Transcript 0:00 What's the difference between building your own IT application and buying one? Well, if you build it, you own the code, the bugs, and the emotional trauma. 0:08 But if you buy it, you own the licenses, the integration nightmares, and a dedicated Teams channel for just complaining about the vendor. [upbeat music] Computers have lots of memory, but no imagination. 0:19 Man is a slow, sloppy, and brilliant thinker. Machines, on the other hand, are fast, accurate, and stupid. AI will do what you tell it to do, but that may be very different from what you had in mind. 0:31 Technology presumes there's just one right way to do things, and there never is. Computers are useless. They can only give you answers. 0:40 For a list of all the ways technology has failed to improve our quality of life, please press three. Welcome to ProcureTech Unpacked. 0:48 I'm Joel Conneen Demers, and this is the show for procurement professionals who want to understand the technology shaping our function. 0:55 Today, we're gonna be discussing build versus buy in the era of AI and if, you know, the fundamental, uh, points behind the discussion have changed or not. But first, welcome back, Matt. How's it going? Hey, Joel. 1:07 Yeah, going good. Yeah, there's a lot going on at the moment, for sure. Um, I'm kind of coping at the moment with, um, another transition in life. 1:16 So my, my son that's at college is moving into his first proper apartment, if you like. So your role changes from being leader to advisor, I'm finding. 1:28 And then at the same time, my other son is, uh, is buying into a business or hoping to, and, and again, I'm, I'm becoming an advisor for that as well, and almost advising him on stuff I have no idea about. 1:40 But just age and experience coming in. So it's, uh... So he's moving out of the house? He's moving out of his college apartment into a grown-up apartment, if you like- Okay... which will be permanent home then. 1:53 So actually, yeah, he's moved out of our home now completely, so we've got a spare room. Y- yeah, I was gonna say, you've got room for your, for a trophy room. [laughs] That's right. Yeah, if only I had some trophies. 2:04 [laughs] I've got one, I've got one on my shelf over there from SIG. I could put that up. That would be good. [laughs] We could do that. Yeah, the whole room around the one, around... Maybe I can send you the cup. 2:13 You can, you can have the cup for a while, uh- Oh, that would be nice... in the meantime. Yeah, if we could just put Focal Point on it for a while, and we can, we can put that in the corner there. Yeah. 2:21 But, uh, yeah, at the same time, I'm, I'm recruiting, so actually there's an analogy there, I think. 2:26 But, uh, yeah, I'm, I'm recruiting a, a few different roles within the company, um, to, to up our team, which is really exciting. 2:36 Um, you know, trying to find the right person and, and make sure that they're doing the right thing is, is for sure interesting. I kind of focus very much on likability because, especially in the sales roles. 2:50 But what I really don't wanna do is just, just find somebody who can just read a script. I want somebody who can actually be a trusted advisor. Um, that's how, how I tend to work. 3:00 That's how, how my colleagues tend to work is, is being an advisor and, and helping people to come to the right, right transition, et cetera. So again, the advisor role comes in. What about you? What have you been up to? 3:14 Yeah, and I was just, I was just gonna add to what you were saying, like, I find that so right, right? And, and being able to, uh, do consultative selling to some re- degree, right? 3:22 So that, you know, you don't wanna sell just to sell, because eventually you wanna be able to deliver, and that's what makes the best long-term partnerships and, and clients, right? 3:31 But yeah, so on, on my end, well, actually, we had, we had talked about a couple things before getting on the show, but I have a story for you that happened quite recently that just popped into mind yesterday, which is... 3:40 So I have a, a, like a 90-year-old neighbor who, uh, lives on the other side, uh, here, and I can see him from, from the back of my house if he's outside in his yard, and he was doing yard work the other day, and then all of a sudden... 3:53 Like, I just glanced over, I saw him, I'm like, "Oh, he's outside." And then all of a sudden, he just keeled over and, like, boom, right in, into the [laughs] into the, uh, into the lawn. 4:02 And I'm going, "Oh, that's not good." And, uh, well, I guess he... Like, he'll get up, right? And so I'm waiting here, ooh, like 10 seconds, 15 seconds go. I probably should go check out what's happening. 4:13 So I go out there. Uh, I tell my wife, like, "Hey, be on standby," right? "We might have to call an ambulance or something." And, uh, and like, I get over there, I go, "Hey, are you okay?" 4:23 Like, you know, "You f- you fell." He's like, "Oh yeah, I just have trouble getting up when I fall these days." I'm like, "Okay, well," I'm like, "Let, let's get you up," right? 4:30 So he's not a small man, so I, like, grab him and, like, bring him up. Uh, and he had a, he has a cane as well, which was on the ground. So, like, he's up. I leave him up. 4:38 I go to grab the cane, and he falls again [laughs] as soon as I let him go. 4:44 And as he falls, his denture comes out of his mouth and, like, hits the ground, and his pal- his pants fall to his ankles and, and I'm going there, like... 4:53 I'm sitting there going like, "Oh no, [laughs] this is not good." [laughs] And then Benny Hill runs past, right? That's right. That's right. [laughs] Couldn't have made that up. And I couldn't have made this up. 5:00 So I go, "Okay," like I'm trying to put his pants up. Not working. So I en- end- ended up, like, bringing him up again, giving him his cane, like helping him with that. 5:09 He put his denture back in his mouth alone, thank God. Uh, [laughs] and then, and then he goes, "I've..." He's all flustered. He's like, "I think I'm gonna sit down." Uh, he's got a ben- there's a bench nearby. 5:20 I'm like, "Yep, that's a good idea. Let's sit you down." Sit him down on the bench. The bench explodes into a thousand pieces. He's down again. [laughs] No way. Oh, my God. Yeah. Yeah, yeah. 5:30 So eventually, he's like, "Hey, I'm just gonna sit here." And like I, I made sure he was able to get back home, uh, you know, safely afterwards and all that, but you could see... 5:39 And my kids were s- watching me from the window, right? So eventually, afterwards, it was having a, the whole discussion around, you know, aging and wanting to be independent and all this good stuff. 5:49 And so it's, it's, it's funny. It kind of ties back with what you've been going through around just, you know, life's, life's milestones, right? That's right. Yeah, and then, you know, you, you, you start to think, okay- 6:01 My first thought on that was, okay, what's the remedy for this is don't fall down- Yeah... if you can't get up when you fall down. But, but everything, it's just a comedy of errors there with the bench and everything. 