Pierre Mitchel, ProcureTech Analyst |- Will AI Replace Procurement Teams? The Future of Spend Analytics and Automation: täielik transkriptsioon

yet, you know, 25 years later, companies still are really bad at having good, detailed, you know, accurate, relatively timely, you know, spend visibility that they can take action on and and and drive some value. 00:02:19.439 --> 00:02:20.719 And why do you think that is? 00:02:20.800 --> 00:02:22.159 And that's that's the question, right? 00:02:22.240 --> 00:02:30.560 Like, if it seems like it's an important, like fundamental building block, you know, why is that not in like the must-have tool set of every CPO? 00:02:30.719 --> 00:02:33.039 And why is it not more pervasive today? 00:02:33.280 --> 00:02:34.400 Yeah, that's a good question. 00:02:34.479 --> 00:02:42.159 I think one is there's always this we ROI everything, you know, it is like the ROI death march on any investment on anything. 00:02:42.240 --> 00:02:49.680 And so, like, when you have an area that's obviously strategic and foundational, but yet it's like, well, what's the ROI on the spend analytics? 00:02:49.840 --> 00:02:57.599 Well, you can be like, technically it's zero because it doesn't necessarily create the savings, but it's like it's the treasure finder, right? 00:02:57.680 --> 00:03:00.800 We're giving you this thing that's actually just telling you where the treasure is. 00:03:00.960 --> 00:03:01.919 That's pretty valuable. 00:03:02.000 --> 00:03:04.000 That's a pretty massive option value, right? 00:03:04.080 --> 00:03:06.479 But sometimes people do get bogged down in that. 00:03:06.560 --> 00:03:08.879 Sometimes it's an IT, we'll derail you. 00:03:09.039 --> 00:03:12.000 It's like, well, we have this giant data warehouse, isn't it great? 00:03:12.080 --> 00:03:17.280 We're already working on it, but they're so you know, it's so slow, not getting what you know, the insights that you need. 00:03:17.439 --> 00:03:21.520 So I think there's a few factors around it, as well as just companies are really dynamic. 00:03:21.599 --> 00:03:24.879 They're buying other companies, we're adding more data into the portfolio. 00:03:25.039 --> 00:03:39.039 They have varying spend taxonomies, and then you can just bog down into this whole, well, we need the mother of all spend taxonomies, and we got to clean all in, you know, you get all these vested interests and legacy data, and you just start, you know, things grind to a halt. 00:03:40.080 --> 00:03:40.800 Long, long answers. 00:03:40.960 --> 00:03:50.080 I think it's probably ROI, just kind of the entropy around all that, you know, and kind of just legacy systems and mindsets and whatever. 00:03:50.159 --> 00:03:52.319 And you just need to have it like running water, right? 00:03:52.400 --> 00:03:54.159 And I'm like, what's the ROI of running water? 00:03:54.400 --> 00:03:59.840 Pretty high, unless you're living in the uh in the in the movie Idiocracy where it's all Brondo. 00:04:00.080 --> 00:04:01.280 Do you know this uh reference? 00:04:01.439 --> 00:04:02.639 It's what plants crave. 00:04:02.879 --> 00:04:07.280 The ROI of water was actually really high in that movie because they couldn't, they couldn't grow plants without it. 00:04:07.439 --> 00:04:11.680 So I'm kind of I'm kind of curious, like, what do you think is like the hardest part about spend analytics? 00:04:12.000 --> 00:04:17.920 And what portion of companies that you meet with do you think actually have some kind of spend visibility capability? 00:04:18.160 --> 00:04:24.480 What you know, it doesn't have to be a structured tool, it could just be a Power BI dashboard, but I think that's a compound question. 00:04:24.560 --> 00:04:26.079 But what's the hard the secret sauce? 00:04:26.160 --> 00:04:27.199 Is it the taxonomy? 00:04:27.279 --> 00:04:30.480 Is it the integrations or is it the actual cutting of the data? 00:04:30.639 --> 00:04:34.160 And then the second is what portion of companies you think actually have something in place. 00:04:34.399 --> 00:04:36.319 Yeah, well, I mean, I think you hit on it. 00:04:36.480 --> 00:04:42.160 So one is kind of just the level of detail, you know, and kind of and also then what is it being used for? 00:04:42.399 --> 00:04:47.600 So if you just bring some basic drive-by sourcing events, hey, we've never really done sourcing very well. 00:04:47.839 --> 00:04:56.800 80-20, got to get a stand-up on basic spend queue, bring in some consultants, you know, just to kind of build those wave charts, like, you know, where are we gonna go after the money? 00:04:56.959 --> 00:04:57.360 Great. 00:04:57.600 --> 00:05:02.399 You can get pretty far just kind of doing some of, you know, just just getting there. 00:05:02.560 --> 00:05:06.560 But as soon as you start like expanding the aperture, just needing to get a little deeper, right? 00:05:06.639 --> 00:05:21.279 As we get into spend planning, or as we get into on the direct side to look at opportunities for as soon as you start like sizing prizes that are a little more complicated and require higher fidelity, more accurate data, things like that. 00:05:21.439 --> 00:05:32.399 You know, now the also the analytics are getting more complicated and the kind of OIT can kind of do it and take, you know, the AP history and files and kind of bring it up into the data warehouse. 00:05:32.560 --> 00:05:33.759 You start transcending that. 00:05:33.920 --> 00:05:35.680 Now you're starting to get and do great. 00:05:35.839 --> 00:05:37.199 How do we model costs? 00:05:37.360 --> 00:05:39.360 You know, how do we actually start persisting that? 00:05:39.439 --> 00:05:41.040 How do we track time series data? 00:05:41.199 --> 00:05:43.360 How do we start getting into forecasting? 00:05:43.439 --> 00:05:54.879 How do we start tying in all these rich set of data that we need to kind of see these more sophisticated spend and supply analytics for risk, for you know, supply chain, for all these different for you know design supports. 00:05:55.040 --> 00:05:57.120 And now you start getting into a different ballgame, right? 00:05:57.199 --> 00:06:01.439

