How Valiance Fixes the Enterprise AI ROI Problem

[00:00:03] What if the biggest barrier to successful AI adoption isn't the technology itself, but the way organizations think about value? My guest today is the CTO and value partner at a company called Valiantys. And they are an AI native consultancy that is taking a rather unusual approach, but refreshing, in how they help organizations adopt AI.

[00:00:27] Instead of charging clients based on time spent, they align their success with the outcomes that their clients achieve. In other words, the focus isn't on deploying AI, it's on delivering measurable value. Now this might sound obvious, but as you'll hear in today's conversation, many organizations are still rushing into AI projects without clearly defining what that success should look like.

[00:00:52] So yes, they are investing heavily in tools and pilots and experimentation, but they often struggle to go from proof of concept or to move beyond proof of concept or demonstrate a meaningful ROI. But my guest today has two decades of experience in digital transformation and enterprise architecture. And we're going to explore why AI is changing the traditional build versus buy equation, how Gen AI is reshaping software development,

[00:01:22] and why some long held assumptions about SaaS, middleware and enterprise tech could be challenged over the next few years. So if you are a business leader trying to separate opportunity from AI noise or just wondering how organizations can move from experimentation to real world results, that gap can be notoriously difficult. I'm hoping to give you some insights that will try and make you think differently about where that value is actually created.

[00:01:51] But enough from me, let me introduce you to my guest right away. So thank you for joining me on the podcast today, Dom. Can you tell everyone listening a little about who you are and what you do? Neil, thank you for having me. I'm very pleased to be on the podcast. And so who am I? I'm Dom Selvin. I'm CTO and value partner at Valiance. We help enterprises get measurable value out of AI quickly.

[00:02:17] And we're relatively unique, although others are trying to catch up in that we only get paid when they actually do realize that value. The role title, I said value partner. CTO is well understood. Value partner, maybe not so well understood. It is quite deliberate. We don't bill on hours. We're an applied AI consultancy that charges against outcomes. So we actually work out what's going to make you as a client make money.

[00:02:45] And then, you know, there's a shared interest in that approach. I'm relatively long in the tooth. You know, I've been doing this for quite a while, 20 odd years, 20 plus years in digital transformation, enterprise architecture. I've most recently had a stint as a global CTO at a global systems integrator. I've been a board member on the Mac Alliance.

[00:03:11] You know, we've spent years arguing for composable architectures and how it actually benefits the enterprise. Even before it became quite a fashionable term and a fashionable enterprise concept, I was pushing that banner quite hard. And that lens has carried very seamlessly into the AI space, particularly into the agentic space that I'm really focused on.

[00:03:38] And it lends itself very, very well to the sort of diverse nature of the architectural estate that most enterprises have that they want to deploy AI into. So, you know, my history has all led to and all coalesced to where I find myself today. And you have had a fantastic career. And looking back, I mean, like myself, you've seen all these transitions from mobile to cloud to now AI. And AI is changing absolutely everything now.

[00:04:06] And I was reading before you joined me on the podcast today, you bravely argued that that traditional consulting model is broken when it comes to AI adoption. So let's start there. What are the biggest mistakes you see organizations continue to make? And why do so many AI projects still fail to reach production? There's a few exaggerated stats last year or the year before. But what is going wrong here? What is happening?

[00:04:31] There was that MIT report on 95% of projects fail or fail to hit ROI, all conflated out of proportion. And, you know, you've got to make these claims to be heard in the maelstrom that is a media these days. But the reality is, you know, most enterprises are spending pretty heavily on AI and those enterprises are starting without actually having a definition of what success looks like.

[00:05:01] You can't really bank on a return that you've not actually specified in the past. That pressure to do AI is incredibly tangible and real. But the discipline to actually define the outcome in the first place, that's what's missing in a lot of these cases. So, you know, I'll hold back to my title.

[00:05:20] You know, identifying the outcome before you start to roll out the technology, figure out what the business value is going to be before you start to do any sort of consulting or any sort of solutionizing is really what you've got to think long and hard about. So, you know, firstly, the project is framed around the technology, not the business problem is trying to solve. And that's, you know, mistake number one. Two, you know, you've not figured out what a metric is going to look like from the outset.