6:12 It's, it's crazy. Oh, yeah. I mean, how do you- So I ended up like re-- I ma- I made this, I fixed this bench later, uh, later on, right? 'Cause it was just a bench of two-by-fours that had- You are so nice... 6:20 had rotten through. Uh, but yeah, otherwise I've been writing reports, uh, hard at work. You know, we, we, we referenced the, uh, conference lull, uh, last show. 6:29 But, uh, yeah, I've been writing stuff on MCP servers, which is coming out shortly, which I'm excited about. 6:34 Um, and, uh, the governance layer of AI and how, you know, with these LLMs hallucinating and being sta-statistical models, which we've covered, you know, there's a lot of, um, yeah, there's a lot of things that you need to master in order to get the outcomes that are promised by, you know- Yeah... 6:51 by AI in the mainstream, so to speak. Yeah, absolutely. Yeah. Well, there's a lot going on out there in the market, that's for sure, so... Yeah. So we should probably get into it, right? Yeah, let's talk about it. 7:03 So- Cool. Well, reminder- Yeah... to follow the show on your podcast platform. And, uh, if you're feeling particularly motivated, a five-star rating always helps us tre-tremendously. So let's get into the show. 7:14 [upbeat music] One of the stories I wanted to talk about was the Fable story and the US government. Okay, so you know, this... 7:26 It's a little while ago now, and, and Fable is actually available, and it's available to use free for a little while. 7:32 It's, it's probably finished by the time [chuckles] this gets released, but, uh, yeah, it's, it's out there. I... 7:38 You know, what really happened, though, in, in that whole thing, you know, the Anthropic's Fable 5 and Mythos 5 models, following a tip from AWS, were, were closed down. 7:48 Um, they were actually restricted by the US government. So it wasn't that the US government showed them, closed them down. They actually restricted them and said they shouldn't be used by foreign nationals, which- 8:00 That's hard to implement... how do you do that? Yeah. How do you do that? [chuckles] It's, it's really tough. It's like I'm a foreign national, but I'm American as well. Um, so where do I sit in that? 8:10 So because they couldn't verify that nationality in real time, um, it suspended it globally for everybody. So this had an immediate impact on, on many people. 8:22 The restriction was eventually lifted on June the 30th, and service was restored the next day. But my opinion on this is that I think everybody behaved fairly and rationally 8:35 within their own narrow remit, within their own silo. But that combined outcome, what actually happened was, was mildly or, yeah, possibly largely ridiculous, I think. 8:49 You know, the government identified what it believed to be a national security issue. Great. That's important for all of us. Anthropic had to comply. Okay. 8:59 Amazon had to comply with Anthropic, but customers around the world suddenly lost access to something that potentially they were using for critical business process. Mm. 9:10 That should really be the, the wake-up call for every enterprise, because this was early in the, in the generation. But if this happens later on and you've got a lot of your stuff using that... 9:21 You know, we talk about data sovereignty as, uh, it means choosing which country contains the server. 9:27 The incident shows that sovereignty also means understanding who has the legal and technical ability to switch off your service. Mm. 9:35 You can have your d-data stored in Britain or somewhere, um, but Washington [chuckles] can still make decisions that impact us all. 9:46 Um, I mean, Washington makes a lot of decisions that impact us all, especially at the moment. Um, so yeah, I don't know what you think about it. Yeah. Well, I think you're, you're spot on, right, in terms of, like... 9:59 Where my head goes is software as a service, like you've always been, you've always had kind of these two layers, right, of like you, you maybe you're just dealing with a service provider, a la focal point, right? 10:11 Uh, and I don't talk to the LLM providers in the back, but if I'm building... 10:15 If that software provider is building functionality that leverages stuff in the back, whether it's an LLM, whether it's, you know, we've seen, uh, web outages as well, right? 10:23 Like you're, you're using, um, uh, Cloudflare in the back or these other services, right? And they go out and, and they become a core piece of like global IT infrastructure, then everything else goes out, right? 10:36 And so, uh, it comes back to your point of, of really thinking around, you know, what are you putting into those, uh, into those applications and, and understanding the supply chain from an IT standpoint, right? 10:49 If we try to br- draw, draw in direct manufacturing principles into this, like what are the different hooks, uh, uh, downstream that might have an effect on me and the risks, uh, that are associated? 11:02 And then the other thing that it also, uh, brings to mind is, uh, there's also open weight and open source models, which I'll get into a little bit later on as well, but, uh, that you might want to have more control over the LLM you're using if you're going to be building mission-critical business processes on top of those, right? 11:19 So I think it's also a signal in that direction. Um, yeah. But those are kind of the, the, the things I, I hear, but I'd love to, like get... Do you, yeah, do you concur? Do you see any other key takeaways here? 11:31 I do concur. I th- I think my other concern, uh, that I, I, it's kind of just come to me is, you know, people are building their own AI agents. They're using, they're using maybe a company, 11:43 company contract with Fable or, or for Fable or whatever. Um, do you know where you're using these solutions within your, within your whole infrastructure? You know, you need to understand also what your, 12:01 what your disaster recovery is on these kinds of things. What is the operational fallback? Um, also understand, you know, what notice do these providers need to give? 12:11 'Cause, you know, there's hundreds of LLM providers out there. It's like SaaS providers, right? And, and some of them are gonna go bust. Some of them are not gonna make it. So be ready, be prepared. 12:22 Are your, are your agents portable? All those kinds of things. Yeah, this is fragile. I think you have to be ready and you have to have a plan, and you have to know where you're using these things. 12:33 So just allowing people to be Wild Wild West and do what the hell they want with, with AI agents and, and AI more generally is, uh, is dangerous, especially for larger companies. It's, it's so funny, right? 12:47 Because, like, every... A- as AI, LL- large language models mature and get integrated into the enterprise, like, it's more of the same, right? 