It's not just kind of the easy stuff, and now the degree of difficulty starts going up. 00:06:01.519 --> 00:06:09.519 And that can be a filter when you're just constantly and just trying to do the easy stuff where you can and just react, use the rest of your time reacting to other stuff, you know. 00:06:09.600 --> 00:06:11.360 So no, it makes a lot of sense. 00:06:11.519 --> 00:06:20.240 I guess a question, something that we talk about a lot whenever we catch up on the phone is the topic of AI because it is it is really in the air and it's everywhere. 00:06:20.319 --> 00:06:21.360 Everyone's talking about it. 00:06:21.439 --> 00:06:26.000 It feels like every vendor is putting AI and agentic AI in their headlines. 00:06:26.240 --> 00:06:30.000 Every company we talk to asks us even about our AI roadmap. 00:06:30.160 --> 00:06:38.079 You know, I want to get just at a high level like what's your pulse on what's happening right now in AI and agentic AI, good, bad, ugly. 00:06:38.319 --> 00:06:46.319 And then I would like to talk about next is what are the areas of procurement tech do you think will be most and least affected uh by the changes that are happening in AI? 00:06:46.399 --> 00:06:50.240 Some things will be probably untouched and some will be completely automated. 00:06:50.399 --> 00:06:54.399 So I'd I'd love to get your take on, you know, let's start with what's going on in the space right now. 00:06:54.480 --> 00:06:55.199 What are you seeing? 00:06:55.439 --> 00:07:04.399 Yeah, well, I mean, unless you've been living under a rock, you know, obviously, you'll think 99% of people understand of, you know, kind of that this is a pretty big deal. 00:07:04.560 --> 00:07:07.600 Certainly there's going to be people who are like, oh, this is another hype cycle. 00:07:07.759 --> 00:07:11.439 And there is, trust me, the hype is obviously in the stratosphere, right? 00:07:11.519 --> 00:07:21.759 But um, this isn't like blockchain, you know, where you're like, oh, it's all about blockchain, and you're looking at it and you're like, it's a freaking transaction registry, you know, that's it's a great one. 00:07:21.920 --> 00:07:25.600 And you know, but uh, you know, it's how's this game changing? 00:07:25.759 --> 00:07:33.600 And yeah, I think AI is fundamentally a sea change, especially now with LLMs and agentic kind of coming into the theme. 00:07:33.680 --> 00:07:37.199 But to me, it's not to me like AI is not the center. 00:07:37.439 --> 00:08:01.040 I mean, AIA is like the energizer bunny that is everywhere, it is fundamentally changing everything about enterprise architectures and digital from the data modeling to process management and orchestration to application development to you know intelligence to right, all of these elements are getting impacted by AI. 00:08:01.279 --> 00:08:10.160 And obviously, with LLMs and just really it's gotten the hold of everybody as to demonstrating more of what it can do in certain use cases. 00:08:10.399 --> 00:08:17.519 But do you want to take those that LLM brain and like drive it into all these little agents to try to do that for all different processes that are needed? 00:08:17.759 --> 00:08:23.040 It's massive and it's important, you know, um, but it's a big topic and there's a lot in there, you know. 00:08:23.360 --> 00:08:23.839 It's true. 00:08:23.920 --> 00:08:29.600 I mean, it's and anyone who says that they know the all the answers, it doesn't because it changes every three months. 00:08:29.759 --> 00:08:37.039 I'm kind of curious, like, I feel that there is a a bit of a spectrum between AI optimism and AI pessimism. 00:08:37.200 --> 00:08:44.399 Like, there's some folks where they are like 10 out of 10, this is you know, this is the savior of all humanity and it can do everything. 00:08:44.480 --> 00:08:49.360 Like you think about like the super intelligence advocates like Sam Altman and CEO of OpenAI. 00:08:49.600 --> 00:08:51.840 On the other side, you have many people that are really skeptical. 00:08:51.919 --> 00:08:54.080 They say, Oh, it's overhyped, it's overblown. 00:08:54.240 --> 00:08:57.440 I I'm kind of curious, like, where do you fall in the spectrum? 00:08:57.600 --> 00:09:00.480 One being like, hey, I I don't think this is as useful as people think. 00:09:00.639 --> 00:09:03.679 Ten being like, I think this is like the silver bullet. 00:09:03.840 --> 00:09:14.559 And was there a moment, because I know there was for me, where I went from being a little bit more skeptical to like, holy shoot, like this is this is much more real than I would have thought. 00:09:14.639 --> 00:09:16.000 It's not a toy anymore, you know? 00:09:16.159 --> 00:09:17.679 So I'm kind of curious about that. 00:09:17.919 --> 00:09:18.240 Yeah. 00:09:18.399 --> 00:09:27.360 Well, I think those people that come from uh different camps quite often, not to go full Yoda, but you know, they have not unlearned what they have learned, right? 00:09:27.519 --> 00:09:34.320 And this is a like procurement executives, you know, and they're never gonna tell you, like, hey, I've turned that crank, you know, in 10 different transformations. 00:09:34.399 --> 00:09:35.360 Why do I need something? 00:09:35.600 --> 00:09:36.879 Yeah, and you might not be able to. 00:09:36.960 --> 00:09:48.159 It doesn't mean you can't turn the crank a lot better and faster and deeper in this, you know, kind of I think anybody that's stuck on both sides of those camps are not really, they're just not seeing the value that they need to open their eyes. 00:09:48.240 --> 00:09:55.200 And so for me, it was always innately curious and I'm and I'm drawn to the new technology and really trying to understand what it is and how it's useful. 00:09:55.360 --> 00:10:01.039 And so for me, for around LLMs and AI, to me, there's kind of this term that's around like neural symbolic. 00:10:01.120 --> 00:10:02.399 So it's two worlds, right? 00:10:02.720 --> 00:10:05.120 Our brains are oriented as neural