[00:05:49] Nobody can actually prove after the fact if you haven't figured out what you're going to compare against at the beginning. You know, you don't measure it. You can't, you can't, you can't quantify it. And in mistake three, you know, the billing model rewards the wrong thing. A consultant, this is why I think the consulting model has changed. If you pay a consultant by the hour, there's no reason to finish early. There's no vested interest in achieving the value goal, that metric that you've identified earlier on.

[00:06:18] And, you know, it's kind of analogous to, you know, hiring a builder by the day without an architectural plan with no blueprint. And there's no impetus for that builder to finish. And actually, there's no impetus on the builder to agree on anything either. Because they're going to keep going for day after day after day and keep getting paid.

[00:06:40] And looking back at both of our careers, I think for years, enterprises were always traditionally encouraged to buy software rather than build it. And you believe AI is turning that logic on its head. And I would agree with you as well. What has changed? How should leaders be rethinking that build versus buy decision now? I've been a big proponent of buy over the years.

[00:07:05] Figuring out what the business problem is, a lot of businesses have similar shapes and similar sort of traits that you can apply a template across the lion's share of the business and have a lowest common denominator approach to running that business. And the reason why buy one, in my view, was because building was actually prohibitively slow.

[00:07:30] It was difficult to articulate the needs of the build. Maintenance was a burden that you would carry forever. However, those two very large factors in a buy consideration have been diminished massively by the introduction of AI. The coding agents are very, very powerful nowadays.

[00:07:55] That ability to spin up custom software at a fraction of the time required. There's still token costs and various other things that factor into this. So, you know, the days of the subscription-based token charging may well be in the past, but that's maybe another podcast in its entirety. But the old maths was pretty straightforward.

[00:08:24] Custom software costs more to build, more to keep alive. And so you did the risk assessment and you bought. Generative AI has massively cut that cost. The first half of that equation has flipped very, very easily. There's still the maintenance cost. You've still got to make sure you keep the lights on and things like that. But the new question isn't, you know, what should we buy?

[00:08:48] It's what's so specific to us as a enterprise that no vendor will ever build it well because it won't, it will be aligned to a lowest common denominator. You still do have to own it. You still do have to govern this. You still do have to maintain this. But it is now an order of magnitude, probably more cheaper to build than it was ever to buy.

[00:09:11] So if generative AI can dramatically reduce the cost and complexity of creating custom software, what does this mean for established SaaS vendors and middleware providers? I've heard a lot around the SaaSpocalypse, especially if they're building businesses around solving some of these problems. Will they lean in on the fear around the responsibilities of support and maintenance? What are you seeing here?

[00:09:37] I'm hearing a lot of rumours of funding for et cetera, lowering for SaaS vendors because of those fears around SaaSpocalypse. Where do you stand on everything that you're seeing here and how it's all unfolding? SaaS is dead. SaaS is dead. That cry has been loud for maybe six to 12 months. And the SaaS providers will adapt.

[00:10:01] They're very big and there's some very smart people working for those SaaS providers. But any company whose product is convenience, they're now competing with a customer who can generate that convenience for themselves on a Tuesday afternoon. Yeah. You know, if they know the requirements, if they know the capabilities that they're looking for, and they know the targeted specific niche that they're catering for, that's a pretty easy build now.

[00:10:30] But the exposed category now is one whose value is simplifying complexity or wrapping a workflow. And that moat... How many times have you heard the word moat these days? It's just been... It's like the word du jour. But anyway, that moat is... It was always rented. It was never really owned. And that middleware was probably the most exposed of all. Integration, glue.

[00:10:55] It was a tax that the businesses were paying because writing that glue, writing that software, was actually quite hard. And it was relatively tedious. You know, mapping from one attribute to another attribute, making sure it all plumbed together. It was difficult and hard to maintain the concentration required for that. But that's no longer hard. That integration is actually no longer hard. SAS probably doesn't vanish, as I said. But the survivors are holding...

[00:11:24] They're holding something that a model can't regenerate. That's the... There's the network. There's the proprietary data. There's regulation position, regulator position. And there is workflow lock-in in there as well. But every per seat subscription now is a question that these buyers can answer differently. You don't necessarily need to pay for a seat that you're not going to use. And in fact, who sits in that seat as well? Is it the agent that sits in that seat or is it the human that sits in that seat?