12:55 'Cause these are the same type of questions or the same types of concerns that, you know, moving stuff to the cloud. Oh, what's in the cloud? Oh, yeah. What's, what's not in the cloud, you know? 13:02 It's, it's kind of the same types of questions that come up around enterprise resiliency, around mapping your business process, around r- knowing where you're using applications and, and different functionalities in the back end. 13:13 So it's, it's, it's funny to me, right? Like, we, [chuckles] we always come back to the same- It does... types of conversations. Yeah. Yeah, and I've, I've been working for 40 years, right? 13:20 I remember this with ICPs early on. So who's my internet content provider- Mm... 13:24 and who's, you know, in the, in the '90s and, and before that, there was a, a kind of a, a splurge of mobile phone operators that suddenly happened, and some of those went out of business. 13:36 And, and prior to that, there were other things going on. It, it's, it's just a constant change. Yeah. 13:43 And then you get the dot-com bust, and the, and the company you were dealing with is no, no longer there from one day to the next. Um- But it seems that at the very least that, like, you have winners, right? 13:52 You have s- more stability over time with, like, the underlying pieces of, uh, IT architecture. Yeah. Um, so I'm sure, I'm sure we'll get there with LLMs as well, right? 14:01 But I think it's, it's, uh, it's funny because this time around, I feel like it was, like, a magic pill moment, right? It was like, "Oh, it's gonna change absolutely everything." But I... 14:09 Yes, but, like, very slowly, and with the same constraints as everything else. [laughs] Yeah. Yeah, very much so. Yeah. Oh, cool. Same result, different problem. The, the second story I wanted to cover was the, um... 14:22 and it's related, but it's this new Chinese Kimi 3 model, uh, that's taking the world- Yeah... by storm. 14:28 So essentially a startup called Moonshot, uh, in China released a model, a new model like Fable, like we just discussed, on Friday, uh, July 17th, and it immediately shot up to the, to near the top of the AI capability le- leaderboard. 14:42 So there's stuff like, uh, Epoch AI and, and a number of others that have these tests for LLM to, to take, and then they can give you a score and essentially give you a ranking in terms of the other models. 14:56 Uh, and this new model outperforms e- essentially all rival, uh, models except Anthropic's Fable 5, whi- which we just discussed, and OpenAI's leading model as well, GPT 5.6, on, uh, overall capability, right? 15:10 So it might perform less, less, um, less high or, or better in other dimensions, but if we look at it overall on the benchmarks, that's where it sat, in third place. Uh, why is this important? 15:22 Well, you know, we had a DeepSeek moment, I think it was a, a year ago, uh, and we've got other Chinese models that are starting to del- deliver essentially frontier level performance, so the same types of performance as, uh, the big, the big US firms, but at a fraction of the cost, right? 15:37 So some- something like 10% of the cost of running, uh, the models, the US models. 15:43 And to me, the how is the most important part because, uh, I don't know if you've seen the graphic, Matt, around, like, AI being a five-layer cake, right? So i- there's, um... I think I have the graphic actually. 15:57 If you've got it, I'd love to see that. I haven't- Yeah. Let me-... I haven't seen it... let me pop that up here. All right, so here, here it is up on my screen. So AI is a five-layer cake. 16:05 I don't know where this graphic is from. I just found it on the, the internet. It's not mine. Uh, but essentially, there's, there's... yeah, there's these five layers, right? 16:12 So one is applications, which is the one that as procurement professionals we see, uh, uh, in our day to day, right? These are the, the focal points, the Aribas, the Zips, the Coupas, the et cetera. 16:23 Anything you're using as an application is the top of that cake. Then, as we just mentioned, right, you're using models in the background. This is the supply chain of your, of your IT software. 16:35 So Fable 5, uh, this new Kimi K3 model, uh, is that second layer of the cake. 16:40 Then there's the infrastructure, so the data centers and all the discussions going around, uh, about, "Hey, we need new data centers to be able to support everything the models needs to provide the value through the applications." 16:52 Uh, underlying of that is chips, so NVIDIA providing the chips to the data centers and or AMD and these other chip providers, and the whole, uh, you know, embargo on the high-performing chips from the US to China. 17:04 And then at the very bottom, you've got energy, right? 17:07 So how are you feeding these chips that rely on these data centers or that are in these data centers on which the models rely on to provide the value in the applications? 17:17 Uh, and I think the biggest things that are different, uh, in China is that they're deploying new energy and, and mostly renewables at a crazy pace, and they're also constrained on the, the performance of the chips they have access to. 17:30 And so they have a, a, a, an advantage on energy production, on energy costs, and they also need to create models for lower performing chips, so they need to be more efficient models than, than the US providers, right? 17:43 They use less energy anyway, possibly. Yeah, as well, right? Yeah. They-- and they, and they-- they'll perform, uh, more efficiently, right? So the algorithms- Yeah... 17:52 uh, how you build the model, you're gonna build it, try to build it with less lines of code, right, and the logic. Yeah. If I can, you know, simplify it to the, the most simple expression here of what that means. 18:03 Yeah, so, yeah, I think on the energy side especially, I was in Utah recently at, in Salt Lake City with a At a conference, interestingly, as always, uh, but it was a Green Cabbage conference. 18:14 And at, at that conference, I was walking through the town of Salt Lake City and somebody approached me and, and was campaigning against a data center was being built. 18:23 And I said to him, "Why are you campaigning against it? Don't you use AI?" Um, and, and he gave me some facts and figures. I said, "Did you find those on AI?" He said, yes. [laughs] Um, but, but more importantly, 18:36 um, he said, "Oh yeah, this, this data center's been put in, and it's, it's increasing the, the temperature around it, and it's using a lot of energy." 18:43 I actually got to the conference and there was somebody there from an en- energy company. 18:48 And I was talking to him about this and he said, "Oh yeah, that data center that's being built in Utah actually has its own power station being built with it alongside it because Utah cannot cope with the amount of energy that's being needed." 