nets, right? 00:10:05.200 --> 00:10:15.440 And as AI has trended towards, you know, we move from to neural nets and deep learning and then to, you know, LLMs and to kind of building on top of that. 00:10:15.679 --> 00:10:23.600 That world is great, but it is it is probabilistic, it is neural, it is not symbolic, it is not a structured reasoning. 00:10:23.679 --> 00:10:31.919 And over on the deterministic side, right, where SaaS apps live, structured data, clear domains, clear algorithms, rules. 00:10:32.080 --> 00:10:33.200 That's kind of been in that world. 00:10:33.279 --> 00:10:41.679 And it's always like, oh, you know, rules are terrible, you know, and on the other side, it's like, well, AI is completely, it's gonna hallucinate everything. 00:10:41.840 --> 00:10:44.320 And the key is making those two things fit together. 00:10:44.399 --> 00:10:51.200 I did a whole presentation at DPW that was all about bringing these worlds together to say, like, yeah, AI doesn't replace SAS. 00:10:51.360 --> 00:10:52.799 AI is SaaS, right? 00:10:52.960 --> 00:11:09.759 And it's like ultimately it demands more from your enterprise architectures, and it is in turn driving a whole new class of enterprise platforms and applications, you know, whether it's around MCP or whether it's you know, all these kind of different areas, those two things working together. 00:11:09.840 --> 00:11:16.639 Because if you're like Cell and Agentic systems, which are basically lampreys on the shark, you know, it's like, hey, this is sharks are dead. 00:11:16.799 --> 00:11:19.120 Well, if you know, you need a shark for you to live. 00:11:19.200 --> 00:11:28.799 You need a data, you need NASA apps, you know, but don't kill the host that you actually live off of unless you yourself are going to build a data model underneath to persist. 00:11:29.200 --> 00:11:37.919 Like, yeah, you can build great self-driving software, but you still need to have a chassis to run it in, or otherwise it's not it's just a concept, you know. 00:11:38.159 --> 00:11:41.360 Agents are a concept until they're taking action on data. 00:11:41.600 --> 00:11:42.399 It's interesting. 00:11:42.639 --> 00:11:44.080 Well, here's a question then. 00:11:44.320 --> 00:11:48.000 You know, so you're you're laying out a uh a perspective that I also share, right? 00:11:48.240 --> 00:11:54.879 A sort of spectrum between uh deterministic systems and pun not intended higher agency systems. 00:11:55.039 --> 00:12:00.159 And I think that on that spectrum you find a lot of different modules or areas of procurement tech. 00:12:00.320 --> 00:12:01.279 You know, what would you say? 00:12:01.360 --> 00:12:08.240 And this is the question I was asking too earlier, what would you say is the area that you think will be least affected by AI? 00:12:08.399 --> 00:12:11.679 And what is the area that you think will be most affected by AI? 00:12:12.000 --> 00:12:25.679 I think anything that involves, so in that previous example, I was basically talking around the world of language models and the world of structured data models, and in the middle is knowledge models, neural symbolic, bringing these two worlds together. 00:12:25.840 --> 00:12:31.759 So anytime there is complexity and scale, complexity when we run language, right? 00:12:31.919 --> 00:12:37.519 That is an area, and so that complexity can be on user interface, on spoken language and chatbot. 00:12:37.679 --> 00:12:38.559 Great, you know, that's fine. 00:12:38.639 --> 00:12:43.039 We improve the user interface to get you know the gate out of our mind into the computer. 00:12:43.200 --> 00:12:48.960 But the other is just the existing data that's out there, and that could be external complex market intelligence that you're bringing in. 00:12:49.039 --> 00:12:50.799 But the perfect one obviously is contracts, right? 00:12:50.879 --> 00:12:56.399 And that's the area where it's had the most value because contracts are written in a language called legalese. 00:12:56.639 --> 00:12:58.480 It's actually multiple languages, right? 00:12:58.639 --> 00:13:02.480 So it's legal ease, but it's German legalese versus whatever. 00:13:02.559 --> 00:13:13.440 So you have all these domains, and you got to be able to sift out those attributes and the metadata to be able to like understand it and then be uh so that's the area where it adds the most value and it will have the biggest impact. 00:13:13.600 --> 00:13:20.080 And you'll probably see that, you know, with in legal services, within obviously already in translation services, right? 00:13:20.159 --> 00:13:20.960 And things like that. 00:13:21.120 --> 00:13:22.559 So it'll work its way through. 00:13:22.639 --> 00:13:40.559 So that's where it's had the most value and impact, and it will continue to disrupt areas, but it also is just gonna create better deterministic tooling to just do better automation and RPA and all that stuff that that was on the deterministic side that just never got implemented because it was too unwieldy. 00:13:40.799 --> 00:13:43.200 This created more value in driving that. 00:13:43.279 --> 00:13:47.279 So obviously, the transactional stuff is just gonna get more automated away. 00:13:47.440 --> 00:13:52.000 And anything also that's really complex has the potential to be impacted. 00:13:52.320 --> 00:13:53.039 I agree with you. 00:13:53.120 --> 00:13:56.480 I think that CLMs are written in all the legal documents are legalese. 00:13:56.960 --> 00:13:59.919 You could also argue that they're written in the language of billable hours. 00:14:00.159 --> 00:14:03.039 So that's how you get very frothy, long documents. 00:14:03.360 --> 00:14:04.559 That's the value prop, right? 00:14:04.639 --> 00:14:07.679 You don't need as many corporate, you know, you don't need legal involved as much. 00:14:07.840 --> 00:14:18.240 They're also they're often the ones that are kicking and screaming, honestly, in a lot of these transformations, because they don't they don't want to