[00:11:52] So, you know, I wonder whether thinking of it in terms of a... You know, the calculator didn't necessarily kill mathematics. All it did was made the skill available to everybody. So it killed the people whose best skill was mathematics. You know, that may be a little bit of a far-fetched analogy or a forced analogy. But the software that does the arithmetic is in that same position. It's a skilled piece of software that anyone could build now.

[00:12:21] And we spoke about at the very beginning of our conversation how many organizations are struggling to prove AI, ROI. Tech does love a good acronym. It's two right there. But especially when you look beyond pilot projects. So when you work with clients, I'm curious, what are the first signs in an AI initiative or an AI project that is creating genuine business value rather than just more excitement and hype that's actually going nowhere?

[00:12:48] Are there any signs that you've got used to seeing now that you can spot early on or is it not till later? Yeah. It's pretty early doors. You know, there's... My northern is slipping in there. But it's seeing these characteristics of adoption and of value. We've not got enough of a data state to make any sort of anything other than empirical observations.

[00:13:17] But I think the first signal is actually quite a boring one. And that's when people stop actually talking about AI and just use it. It's actually just disappeared into the workflow. That's their day-to-day life. The excitement was in the demo. That excitement does live in the demo.

[00:13:35] But the value that the businesses realise that you as individuals within that business, that's going to be realised on a Monday morning in the routine that no one narrates anymore. It's embedded in the day-to-day life that it's now a de facto part of the state's quote. It's not your fault that your agentic AI systems are acting outside of compliance.

[00:14:03] There are just too many data sets to guardrail them all. But with Denodo, you can now organise your hundreds of data sources into one layer and govern your agents with a single approach. Try it now with Denodo by visiting denodo.com to learn more. There's also this growing belief that code itself is becoming less of a differentiator.

[00:14:29] So if that is true, where should organisations be focusing their investment if they want to create that long-lasting competitive advantage that they're searching for? When everyone has access to the same models, then you mentioned the code. The code they write starts to converge because everybody's got access to the same model. Who is it who said, if you try the same thing and expect a different outcome, that's the definition of madness?

[00:14:59] I think I've completely horrifically paraphrased that badly. But I think you get the point. The advantage moves to the things that the model has never seen. So this is the thing that the model's never been trained on. And that's your business. That's your institutional knowledge. The code that was produced by human minds was the moat because it was hard and it was scarce.

[00:15:26] It was learnt over many, many years. There's this sort of myth of the 10x developer who's able to turn out code and so on and so forth. But generative AI has actually made that abundant and cheap. And abundant things do not differentiate. The durable advantage that businesses will really see is underneath the code. It's the data that you own.

[00:15:54] That's institutional knowledge that you've accumulated over time. There's the systems that you have connected. The people who have accumulated that knowledge capital as well. These are assets that a competitor in the same vertical with the same model access, they can't replicate it. Because they're a product of that particular enterprise's specific history. Where they've come from. What decisions they've made in the past and so on. Code absolutely you can see on top.

[00:16:24] It's far easier to copy than it used to. The way to think about it, and it's kind of nice on this, is that if you've got two chefs, same knife, same recipe, the one that's got the relationship with the farm will probably produce a better dish. Even though they've got the same models, same knife, same recipe and so on, the person who's got the same historical relationships and so on and so forth is going to have an edge on the other one.

[00:16:52] So if you invest in what you only have, then everything a competitor can prompt their way to is by definition not your advantage. Love that. And we're doing a little research on you before you join me on the podcast today. I was reading how you often talk about proprietary data, ontologies, integration depth and governance. How all these things are collectively becoming the new sources of value.

[00:17:16] Can you expand on that and why these areas matter so much in an AI driven world and also how leaders in organisations can strengthen them too? Yeah, I mean, anybody who follows me or listens to anything I've got to say, I think every second word that I use is ontology. And in fact, my wife, she actually hates the term now because it's one of those. I've said it so often, it lost all meaning to her.

[00:17:40] But the AI is really only useful when it understands the context that it's operating in. If you strip away any of that context, you get answers that sound right, confidently wrong, but they can't be trusted inside of real business. Everyone's heard of hallucinations. Everyone has heard of the sort of compounding errors and things like that that happen within an AI.

[00:18:02] That it's it's it's it's sycophancy and it's its ability to confidently state erroneous information as fact is born out of the fact that it is a statistical model that predicts the next word makes most likely word to come out of a sequence of word based on the nearest neighbours behind it. Now, in a business context, that's not good enough. Yeah.