19:02 Um, so again, we're, we're adding fossil fuel infrastructure to, to power this new generation of AI. It's interesting to hear that China is using more renewable sources. Yeah. 19:16 They're doing the same thing, but they're building like a solar- Yeah... farm right next to the data center, right? And I, and I love that. I think that's, I think that's great. 19:25 Now, obviously I have some concerns around all of this and, and my biggest concern, as always, is data security. Yeah. Um, you know, if you start throwing your, your data into Kimi Ki- into Kimi, and 19:40 suddenly that data is, is no longer sitting in your data warehouse. Yeah. That data is actually sitting now in, in China. Do you have concerns about copyright? 19:48 Most people, at least in the Western world, have concerns about copyright, for sure. So that kind of comes to my mind immediately. Yeah, but I think it's, it's interesting to study the Chinese model in any case, right? 20:01 Even if you're not gonna use it in the sense where... Like another, another point that's interesting is, uh, that the mo- this model is an open weight model. So not to be confused with open source, right? 20:10 Open weight means that you can download and modify the parameters, uh, at, before calling the, the, the LLM, as opposed to closed models like Claude or ChatGPT, where you just have access to the API, you can't change the parameters. 20:24 So I, I liken it to like, it's like getting a DVD versus streaming on a platform. Like if I get the DVD, I could do all sorts of cool stuff with it, [chuckles] right? 20:32 I can rip, rip the video off the DVD and do all this stuff, whereas I can't do that with streaming. 20:38 And so it gives you a lot more flexibility in terms of what you can do with the model if you, if you start using it, right? 20:43 So there's all these like little layers of, uh, intricacies, details that, that end up mattering in terms of like the cost and the performance that you're getting at the application level, that I think the Chinese model is gonna just inform other providers, like maybe not these leading US ones that wanna capture the market, I think essentially, but the, um, you know, the, the 105 other ones that we detailed on a previous show, for example. 21:08 Yeah. No, and, uh, yeah, I think you're absolutely right, and people will learn and, and people will continue to, to develop, for sure. 21:15 I think what it, what it's saying to me is, you know, that's the best individual model maybe at the moment for this, this task. Tomorrow it won't be. You know? What's the half-life of, of these LLM models? 21:27 I think the half-life [chuckles] is getting shorter and shorter. Um, as you know, some have advantages, others have advantages. And like you said, Fable is the, the biggest, closest comparator to this, this solution. 21:41 Tomorrow it won't be. There'll be something else. Um, you know- And you, you might not need the highest performance model- No... to achieve the outcome you're looking for, right? 21:50 I think that's the other thing, like we get bogged down in like, "Who's the best?" Right? And it's kind of the same thing as like the Gartner- Yeah... Magic Quadrants, right? 21:56 But it might not matter, like what's the right fit for the task at, at hand, right? Well, define best. 22:02 You know, once we s- once we start talking about tokens and, and costs of, of models and everything else, I mean, that has to be taken into account as well. 22:11 It's like software, you know, you can, you can buy the best possible contract life cycle management solution out there, um, and it'll cost you $2 million a year and it'll do everything for you. 22:21 Do you want it to do everything for you? Your... If your legal team doesn't use the redlining function, if the, the rest of the company is not even touching it, they're just using DocuSign, why do you need it- Yeah... 22:33 at the end of the day? You're still paying for it. [laughs] Yeah. Yeah, and you... Yeah, you'll be on a contract for three years to pay for it [chuckles] probably as well. Yep. So, so- Yeah... 22:44 you know, if I, if I try to wrap up that story, to me the, the takeaway, or at least what I'm seeing is like LLMs are a new spend category, right? Mm-hmm. Especially if you're gonna be 22:53 having high usage of it in your business. Uh, and so, you know, w- what... It go- It ties back to you saying like, "Where do we use it? Why do we use it? 23:02 Uh, what's the should cost for these different LLM providers based on the scenarios we wanna use them for? How are tokens being charged for?" Uh, right? 23:10 And we defined tokens, I think, in the last show, in, uh, one of the last shows. Um, and just don't give this to like an IT category manager, right? 23:17 If you're seriously gonna make investments with IT around using LLM-supported agent and agentic, uh, processes in your business, I think you have to consider creating a new, a new category in your procurement team, uh, 'cause it's gonna become significant spend over time. 23:34 I, yeah, I agree. I think it's, it's almost becoming a utility type of category for me as well. 23:39 You can buy it from different places, exactly the same, same thing, and you need to be able to switch from one gas supplier to another gas supplier seamlessly and easily, which you can do 'cause gas is gas. Mm. 23:51 But, um, you need, you need to have portable models, portable agents that you can move between models, for sure, as, as time goes on. And, and that category's gonna be a difficult one to manage for sure. 24:03 I was just talking to a category manager for... He's actually direct materials and, and he said, "Oh, I'm struggling to find suppliers." 24:10 I don't think you'll struggle to find suppliers for AI very quickly, but you will struggle to filter them. 24:16 And understand which of those hundred and plus, um, suppliers I should be using and which should I be majoring on. 24:24 It's, uh, yeah, for procurement, it's a, it's a big thing, not just for procurement technology either, for procure-- for technology generally in your company. Hand in hand with, with IT, uh, absolutely. Yeah. 24:36 Uh, all right. I think that's, that's a good way to end Market Signal, so let's get into the deep dive. [upbeat music] All right. 24:46 Today's, uh, build versus buy in the age of AI is what we promised at the, the top of the show here. 24:51 So w- there's three big, uh, blocks of, uh, or at least three big takeaways, three big, uh, things we wanna leave you with on this topic on the show today. 24:59 Uh, AI has made building software so much easier, uh, but to me it doesn't make owning it any smarter. Uh, and I think we've been alluding to that in the Market Seg- Signal segment too. 25:10 So let's share a three-part framework for how to think about it. Number one, build versus buy is not a binary decision, even though there's only the two terms. To me, it's a continuum. 25:20 So where you land on that continuum determines how much risk, cost, maintenance burden you're ready to absorb versus you wanna, uh, put, put outside to the world. 