be nobody wants to be, nobody wants to be left behind, nobody wants to be obsoleted. 00:14:18.399 --> 00:14:19.679 You know, it is interesting. 00:14:19.759 --> 00:14:22.000 Like, you know, I used to think so. 00:14:22.080 --> 00:14:31.279 My sister started law school on Monday here in San Francisco, and uh it's the first time I've had like someone from my nuclear family out in California. 00:14:31.360 --> 00:14:37.120 So I'm very excited she'll be here for at least three years, assuming she uh continues enjoying her legal training. 00:14:37.360 --> 00:14:38.240 But it is interesting. 00:14:38.320 --> 00:14:45.360 Like I know a lot of people who are lawyers, and I think, oh, you know, with Harvey and all these legal tech companies coming out, is there a risk? 00:14:45.519 --> 00:14:46.639 Do you feel risk at your job? 00:14:46.720 --> 00:14:53.679 And they're like, no, we're so busy and we're so underwater all the time that I would love to have a way to offload some of our work. 00:14:53.840 --> 00:15:04.480 And so they're not, I think today feeling like the, you know, the crunch of, you know, hey, it's it's replacing our work as much as the uh the opportunity to to do stuff a little bit better and faster. 00:15:04.639 --> 00:15:19.759 You know, I will say in our business, we've used, we just use Chat GPT off the straight off the internet with uh obviously in our enterprise environment to review basic contracts like NDAs, because 90% of the time it's pretty vanilla NDAs. 00:15:19.840 --> 00:15:23.519 And so I just want the system to tell me, hey, are there any major red flags? 00:15:23.679 --> 00:15:32.559 Now we do have a commercial council who will sometimes send stuff to, but a lot of times if it's really clear that it's fine, then I say, okay, let's just go and keep moving, right? 00:15:32.720 --> 00:15:34.240 So I think legal makes sense. 00:15:34.399 --> 00:15:38.320 What are the areas that you feel like are gonna be like unchanged by AI? 00:15:38.399 --> 00:15:41.919 Because that people often ask the question, what's changing in the next five to 10 years? 00:15:42.000 --> 00:15:46.720 But they very rarely ask the question, you know, what's not gonna change the next five to ten years? 00:15:47.039 --> 00:15:55.440 Well, I think IT operations, I mean, obviously, you know, plumbers and anybody that's you know physically gonna, you know, on site your operating, you know, they'll be fine. 00:15:55.600 --> 00:16:07.279 I do think just on IT infrastructure, they're gonna be plenty busy just moving from you know legacy tech debt to moving to more modern architectures, right? 00:16:07.360 --> 00:16:19.200 Going from old, you know, integrations to modern microservices and APIs now to composable apps and you know, agentic and you know MCP, just building the whole new infrastructure. 00:16:19.360 --> 00:16:20.320 It's still gotta work. 00:16:20.399 --> 00:16:30.639 And honestly, I mean, AI is introducing as much complexity as it is helping to solve the complexity in in your business operations, right? 00:16:30.720 --> 00:16:38.159 So, and that's probably a temporary thing because right now there's this giant land grab and the you know, the battle of the tech guerrillas, right? 00:16:38.240 --> 00:16:44.320 And yeah, things are getting more commoditized all the time through the LLMs and all the hyperscalers, right? 00:16:44.480 --> 00:16:52.320 They're kind of just you know, invoice scanning and processing, and you know it's just an abekuan truly like you know, uh infrastructure and plumbing, right? 00:16:52.480 --> 00:16:56.480 So I think they'll be I think IT will actually be fine. 00:16:56.639 --> 00:17:04.960 It's the the folks that are just actually doing sneaker net kind of operations and just fingers on keyboards that repeatable. 00:17:05.039 --> 00:17:06.720 That's you better start elevating. 00:17:07.119 --> 00:17:08.319 What's what's sneaker net? 00:17:08.640 --> 00:17:18.480 Oh, sneaker net is like you basically go from one system over or or swivel net, go from one system to another and like take one from one system into another, or you swivel in your chair from one to the other. 00:17:18.640 --> 00:17:20.079 So it's kind of like RPA is stuff. 00:17:20.559 --> 00:17:26.720 Anytime you're doing like redundant, you know, hands-on keyboard activity, it's the algorithm is coming for you. 00:17:26.799 --> 00:17:31.440 I mean, well, you know, you mentioned something earlier that I I wanted that I was sort of curious about. 00:17:31.519 --> 00:17:37.039 You said that RPA hasn't really borne out full promise, but now you feel like it it might be able to. 00:17:37.279 --> 00:17:45.359 What is it that LLMs or AI can provide to RPA or automation more broadly that hasn't been in the possible in the past? 00:17:45.440 --> 00:17:46.880 Like what's the relationship there? 00:17:47.200 --> 00:17:47.519 Sure. 00:17:47.680 --> 00:17:55.039 So one of the challenges when you start layering, let's chit just take uh a term orchestration, process orchestration. 00:17:55.119 --> 00:18:01.920 We take a subset of that, which is basically workflow on steroids, which it's basically, you know, and you get this automation. 00:18:02.079 --> 00:18:13.119 Part of that is kind of that automation with the end user and the stuff where there were, you know, hands on keyboards and being able to better automate and a lot of this through experience management tools and other things. 00:18:13.200 --> 00:18:20.000 And the whole technology has kind of evolved from RPA to intelligent automation to agentic, you know, automation, etc. 00:18:20.319 --> 00:18:26.720 But a lot of those integrations, because ultimately when you layer something on top, you got to integrate to all the crap underneath, right? 00:18:26.799 --> 00:18:31.839 It's like one of the most complicated surgeries on the face of the earth is face transplant surgeries. 00:18:32.000 --> 00:18:53.119 And you're like, oh, it's just not to go gory here, but it's