[00:18:27] You've got to be able to trust your systems that that are being used to power your business. And if you don't have something that's describing. So let's work from the bottom up. So your proprietary data is yours. It's raw and it is fact about that that you've accumulated about your business, but it doesn't have any semantic meaning to the business as a whole.

[00:18:50] You know, what is in table A, column B and, you know, cell C is exactly that. It's the data that sits in table A and so forth.

[00:19:01] If you apply a semantic layering on top of a semantic meaning on top of that, how it actually impacts the business in a finance context or in a operations context or in an engineering context, then it actually actually elevates it from being raw data to being actually relevant and intelligible information that both a human and an agent, an AI can reason against.

[00:19:28] So that proprietary data is important, but the ontology, the nouns and verbs and relationships and meaning that describes the underlying data is massively important for agents to be able to reason properly against them. You know, you need to know what a customer is. You need to know what a return means. You need to know how an order relates to a contract. You need to know how an individual pays over time, you know, whether they expected to hit the full 90 day payment terms or whether they pay early.

[00:19:58] You know, those sorts of things are important aspects that the data tells you, but you have to interpret. This magic layer gives you that interpretation. I mean, the integration depth really matters because, you know, a brilliant system that can't reach the place where work happens is actually a clever toy. It doesn't really give you much more than just being a bit shiny. And then you mentioned governance as well.

[00:20:23] And it's something that's kind of invisible. People don't realize quite how impactful it is to the entire estate upon which, you know, your software is all working. That governance tells the agent what it can access. It tells the human what they can access.

[00:20:43] It tells the agent, you know, what's what this particular entity within the enterprise estate means and why it means that and where it can be used and how it can be used. That governance isn't something that just sits after the fact. It's woven and interspersed through the entire architectural estate as a first class citizen. And everything flows through that governance ether to be able to say, yes, this is OK. This is not OK.

[00:21:12] And and this is secure. This is not secure. You can't have a an enterprise stack without governance. 100 percent with you there. And I think over the last, what, three years, we've seen an AI gold rush of sorts where nobody wants to get left behind. There's been a fair amount of bandwagon jumping as well.

[00:21:35] And many executives, of course, felt that pressure to move quickly with AI while also managing risk, compliance and security concerns, as you mentioned there. So what's the secret to striking that right balance between experimentation and wanting to get out there and lead the way with accountability and making sure you are generating business value and better outcomes, etc? I mean, instinct is to treat speed and controls a trade off. Yeah.

[00:22:04] You know, you do not in reality. The thing that actually slows the enterprise down is when, as I said previously, the governance bolted on after the fact. You can't you can't you can't sort of say, I'm going to build the platform and then I'll add a layer of of compliance and governance afterwards because it's it's just going to be too painful to intersperse within the other the systems.

[00:22:34] Governance, security, accountability designed from the start are an accelerant, not a friction vector. And they are what lets you scale that pilot. Without having to throw away the initial code and just keeping the learnings and then start from scratch, because you can you can you can roll it out because it has to actually meets the risk requirements of the business. The leaders, they don't need to move slower. It's all very well saying you're going too fast.

[00:23:04] You're trying to adopt things too quickly. And the reality is the space is moving fast and you can't be left left behind. But you as leaders, you have to move more deliberately. You have to know what you're testing. You have to know what success looks like. You know, going back to what we described earlier, identifying the target value and identifying the near and the midterm horizons of what you actually want to achieve.

[00:23:28] If you have that north star to aim for, then you can move at that's at the speed that AI allows you to. Containment of the blast radius is also important. You know, your pilots will be quite self-contained, but the holistic rollout to the entire business is not going to be self-contained. So you need to make sure that you iteratively build it all out. The accountability is not the attacks on moving fast, but it is the permission to do so as long as you take that accountability into account.

[00:23:59] And I think the changes that we've seen over the last three years make it almost impossible to predict the future or see how quickly changes are going to take hold. But if I was to pull out a virtual crystal ball here, ask you to look three to five years ahead, how do you see enterprise technology stacks looking in the future if your vision of an AI native organization becomes reality?