25:30 And there's not one where you're gonna get zero risk, zero burden, zero cost, right? Uh, like you're-- there's no optimal decision. It's all a qu- a question of trade-offs. 25:39 Uh, and most organizations anchor the discussion in the extremes, but they miss all the good positions in the middle of that continuum. Uh, so if I could illustrate it with four different positions. 25:50 Let's take number one, like a full custom build, right? We're at the end of this continuum. Your team, uh, does all the, the work, fully codes the platform, uses open source LLMs. You use your infrastructure, right? 26:03 You've got a server in the basement [chuckles] of one of your offices. You own every bug, every security patch. Like, you need to keep on top of all this stuff with your IT team. 26:12 Uh, and so that's like the one end of the spectrum, which, you know, unless you're a, unless you're a software provider, [chuckles] I probably don't recommend. Uh, or unless you're wanting to do really unique stuff. 26:23 Then you move towards buying, so building on a generic platform, for example. So you've got all these Microsoft Copilot studios, n8n make platforms that are low code, no code. 26:35 It's still your team doing everything, but you can have access to frontier LLMs, uh, shared infrastructure between clients of these generic platforms. 26:44 And so it gives you a very low barrier to entry to start building stuff, but there's still a partial burden around needing to, to, to build the logic, right? 26:52 There's zero pr-procurement logic out of the box in a lot of these platforms. 26:57 Uh, but your IT will like these platforms because they say, "Hey, everybody can use the same, the same platform, and they've got all of the same capabilities," but they are lacking the domain specific knowledge, which I think is important. 27:08 That's true. Which leads me to the third one, which is buying a, a procurement platform, uh, with an AI workflow studio. 27:15 So I covered this recently on LinkedIn of like every big major provider on the Gartner, uh, Magic Quadrant for source to pay suites now provides an AI workflow studio for you to build your own agents and/or they give you agents out of the box. 27:30 So you have a software vendor that has domain specific knowledge, uh, and your team is more in configuration mode than in like creation mode and, and, um, and build mode. So the vendor provides the LLMs. 27:41 It's the vendor infrastructure. The vendor's maintaining the software. 27:45 You really have to be more on the business process and business analysis side of things, of thinking through, okay, how do we stitch together business processes? 27:52 How do we work with IT to ensure we have end to end processes across cross-functional applications and all this good stuff? 27:58 And I think this is probably where most mature procurement functions should be operating if they aren't already. 28:05 Uh, and, and then you have buying a procurement platform, but only with prepackaged agents, and that's probably, you know, the far end of, of buy. Is like, "Hey, we don't wanna build anything. 28:15 Like, we're just gonna-- we just want the vendor to spoon feed us, uh, uh, agents and, and agentic functionality fully built by the vendor out of the box, key in hand. I just pay the so- the software license." 28:26 It's probably the fastest time to value, but it's also the least amount of flexibility. 28:31 Uh, and so wherever you wanna put yourself on that continuum, uh, I think is, is a, it's an interesting way or a useful way to think about build versus buy. Any thoughts, Matt? 28:41 I think it depends on, on your company, right? Where you wanna be as well. 28:45 If you're a kind of a, I won't say cookie cutter, but I will say cookie cutter company, um, who does standard things is, you know, a smaller company maybe, you know, cost of, of technology is very important to you. 29:00 Um, yeah, coming in with that prepackaged, pre-agented, um, solution absolutely can make sense for you for sure. I think for me, 29:11 the biggest reason that the people are generally thinking about building is they wanna get away from that vendor dependency. Um, and I get that. I really do. 29:23 You know, I'm a, I'm a software guy and have been for, for many years. But honestly, it, it, it just shifts it from being a vendor dependency to another dependency. You're now dependent on people within your company. 29:34 As you, as you said, you've got the server in the basement or maybe you're contracting with Azure or AWS or some... or Google. 29:42 But basically, you, you've got your data, you've got your information, you've got your processes, you've got your solution that you've built. Why would you do that? 29:53 What is the, what is the importance to you as a company of actually building it yourself? Is it to avoid cost? Is it to simply avoid that vendor dependency so you have full control? 30:06 If you want that full control, what does that actually mean to you? Because that means now you need to have your own infrastructure in place. You need to have your own developers in place. 30:16 You have to have your own ownership in place. Um, so again, you're dependent. You get rid of that subscription invoice, I get that, but you have to think about how do I continue to develop this solution? 30:29 You know, typically software companies are spending 40% of their revenue on R&D. That's a typical spend for a SaaS company at the moment. It's, it's massive. 30:41 Um, if you're saving $200,000 a year, are you willing to put in 40% of that, $80,000 a year, to continue to develop your solution? But more importantly, is that enough as well? Yeah. 30:56 Because the company you would have bought it from has 100 customers, 200 customers, and they're using 40% of that revenue to develop their product going forward. So it's, you know, you are gonna fall behind. 31:09 It's like small software companies fall behind big software companies often, until the big software company gets too big that they can't help tripping over their own feet. But that's basically what happens. So, 31:23 you know, you can sit it in different places. I think the other thing for me is, like I said, what is the value to you as a company? For me, 31:32 it has to be a business differentiator therefore, because the economics of building a solution otherwise don't actually make sense unless you're a really, really bad negotiator. 31:42 So the economics don't make sense, so therefore it has to be an advantage to you commercially. Mm. 31:50 Now, I know one company in, in Europe that's built their own software solution for procurement, and it is a business differentiator. 31:57 They've actually built it in that it, it actually offers their customers a much better service because they're b- able to buy stuff more effectively, more efficiently. Um, that, that makes sense. 