like, oh, it's scanning, but it's a but it's like a day and a half of hooking up the new face to like literally all the systems, the circular, and because so integration is a major issue, especially with security, and that those are vectors, threat vectors, and when it's built, you don't want to mess with it, right? 00:18:53.200 --> 00:18:55.839 You know, which is kind of one of the things where people don't want to upgrade. 00:18:56.079 --> 00:19:14.160 One of the key things is taking these hard-coded integrations, taking these very fragile APIs, and and ultimately being able to get to a higher level of integration where you can say, you know, and in the business system to say, I want to turn a PR into a PO, or I want to take an RFQ and turn it, you know, multi-round bidding event. 00:19:14.240 --> 00:19:17.119 And I just want to say it conversationally at a high level. 00:19:17.200 --> 00:19:19.119 And then the AAI is like, got it. 00:19:19.279 --> 00:19:27.680 Well, I know that you're running six different business systems, you know, some ERP, some S T, I know the dialects that they speak, I know what's required for them. 00:19:27.839 --> 00:19:32.960 I know that one doesn't have it, so I need to either write it myself, you know, or I do that translation. 00:19:33.039 --> 00:20:09.759 And so if you look at some of the things that are happening now in, you know, uh genic, whether it's around A to A or Model Context Protocol, MCP, they're linking these low-level integrations and APIs up to higher levels of abstraction so that you can now, and this is the next generation systems will be writing to these kind of data models that are not in one vendor, but it's to a general purpose object, you know, so it's a virtual, you know, vendor master or you know, bill of material or whatever is and underneath the covers, the tech helps you build the deterministic tooling that actually integrates to all that stuff underneath. 00:20:09.839 --> 00:20:33.279 So now you basically have a you know supplier information management functionality built right in, and it's built right in because the AEI helps you do that mapping between the modeling of the business processes and the user and the data to tools and all those and and basically all the apps that need to be cared and fed for until they're replaced by the next generation systems. 00:20:33.359 --> 00:20:35.599 And that was always the problem with RPA, very fragile. 00:20:36.079 --> 00:20:36.480 Fragile. 00:20:36.559 --> 00:20:47.599 That's that's the word that I had also heard is that you would build one of these automations, and then the minute you that a software vendor changes a pixel, then all of a sudden you have to go and reconnect to the pipes. 00:20:47.680 --> 00:20:50.880 And uh in a way, weird way, it kind of seems like it would stifle innovation. 00:20:50.960 --> 00:20:58.640 Because if you're like, hey, I'm worried that if I if I walk take one step, I'm gonna you know be a bull in a China shop on RPA, then nobody wants to do it. 00:20:58.799 --> 00:21:03.359 So I think it makes perfect sense that that LLMs can make that process more robust. 00:21:03.599 --> 00:21:04.720 It is, it is interesting. 00:21:04.799 --> 00:21:11.759 So I I guess a question I'm curious about though is at what point RPA away people versus enhance them to do more stuff, right? 00:21:11.839 --> 00:21:18.000 So there's two possible futures, or maybe maybe they're both the the future just at different timescales. 00:21:18.240 --> 00:21:25.839 One is saying, hey, we're gonna take people that we have in our organization and we're gonna supercharge them with AI tools so they can do the work of like 10 people. 00:21:26.000 --> 00:21:32.559 Uh there's another mindset where some companies come to the market and they say, hey, what we're gonna do is we're gonna build a, you know, we're competing with people. 00:21:32.640 --> 00:21:38.079 So if you think about in the case of customer service, companies like Diagon, is that what it's called? 00:21:38.160 --> 00:21:38.960 Diagon, right? 00:21:39.039 --> 00:21:39.839 Yeah, Diagon. 00:21:39.920 --> 00:21:48.799 They were they're trying to replace folks who are actually at these call centers with AI agents, like customer service agents, not agentic agents. 00:21:48.880 --> 00:21:54.400 I guess they're both, but but you know, where do you think is the role for like human enhancement versus replacement in the procurement space? 00:21:54.720 --> 00:22:03.119 Yeah, well, I think you know, you you've highlighted it just you can look to analog areas, right, where you can just So customer support, right? 00:22:03.200 --> 00:22:05.680 You can say, well, that's like procurement help desk, right? 00:22:05.759 --> 00:22:11.680 Or any kind of shares, any kind of business function, you know, that's supporting, you know, the business units, right? 00:22:11.839 --> 00:22:17.519 So, and basically managing that knowledge and the FAQs and the triage, you're building a knowledge base, right? 