets, talks about a bunch of documents. There's a ton of bias in there. People present. They run out of time. All that kind of stuff. And so now we're helping make that IC committee more efficient.

[00:23:34] So we fixed the upfront piece, which is like you can review more deals per week, create a bottleneck at the IC committee, and then you take down that problem. And there will be another bottleneck and another bottleneck and another bottleneck. And we just keep going down the value stream until we've punched the whole thing. And you have an AI native value stream and you have an AI native company. And you kind of get these synergistic compounding effects of doing that in a sequential way.

[00:24:01] Notice there are chatbots in here, but this isn't a chatbot product. This is like hard yards of figuring out a business process, data in, what is the outcome, and then build in and wrap in and bend in these LLMs to that process so that you can kind of get the outcomes that we want to get. Love that. And when it comes to transformation, so much has changed. The technology has changed. But the core foundations largely remain the same.

[00:24:28] And looking back, you've spent time with JP Morgan Chase and Accenture. So you've seen some of those large-scale transformation efforts over the years very up close. So what separates the organizations that do actually scale the transformation, especially AI, from those that stall after the few use cases? What are they doing different? I think it's, you know, a lot has changed since, you know, I was at Accenture and even JP Morgan. And technology has changed.

[00:24:57] The models are better. There's new companies emerging. There's new patterns in the market. There is more appetite for change, to be fair, to the U.S. industry. There are still not many leaders that have full conviction to go, you know, to go all in. I think, you know, whether it's, this isn't something that you can do. Like, you have to have a healthy business, first and foremost. This isn't getting you out of a hole, necessarily.

[00:25:27] Like, you have to start from a foundation of, like, we have a healthy business. But then you have to have a leader that can kind of, you know, look through the cycle and say, hey, we're not solving for next quarter or even next year. Like, this is a five-year journey to remain competitive. We're going to have to do things differently. And whether that's transforming legacy from within.

[00:25:49] And we're doing that with, you know, Guggenheim Investments, Guggenheim Securities, you know, Cadillac F1, G1001, Dodgers, and some of the other companies outside of TWG that we're working with. But another way of doing it is actually start completely new.

[00:26:07] Have an air gap tech stack, second people over there, new office, take them out of the day-to-day and say, hey, we're going to build an AI native company from scratch and then move volume over to that company, you know, over time. And so both models can work. You have to have a lot of conviction to do the second model, which is starting a new company, because you're not seeing these, like, early wins as you would if you were transforming legacy.

[00:26:36] There's not all those tangible things that you can point to and say we're making progress, but you potentially can move faster. If you try and transform the legacy, you have, you know, you have a big human component that you have to take care of, which is change management. You have to bring the organization along for the ride in the journey. Otherwise, you know, you're going to get, you know, adverse reactions all over the place and people are not going to adopt the products that you build. Love it. Such a great example again.

[00:27:04] And as we look ahead and AI inevitably will become part of the critical infrastructure and everything from finance, insurance to motorsport there. I mean, what new risks or maybe unintended consequences should leaders be thinking about right now? So I think over the last few years, we've all got excited by the shiny, shiny to a certain degree, but any risks or unintended consequences they should be on the lookout for too?

[00:27:29] I think, you know, I think everyone probably talks about hallucinations and auto risk and stuff like that, but you should really have, it's not a solved technology problem, but you should really have a robust AI governance process to be able to handle that. Whether it's human in the loop where it makes sense, not in the loop where it doesn't defensible, you know, policies such that if an auditor comes in, you could say, hey, this hallucinated, this made an error.

[00:27:57] This was an unintended consequence, but like, you know, we all agreed that this was safe. And so we'll monitor for that and you're able to kind of detect those things, but, you know, you're able to report on that and be open and transparent. And by the way, regulators just want to know, do you have a policy? And are you doing what you said in that policy? And so they're not inventing rules to catch you out. It's kind of the common misconception. It's like, do you have an AI policy?

[00:28:27] Did you write it recently? Do you update it often? And when we come in with zero notice, can you prove that you're doing what you said? Like that's the bar with regulators. But yeah, I mean, those are some of the key things there. Love it. And for anybody listening that are interested in learning more about how you're helping enterprises move beyond AI adoption to intelligence-led transformation, doing it the right way, where would you like me to point everyone listening?

[00:28:56] So go to TWG.ai. And you can also visit our LinkedIn page as well, TWG.ai. Awesome. I'll add links to everything that you mentioned there. And just as finally, as you look ahead yourselves, what excites you about everything that you're seeing, hearing, that you're working on at the moment? You've got access to so much information, so many big names. You're working with Palantir. What excites you right now? You know, I think we're at the heart of it. And so it's a fun time to be in technology.

[00:29:26] We have amazing partners with Palantir. We partner with all the model providers. And we have an amazing portfolio of companies within TWG. And so we've got, you know, our bread and butter in financial services and insurance. But we've got a huge sports portfolio from, you know, the Dodgers, Lakers, all the motorsports division, professional women's hockey league, even rodeo. We're doing AI in rodeo. And so, you know, for a lot of people, the diversity is really fun over here.

[00:29:56] I need to get you back on to talk about AI and rodeo. So there's a podcast episode right there. I think when we're talking about AI, there has traditionally been so much smoke and mirrors and hype. And this year, I've been to so many tech conferences. I've noticed a big shift that everything is focused now towards business outcomes, towards ROI rather than another product demo. And what I love about what you guys are doing there is you're talking that language, maximizing output, decision making, and delivering competitive advantage.

[00:30:25] I think many organizations lost their way being distracted by the shiny, but it's great to get back to this now and love the work you're doing. But thanks for sharing your story today. Thanks, Neil. So a big thank you to my guests for joining me and sharing such a practical view of enterprise AI that takes us well beyond the hype cycle. And one of the many things that stood out to me was that reminder that successful AI adoption is rarely about having access to the best model alone.

[00:30:52] It's about ownership, governance, data trust, business process design, the discipline to measure value in dollars rather than dashboards full of pilot projects. The unsexy stuff, that is what's getting cool. And for business leaders, I think this conversation raises an important question. Are you building AI into the way your organization actually operates? Or are you still treating it as just another technology experiment? Let me know.

[00:31:21] You can reach me at techtalksnetwork.com. But more than anything, big thank you to you for listening. I'll be back soon with another conversation at the crossroads of business and technology. And hopefully I'll get to speak with you all again then. But that's it for today. Bye for now.