32:09 If you can do that, then fine. 32:11 But where I've seen build make sense over buy is very large retail companies, you know, the Walmarts, the Home Depots of this world, who have 3,000 people in their IT department building stuff, but they build the customer-facing stuff. 32:25 It's a differentiator. This is why we are different from Lowe's. This is why we are different from Target, and it, and it actually brings a value to the company in a different place. 32:37 So I think commercially you need to think about it. Tactically, you need to think about how you're gonna continue to manage it as well. 32:44 I, I couldn't agree more around the fact of like, does it create a competitive advantage for you in some way, shape, or form, right? 32:50 And I think the oth- the other point I'll add before we go to the second takeaway is it doesn't have to be the same choice for the whole organization, right? 32:57 A- and/or the same processes, 'cause you can sit on different parts of that continuum at the same time for different problems you're trying to solve. 33:05 So an MS Copilot or like a generic platform, cross-functional, may be the best fit for cross-functional workflows, right? 33:13 If it's supported by IT end-to-end and, and that's how you're gonna get the value from, from inte- the value of the processes and the integration between functions. 33:21 Whereas self-contained procurement workflows on a, on a procurement platform, something fit for purpose for procurement may help you get the results you actually need on a specific set of processes, uh, within procurement, right? 33:34 So I think that's, that's the other thing I'd call out in this discussion, but I think you're bang on, right? Get... The, the-- Yeah. 33:40 We often underestimate the cost, and we'll get to that in, in, um, of building in number three. 33:46 So number two, bui- buy to build to me is, is the new paradigm that we're essentially pointing to, which is leading teams aren't choosing between building and buying. 33:55 They're doing both in the right sequence and at the right layer. So they would buy the platform or the platforms, right? 34:02 So if you have a generic company-wide platform and like most of your, uh, your, your functions are using the company-wide platform, and then procurement for, in this specific example would be where we're creating differentiation, uh, if not by build, well, by a speci- a procurement specific-- having a procurement specific platforms with built-in domain expertise, with out-of-the-box processes and agents that we're able to, you know, reduce the time to value by, by implementing it. 34:29 Uh, and then we can copy and modify them to fit our context, right? 34:32 We may build cross-functional agents and, and workflows on the, the company-wide platform and procurement specific agents and workflows on the procurement specific, uh, platform and, and as much as possible reuse existing content, and then only build from scratch on the edge cases that generally require it, right? 34:51 And, and sometimes it'll be just stupid integration use cases like we have 18 ERPs and we need, we need to copy paste the same thing in eight-- across the board, right? 35:01 Or make sense of data across different architectural, uh, issues. 35:06 But that's, that's how I'm thinking about it is how can I get the best of both worlds and how can I, how can, how can I buy to build and not just buy to be constrained, uh, in a platform, which I think is, is how people think about buy a lot. 35:19 Yeah, I think so. And, you know, opening-- I have concerns about opening up platforms. I mean, most platforms have some APIs and, and they'll talk about open APIs. Um, but what is actually open? 35:33 What do you actually want to be open as well? Yeah. Um, you know, if you've got four elements of data, then fine, it's not as big an issue. 35:41 You can probably control it, you can, you can understand it and, and keep somebody accountable for it at the same time. 35:49 If you've got a huge amount of data, I mean, most of these procurement systems have a massive amount of data. Who do you want to expose that to? Who do you want to allow to build agents on top of that data as well? 36:02 Because the, the data is gonna escape. It's gonna get out into the wild and, and there's no putting the genie back in the bottle at that point as well. So, you know, your data is accessible. It needs to be portable. 36:15 Um, ex- you know, external systems and agents can use it with appropriate permi- permissions, though. So where is your permission police? Who's looking after that? You have to be... 36:28 Again, I, I said earlier in the show, Wild Wild West. I think that's where we are at the moment. It's kind of like the internet was in the '90s, um, that everybody can do everything. But be careful on your data. 36:41 You know, opening an AI, an API and, and allowing AIs to, to connect to it, it sounds fantastic, but if you lose control of your data, you lose control of your copyright, you lose control of your business advantages. Mm. 36:57 If your, if your supplier data gets out there to your competitors, how damaging is that to you? Think about that, 'cause it's probably pretty damaging. Uh, you know, indirect procurement may be less so. 37:12 You know, if, if, if your Staples data gets out there, it's not as big a deal, but it's still a deal. Yeah. It's still important, and it, and it opens up other risks as well. It's not just about the data. 37:23 So yeah, and then you need that auditable history of, well, of, of who made the decision when, and why did they make that decision? If you're letting people make their own AI agents 37:35 and you don't even know they've done it, that could be a problem. Yeah. The genie's- That's-... out of the bottle, [chuckles] so to speak. Yeah. Exactly. Exactly. 37:42 So that, that's my, my huge concerns, and I often come back to these security concerns because it's important to businesses. Yep. 37:50 You know, if you've got direct mate- you're an engineering company, you've got direct material information in your, in your data bank and it, and it escapes, sayonara. 38:03 But I, yeah, and I think the, you know, you're absolutely right. 38:06 Uh, but I think I wanna give that control, you know, at, to, to the procurement departments, to the IT teams of saying, "Hey, we have all of the APIs available. 38:16 You can get access to all the data in the backend," but we also have the permission layer, um, built into those APIs and potentially that MCP server. 38:25 And if that's a new term for you, [chuckles] we'll, we'll probably put out an episode on that. 38:29 But essentially, giving LLMs access to the backend data of my transactional systems, I wanna be able to do that when I want to do that, but I also wanna be able to block systematically, um, systematically fields and/or transactions and/or, you know, data objects, uh, depending on how I wanna architecture my system. 38:49 So I think it's, like, fully open with all the guardrails necessary to, for you to configure how open you, you make it, uh, from the default, right? Yeah. And predefined models that you're allowed to use as well, right? 