00:22:17.599 --> 00:22:23.759 And and whether you're deploying that right out to the end users or whether it's through a COE or you know, whatever it is, that's great. 00:22:23.839 --> 00:22:25.599 It's a knowledge management exercise. 00:22:25.759 --> 00:22:30.000 And yeah, it's just gonna make them more productive and you're gonna need to hire less of them. 00:22:30.079 --> 00:22:35.440 And maybe there's gonna be some you know reduction workforce that's needed because people are just not as busy, right? 00:22:35.519 --> 00:22:40.799 But you see this also, you know, you come from the manufacturing world, it's the same thing in manufacturing, right? 00:22:40.880 --> 00:22:47.279 You know, when you go to more highly automated environments and this question of, oh, we're gonna replace the people with robots. 00:22:47.440 --> 00:23:06.079 Well, yeah, I mean, when you're gonna bring in more sophisticated automation and whether it's purpose-built, human, like you

know, the Japanese mindset around flexible manufacturing and building small machines and all that, or whether it's the larger robotic systems, or whether, I mean, you work to Tesla, the mega, mega system, it's gonna have an impact on labor, right? 00:23:06.160 --> 00:23:16.640 And you're gonna have to take resources out that are no longer capable of kind of being upskills to kind of be the transformers or the work centers themselves, right? 00:23:16.799 --> 00:23:19.039 And so the same with procurement. 00:23:19.759 --> 00:23:20.480 Oh, it's interesting. 00:23:20.640 --> 00:23:33.920 You know, I mean, my only pushback on this concept of workforce replacement and procurement is that I feel that the cost of the person relative to the amount of dollar value that's being spent is is very small. 00:23:34.079 --> 00:23:41.200 And so, you know, when I used to be a Waymo, as you remember, and I remember I talked to a lot of entrepreneurs who are building different autonomous vehicles, right? 00:23:41.359 --> 00:23:46.880 So Waymo was looking to replace the driver and build a ride share like an Uber or Lyft without a driver. 00:23:47.039 --> 00:23:49.359 And in that case, the driver is like the biggest expense. 00:23:49.440 --> 00:23:51.440 So it makes perfect sense to try to automate. 00:23:51.599 --> 00:23:55.440 And there's also experience improvement that I I'll talk about my opinions on later. 00:23:55.599 --> 00:24:01.839 The second kinds of examples I would see is people would say, okay, what if we could make autonomous uh shipping container boats? 00:24:02.079 --> 00:24:17.920 And I remember thinking like that sounds like not a good idea because maybe you could add sensors and hardware and software to help facilitate autonomy on a transatlantic cargo ship, but there's$300 million of goods on there. 00:24:18.079 --> 00:24:25.359 And the labor of the people, even if you had five or 10 or 20, it's a rounding error versus the cost of the goods. 00:24:25.519 --> 00:24:34.240 And so I feel like there is like some in my mind, I have this a visualization, which is like the ratio of the cost of the human versus the the potential impact one way or another. 00:24:34.400 --> 00:24:43.839 And in procurement, it feels like okay, well, I know teams of I was on site on Tuesday with a team that's like seven people managing like billion four of spend or two billion of spend, something like that. 00:24:43.920 --> 00:24:50.240 And so you'd think, okay, actually don't reduce those people, just like get higher quality people and give them more tooling. 00:24:50.400 --> 00:24:52.559 So I'm kind of curious how that sits with you. 00:24:52.880 --> 00:24:57.200 Well, I've been involved in the procurement benchmarking area for a very long time. 00:24:57.359 --> 00:24:59.839 And and uh I would say that you're absolutely right. 00:25:00.000 --> 00:25:09.680 And there's not a function that probably has a higher ROI than you know, procurement, you know, in terms of it's literally 5x, 7x, 10x, right? 00:25:09.839 --> 00:25:17.200 So you're like, well, if I'm comparing that payback versus my cost of capital, you know, yeah, that seems like a pretty good investment area. 00:25:17.279 --> 00:25:20.000 Now you can get into, well, procurement's cherry picking. 00:25:20.160 --> 00:25:21.440 Do we believe the numbers? 00:25:21.599 --> 00:25:46.319 You know, it's the whole system around savings, you know, you get into this whole dialogue, but regardless, I mean, even if it's 2x or 3x, it's like, wow, we should really move on the I, increase the R, you know, and give them better tools, and let's get it to 5x, 10x, 20x, just through creating more say, you know, savings revenue enhancement, you know, working, whatever it is, right? 00:25:46.400 --> 00:25:49.359 Just create more value to give those resources better opportunity. 00:25:49.440 --> 00:25:49.920 It's so funny. 00:25:50.000 --> 00:25:52.319 I was involved when I was like, this is classic, right? 00:25:52.400 --> 00:26:10.000 When I was doing a benchmarking project, and you know, um, this was back, you know, when I was originally at Hackett, and it was like, you know, we're telling this large you know, defense company, they're like, why are you spending all your time telling us about all this money that we're spending on P2P and this transactional stuff when all the value is in the strategic area? 00:26:10.240 --> 00:26:15.359 Yeah, because you're spending about 60% of your budget on that low impact, you know, stuff. 00:26:15.519 --> 00:26:24.079 So you're basically wasting its expense in you know over half of your money that you're given versus actually creating investment, right? 00:26:24.160 --> 00:26:31.519 And the difference between expense and investment is that it's not just PO processing, you're actually finding new opportunities for value creation. 