[00:24:24] And any assumptions about software and consulting that you see will be quickly outdated in the next few years? How do you see all this playing out? Any prediction in this space is futile. You can see how quickly they age and how slowly we are to predict things, but how quick things come into pass.

[00:24:48] So I think I'm going to sidestep your question a little bit, if you don't mind, and talk about the pattern that I actually see rather than any predictions. I think the cycle has compressed so far that what was a three-year plan has actually now lived in three months in reality. So as I say, the question is more the pattern, not the timeline. And I think the pattern runs in three steps. You know, there's always the rule of three, isn't there?

[00:25:17] The order actually isn't negotiable, though. So first, the organizations realize that they need automation. And they sort of kind of say AI is the solution to that. And they say, let's roll some agents out. Let's make that happen. But then they realize that actually that automation needs to run on their own data. That's that differentiation thing we talked about earlier, the unique aspect rather than just using the models that are available to everyone. And so, yes, great. We've got automation and we've got it working against our own data.

[00:25:47] But the third one is the realization that it has to run against data that actually means something to the business. Now, step one gives you agents working in isolation. That's great. There is some automation there. Each reasoning of the open weights of a model, which everyone has access to. Clucination and scale. Nobody's aligned. If you can answer yes to all three of those patterns, then you've got a digital twin.

[00:26:15] That is an ontologically backed enterprise where the intelligence layer, not the application layer, holds the value. Stack does get adaptive. It's less rigid. One size fits all software. Most of the value does sit in how the knowledge is structured, how the systems connect and how decisions are actually governed behind the seats as well.

[00:26:39] And I think the assumption that's going to age worse is that the assumption that actually has aged worse is that AI lives in the model. It doesn't. It lives in the context you wrap around it. Everyone has access to the model. Almost no one has done the context work, owning your own intelligence and describing it in a manner that humans and agents can both reason against.

[00:27:06] And I think that is a thought-provoking moment to finish on today. And for everybody listening, Valiance is an AI native consultancy that is built to fix broken legacy consulting model, as well as where enterprises overspend on AI initiatives that rarely reach production. I'm hoping that we've hit a nerve with a few people today, set off a few light bulb moments. And for those people that are interested in maybe continuing this conversation with you, where should I point everyone listening to that?

[00:27:36] Well, that would be amazing. Thank you. Thank you, Neil. I think the best place is to get hold of us. Look at the Valiance website, valiance.ai. Look us up on LinkedIn, on Dom Selvon, and then there's the founders, Tarek Nasir, Anita Rajdev, and Rad Parvin. All our thinking, all our thought leadership on enterprise AI, on architecture, on adoption, on people first as well can be found in those two locations.

[00:28:02] If the practical side of making AI actually work inside a business is the conversation that you care about, that's the one that we're in. That's the one we're living and breathing every day. Come and argue with us. Come and have a conversation with us, and we'd be thrilled to have that. Excellent. Well, we covered so much today from the build versus buy inversion in this AI world that we find ourselves fixing that enterprise AI ROI problem. And I will include links to everything that you mentioned.

[00:28:32] I'll also include a link to your LinkedIn. Just make it easy for people to get hold of you. But more than anything, thank you for starting this conversation today and also talking about a language everyone can understand. So much value there. But thanks for joining me today. And Bridget, Neil. Thank you very much for having me. I loved Dom's relentless focus on outcomes. At a time where many AI discussions center around AI, around models, tools, and technical capabilities,

[00:28:57] he reportedly brought this conversation back to, hey, what problem are we trying to solve here? And yes, this does sound incredibly obvious, but it's surprising how often organizations skip that crucial step in their rush to just adopt new technologies.

[00:29:13] And perhaps the biggest takeaway is that AI success seems to have less to do with technology choices and much more to do with clarity, clear outcomes, clear measurements, clear governance, and a better understanding of where value is created in the business. Yep. Always comes back to value and improving business outcomes. But I'd love to hear your thoughts on this. Do you think AI is fundamentally changing the economics of software and consulting?

[00:29:39] Where do you believe the real sources of competitive advantage will come from in this AI world, especially when everyone's using the same models? So much food for thought here. As always, techtalksnetwork.com. Send me a message. Let me know. But I'm afraid, yeah, I know time goes quick when you're having fun and we're out of time already. So I'll be back again tomorrow with another guest. But thanks for listening as always. Bye for now.

[00:30:04] Bye for now.