39:03 So, uh, do we allow Kim? Do we allow all the different systems out there, ChatGPT, et cetera? Or, or do we restrict it to just the ones that are approved? And, um- Yeah, that we've vetted, uh- Yeah, that we've vetted... 39:15 like for a sec. Yep. Absolutely. Yeah, absolutely. Yeah. It's, it's- Yeah. Like I say, could be Wild Wild West. Let's get it under control. Let's get some marshals out there. 39:25 Uh, and then number three, whatever, wherever you build, don't forget the hidden costs. 39:30 So even if you're building on a software platf- an existing software platform that gives you prepackaged content and you're just modifying existing workflows and/or agentic capability, you know, the- there's something... 39:42 Y- there's gonna be costs associated to that, that you should be aware of, right? 39:46 So the, there's something romantic about building, I think you alluded to it earlier, where, you know, especially if you've been dealing with shady software vendors for years, you're tired of the antics, right? 39:56 Build, build, you go, "Oh, we could just do it ourselves." But building adds software development to your core business permanently, right? 40:04 Not for the project perma- indefinitely, and I think on that continuum that we've been discussing, like, there is an acceptable amount of build that you need to do to, to, to unlock value. 40:14 Um, but everything that you build, right, needs, uh, needs documentation, needs version management, needs release management, needs, needs a parent, so to speak. 40:26 Uh, and that's the same thing with just deploying an application without talking about the agentic layer. Um, and I think it, it remains true, right? 40:34 So it, it, it's gonna require your attention permanently, and that's redirected away from other things, and the biggest, the biggest, um, the biggest thing, the biggest thing that's at a premium in business, at least in my opinion, is focus, right? 40:47 Focus and, and ability to concentrate on a g- on a smaller set of ob- of objectives, and everything that you build, that needs to be in that, in, in those, uh, enabling those objectives. 40:59 And so, you know, yes, it's different now because you can use AI to, to generate code very quickly, sure, but AI can also write blog posts very quickly, and we've all seen the LLM slop across the internet together, right? 41:12 And so the, the same thing applies to code. Like, it's, it's much easier... Like, it's just the lowest common denominator has raised up a little bit, but it, but you still... 41:20 Like, it's very easy to generate average code, right? Or an average blog post, and that might not be s- what you need, um, to, to get to the outcomes you're looking for. 41:31 Uh, and another thing is that the talent rea- rea- reality which you alluded to earlier in the show, Matt, right? You're not competing with other procurement teams for this talent. 41:39 You're competing with every procure tech startup that's looking for AI and, and procurement function- or, uh, experience. 41:49 And if you're looking for pure AI, uh, expertise, then you're competing with the Googles and the Metas [chuckles] and the OpenAIs of the world. Uh, and so, you know, 41:59 I think, uh, there was ManpowerGroup's 2026 Talent Shortage Survey. 42:04 Uh, they had 39,000 employers respond to it across 41 countries, and they found for the first time that AI skills are the hardest skill for the world, uh, in the world for employers to find in employees, and it's overtaking traditional engineering and IT with 72% of those 39,000 reporting difficulty filling roles that had any sort of AI capabilities mentioned in the job description. 42:28 So to, to me, these are all the things that, uh, you need to consider when you're gonna start building stuff and then wanting to maintain it over time. Uh, and, and yeah. So I'd love to hear your thoughts, Matt. 42:39 Yeah, and I th- and I think the AI... Yeah, yeah, we need AI skills, absolutely, but the AI skills without Business skills behind them or business knowledge behind them are, are useless. 42:50 You're just gonna end up, like you say, with slop. Um, even if you've got a person writing the code or, or writing the prompts or, or whatever, um, without having the right guidance from business people. 43:04 And to have an AI per- person, somebody who knows how to do AI very, very well and knows business, I mean, that's the new business unicorn, right? Yep. So if you've got one... 43:15 Anybody out there that's got one of those, call me or just keep really, really quiet about it, 'cause keep them in a locked room, 'cause they are gold dust. 43:23 They're the people you need to keep in your business at the moment especially. And that's how I define a business analyst in, in 2026, right? 43:30 Because, like, I've always te- tried to pride myself on being at that intersection of business or procurement, right, and IT, ERP systems, such to pay suites, uh, best-of-breed applications, and now LLMs and AI, right? 43:44 And how do we bring those two together in a specific context to, to get to the objectives that, that we're looking to achieve. Mm-hmm. So, but I agree with you. 43:52 It's very rare 'cause I have to get calls like, "Hey, do you know somebody else like you?" Uh, not really. [laughs] Very true. Not often. There's a couple, but, uh, they don't wanna be bothered, right? 44:04 Uh, so it's pretty- [laughs] They can name their price. Yeah. Away you go. Yeah. That's good. That's right. Cool. So, so three questions just to wrap it all up here. One, have you counted the people? 44:13 So, so have you looked at the entire cost, right? The developers forever, and your management team's attention, uh, dedicated to whatever you build, but maintaining it over time, right? Building is the easy part. 44:26 Maintaining is the hard part. 44:27 Uh, what happens the day who, uh, the, the people who understand what was built, uh, and probably haven't documented it to, you know, a very high standard or w- the standard that should be used, um, what happens when they resign? 44:40 What happens when they leave, right? And what are your fallbacks? How are you managing that risk? Uh, because when you buy a software subscription, like, you're delegating that risk, right? 44:49 Uh, and then number three, are we a procurement team or are we a software company that happens to do procurement, right? If you... 44:56 For the, for the overwhelming majority, I think it's the former, and then it's just a matter of, as you say, Matt, like finding people that can understand, uh, the intersection of business and tech or procurement and tech a- and, and shape it in our context to get to the objectives, uh, while considering the TCO. 45:12 Mm-hmm. Uh, so I'd leave you with those three questions. I don't know if you, you, you'd leave, uh, the audience with another one. No, I, I think that's absolutely right. 