00:26:31.680 --> 00:26:40.480 And obviously, the direct side is where the value opportunities open up a lot more because now you're talking about products and revenue and you know it, you know, et cetera. 00:26:40.720 --> 00:26:41.359 It's interesting. 00:26:41.599 --> 00:26:48.799 Do you feel like there is a difference between the desire to automate the role in direct versus indirect? 00:26:49.039 --> 00:27:06.160 Because I feel like we've seen some of these sort of automation, autonomous sourcing solutions, but they've always been very tail spend oriented, like anything that is the unmanaged, you know, the island of misfit toys, the unmanaged of the unmanaged of the unmanaged, because that's where it's a no-brainer versus direct. 00:27:06.240 --> 00:27:07.759 You've always considered as managed. 00:27:07.839 --> 00:27:18.240 So I'm kind of I'm kind of curious if you feel that there's the role of automation is actually maybe still uniform, or if there's actually really different ways that people should think about this stuff for direct versus indirect versus tail. 00:27:18.480 --> 00:27:26.960 Yeah, if I you

know, well, forgive me if I kind of if this is heresy, but I'm gonna maybe ignore the direct and indirect distinction for now. 00:27:27.200 --> 00:27:29.920 Low complexity versus high complexity, right? 00:27:30.160 --> 00:27:44.000 Because on the tail spend, that's it, that's indirect and it's goods, and it's more simple, and it is easier to automate because you can build business rules around it, you can, and increasingly on the service side, you can kind of catalogize those things. 00:27:44.160 --> 00:27:51.279 You know, you can build out your AVLs and you know, kind of keep things relatively simple, and it's kind of low, lower spend, it's not as critical. 00:27:51.440 --> 00:27:54.400 On the direct side, you also have C parts, right? 00:27:54.559 --> 00:28:00.960 You also have things that are less strategic, and there are places where you know you're making allocation decisions, right? 00:28:01.039 --> 00:28:05.599 So coming out of a supply planning run, oh, I have what are my best allocations? 00:28:05.759 --> 00:28:13.279 Well, you can build business rules in there and you can do some, it's kind of like tail spend, you know, to kind of do some more tactical sourcing, almost three bids in a buy. 00:28:13.359 --> 00:28:15.200 You actually do three bids in a buy, right? 00:28:15.279 --> 00:28:33.519 If it's like two equivalent parts of so I think you can do that autonomous sourcing on simpler direct stuff as well, but it also does help on the most strategic, you know, direct, you know, areas because of either mega market baskets and the complexity of that, you know. 00:28:33.680 --> 00:28:36.480 And hey, the more we can apply automation, right? 00:28:36.640 --> 00:28:46.160 And use AI to help with that, that's a good thing, even if it's just helping the bidders participate in the event, understand what they're doing, to just get better outcomes in the execution of it. 00:28:46.240 --> 00:28:47.440 There's nothing wrong with that. 00:28:47.680 --> 00:28:49.759 No, I I think it I think it is interesting. 00:28:50.000 --> 00:29:03.039 I um I like the idea of getting away from direct and indirect as the main because it is very much like a an accounting-driven distinction, uh, because COGS is direct and SGNA is indirect. 00:29:03.119 --> 00:29:06.240 It's consumed by the business versus going into the product for the customer. 00:29:06.400 --> 00:29:07.599 I will say that you are right. 00:29:07.680 --> 00:29:17.039 And when I think about how we market our business, we always say that we do direct and strategic indirect, but what we don't really focus on is tactical or tail indirect. 00:29:17.279 --> 00:29:30.799 And it's funny because some of the most complicated events are not the direct material, some of the most complicated are the really large indirect packaging, uh, logistics with scenarios and constraints and award optimization. 00:29:31.039 --> 00:29:34.720 Yeah, energy, renewables through renewable energy, really huge. 00:29:34.880 --> 00:29:39.359 I mean, buying, I mean, I can only imagine how strategic it is buying GPUs right now. 00:29:39.440 --> 00:29:42.640 You know, that is it's that is a full domain. 00:29:42.799 --> 00:29:48.079 You should hire someone just to do that, maybe teams of people just to do that if you're at one of these hyperscalers. 00:29:48.160 --> 00:29:51.519 You know, so we don't draw that distinction, but but it is interesting. 00:29:51.599 --> 00:29:58.720 Like, I think with it's sort of hard for me to have like a very concrete opinion on this because I think automation is easier as you go simpler. 00:29:58.880 --> 00:30:00.880 So tail three bids in a buy. 00:30:01.039 --> 00:30:05.599 You don't want to spend ten doll worth of a person's effort to buy a$5 stapler. 00:30:05.680 --> 00:30:09.440 You know, that was the classic, that was the classic OG Ariba example. 00:30:09.599 --> 00:30:17.119 And so there the problem that people should solve is like how do you transact as easily and quickly as possible and just get the stuff and then don't worry about it. 00:30:17.279 --> 00:30:22.240 But the prize is still significantly bigger on the large scale sourcing side, right? 00:30:22.319 --> 00:30:34.000 If you think about direct and the large strategic indirect, you know, you're never going to optimize the tail enough to make a difference when you can make like a small change there and it yields huge results, but it's more difficult to automate. 