45:20 I think the, you know, especially that last one, you know, are we a procurement team or a technology team or a software company, I, I think it's really, really important. 45:29 I've actually seen companies who have, have built a software solution and then taken it to market and they become a software company s- primary and a, and a procurement organization secondary. So, 45:42 uh, but that's not typical, and that's probably not where you wanna be. You know, you're a procurement company. 45:47 You're there to, to make sure that the company is sustainable, that the, you're saving as much money as possible, that you're, um, defining risk as you go through, et cetera, uh, dealing with the right people. 46:00 Um, you're not there to build software. But if it's a business advantage, then- Yeah... think about it. It's gotta be a business advantage. That's my, that's my critical takeaway there. I love that. 46:12 It shouldn't just be to avoid cost, because you won't. You will not- Yeah... avoid cost by building your own in any way. 46:19 So just to, to leave you with a closing illustration here, and I don't think we're gonna have time to play our game today, Matt. We've ran long on this build versus buy, but we'll keep it for next time. 46:28 Uh, but to g- to give you a, a little analogy here, right? Think about, you know, that dream of building a cabin in the woods. 46:34 I may be the only person that has this dream, but like, you know, remote cabin in the woods, uh, is, is exactly the same thing, right? 46:41 It's like, it's like buying s- or, or building that cabin in the woods is like building software. So I go out there, you know, I take two weeks vacation. I, I, I build it. It's super impressive at first. 46:52 Friend comes, uh, friend comes by, he marvels at it. Uh, I built it with my own two hands. I sacrificed my whole summer, and now I can finally enjoy it, right? 47:01 But then the roof starts leaking, and the foundation shifts, and the termi- termites find the beams, and the plumbing freezes. 47:08 And now every weekend I'm driving up there is to patch problems, and I could have just rented a place, right, with condo fees for the same amount of spend, if not less, and do the things that, you know, I had actually envisioned in that dream of the cabin instead of fixing the cabin the entire time, right? 47:24 So if you're a fixer-upper and that brings you joy, that's fine too. 47:29 Uh, but I think that that's the, the, the image I wanna leave you with is that everything is in, in a constant sta- state of decay, and just because you can build something doesn't mean you should. 47:39 Uh, so buy the right foundations, configure them, extend them where they need to. 47:44 You're gonna do a bit of build somewhere along the continuum, and make sure you're putting your energy where it counts, which is your strategy, your suppliers, your people. Matt, any parting words? 47:55 No, I, I, I completely agree with that analogy. Uh, don't you have a cabin in the woods? Didn't we see you cutting your path through to the cabin in the woods at some point? But, um, no, I, I completely agree. 48:04 I, I think the, the other analogy you could use is, you know, y- you pay for the man, not the hammer. Um, you know, building your own hammer doesn't mean that you've, you've built a, a sustainable, ongoing solution. 48:17 Um, because, you know, you, you want to make sure that it's gonna continue to develop and do the job you need it to do as you, as you move forward. 48:26 So that means not necessarily choosing the, the solution that's already in your stack as well, because just because you already have a solution that you've had for 10 years doesn't mean it's right for you. 48:38 It doesn't mean it's the right place to build your agents. It's, um... Also, don't think about, you know, IT should have an impact on, on decision-making. But it should not be the final decision maker, I think. 48:52 The business should always be the final decision maker on what solution to, to use and to buy, because it's a business requirement. It's not an IT requirement. 49:03 So just buying a generic, um, that you can build a procurement stack around isn't, isn't the right one. You're always looking at service with a software as your solution. I think that's the, the critical thing. 49:15 So make good decisions for your business. I love it. I love it. And so as we wrap up here, uh, what are you, what are you up to in the next couple, uh, weeks here before we, we meet again? 49:27 I think in the next couple of weeks, you know, I'm gonna be... You know, I'm recruiting at the moment, so I'm gonna be onboarding the team, um, as we speak. 49:35 Um, but what I'm really excited about is actually vacation time. Um, so we've got, got vacation coming up. Pascal and I are gonna be traveling off to Switzerland with some very dear friends. 49:45 We're meeting with family there first, showing them around where we used to live, and then we, we end up in Milan, and the last, the last couple of days will be in Milan, and that's where I proposed to Pascal 25 years ago. 49:58 So- Oh... I'm excited to do that. To repropose. [laughs] To repropose. Continue the proposal. Yeah. Are you- Or finish the proposal. Yeah. Are you, are you, are you, uh, are you extending... 50:10 Are you renewing this contract? I am renewing. [laughs] Uh, yeah, and I'm making it very, very, uh, very long extension as well. It's- I like it... this one works. This was a, a good partnership. Very cool. Very cool. 50:23 Uh- And how about you? What are you up to? Yeah, I'm still, um, still writing, writing a, a couple things. 50:29 So I'm writing a deep dive on the history of, of catalogs, uh, supplier catalogs, believe it or not, and what it means for procurement teams and how they should think about catalog management and catalog programs for the future. 50:41 Also exploring the depths of, of agentic governance still, uh, you know, on source to pay platforms and, and what it means for budgeting for tokens, budgeting for AI usage in the future, and how we can give procurement the controls to be able to do that. 50:54 Uh, and I have a couple, yeah, a couple little camping trips as well, so that should be a lot of fun. But, uh, I'm, I'm kinda jealous of your, yeah, your Switzerland trip. 51:02 I'm, I'm, uh, curious to know how that went la- next time we chat. Yeah, let's talk about it. I'm excited to hear about the camping as well. That sounds, sounds fun. Maybe you'll find somewhere to put your cabin. 51:12 Yeah, maybe. [laughs] So everybody, thanks for tuning in again this time around, and, uh, we'll see you next time. Thank you, everybody. That's a wrap on this episode of ProcureTech Unpacked. 51:25 If this one resonated, subscribe wherever you get your podcasts and sign up for the ProcureTech newsletter for weekly insights between episodes. 51:33 If something we covered sparked a question or an idea, we'd love to hear from you. All the links for the reports we discussed are in the show notes. We'll see you next time. [outro music] Resonate.