00:30:34.160 --> 00:30:36.880 And so I don't think people have looked at it as much. 00:30:37.039 --> 00:30:45.119 One of the beauty around direct is that the supply chain really focuses your mind in a very structured way. 00:30:45.359 --> 00:30:50.799 At the end of the day, products are going to go to customers and they're gonna be profitable, unprofitable, right? 00:30:50.960 --> 00:31:05.599 And everything is synchronized to standing up a supply chain and the cash invested in it to get those products to be pro to be sold profitably and to build, you know, all the IP and stuff that you want into them. 00:31:05.759 --> 00:31:22.079 So, in a sense, direct and the supply chain is a very bounded and deterministic thing, but it's also very complicated, right, in terms of where the relationships between costs and risk and resources and you know all that cost elements, right? 00:31:22.160 --> 00:31:30.400 So it is a very rich data model and knowledge model upon which to apply the analytics and AI and things like that. 00:31:30.640 --> 00:31:34.079 So it's hugely more complicated, but it is very deterministic. 00:31:34.160 --> 00:31:36.720 And that is the beauty, you know, that I've always loved. 00:31:36.799 --> 00:31:41.920 I come from a supply chain background, you know, like you kind of fall into procurement because you're like freaking procurement. 00:31:42.000 --> 00:31:44.240 I mean, come on, doing deals and pushing POs. 00:31:44.480 --> 00:31:46.000 I'm into network design, right? 00:31:46.079 --> 00:31:49.119 I'm

into, you know, the really strategic stuff. 00:31:49.279 --> 00:31:59.759 And if you look at procurement in the right way, it is kind of how do you design an extended supply network rather than just, you know, kind of you know, point-in-time episodic sourcing around whatever. 00:31:59.920 --> 00:32:01.359 You know, yeah, it's interesting. 00:32:01.519 --> 00:32:11.200 I mean, the point that I was going to make, and I forgot about it, but I've but I've regained my assentiency is the question of whether you know autonomous sourcing is truly preferred by suppliers. 00:32:11.359 --> 00:32:13.519 I think that's been a big claim that's been made. 00:32:13.680 --> 00:32:33.039 But the earliest case studies we've seen of tail automation that have been alleged to be successful have really been with companies like Walmart at that scale, where Walmart could say suppliers will even physically move to, you know, will physically move offices to Arkansas, to Bentonville, in order to better serve Walmart. 00:32:33.119 --> 00:32:35.119 So of course they will negotiate with a bot. 00:32:35.200 --> 00:32:39.599 You know, I remember I was talking with Ben, our head of marketing, two days ago or yesterday. 00:32:39.759 --> 00:32:44.160 And he said that once he had an AI virtual assistant, you know, five or six years ago. 00:32:44.240 --> 00:32:49.279 And I I had I tried it once, and we both remember people getting really furious about it. 00:32:49.359 --> 00:32:56.160 Like if I'm going to dedicate my time, I don't want to be scheduling with a bot, I want to be scheduling with a person or book it myself directly. 00:32:56.240 --> 00:32:57.599 So I'm yeah, I'm kind of curious. 00:32:57.759 --> 00:33:01.039 You know, do you think, you know, how do you think about autonomous sourcing? 00:33:01.200 --> 00:33:06.079 Do you feel like it is exclusively the domain of the largest of enterprises? 00:33:06.319 --> 00:33:07.599 Could it apply to everybody? 00:33:07.759 --> 00:33:09.680 Do suppliers actually like it? 00:33:09.839 --> 00:33:12.160 Or is that marketing stuff, you know? 00:33:12.400 --> 00:33:14.480 Yeah, my gosh, there's a lot, there's a lot in there. 00:33:15.039 --> 00:33:25.039 I mean, one is kind of the the sourcing strategy, category strategy, setting up the right design, setting up the right playbooks, thinking through negotiation, what is the best formats for this? 00:33:25.119 --> 00:33:27.519 How do we best lot the items, you know? 00:33:27.599 --> 00:33:32.160 And this is kind of you know, optimization and kind of scenario planning and things like that, right? 00:33:32.240 --> 00:33:38.559 So there's a lot of this upfront stuff that is really important, independent of like before you even get to talking to the suppliers. 00:33:38.640 --> 00:33:57.759 Although if you do it right, um and you do something like, and I just pick kind of the sourcing optimization, it the only thing the thing that's technology is useful for is just opening up the market basket and then saying, Jerry, pick the crap out of it, do whatever you can do, give me what, give me bundled bids, give me whatever optimizes your cost structure, right? 00:33:58.079 --> 00:34:03.279 And then kind of layer that stuff in, apply some constraints, and see what the market can really tell us, right? 00:34:03.519 --> 00:34:08.639 And then we'll figure out kind of how to do the final, you know, kind of how do we set up the events and do the negotiations of it. 00:34:08.719 --> 00:34:13.199 And then in terms of the supplier payback, I don't have enough good data on it, right? 00:34:13.280 --> 00:34:25.679 I get definitely different experiences around it, but I'm generally in agreement with you that suppliers are old hands, right, at seeing what buyers present to them. 00:34:25.840 --> 00:34:27.920 So they've seen all the tricks in the book, right? 00:34:28.000 --> 00:34:30.159 They've seen this so many times before. 00:34:30.400 --> 00:34:32.000 They have their playbooks, right? 00:34:32.079 --> 00:34:33.679 They know what they're going to do. 00:34:33.920 --> 00:34:37.679 So they can pretty much size you up based on just how you're in. 00:34:37.760 --> 00:34:43.840 And you know, it's pretty obvious whether they're gonna be you're gonna be interacting with a bot or not quite often, right? 00:34:44.079 --> 00:34:50.079 You see that with vet buyers getting AI-driven RFP responses from the tech suppliers. 00:34:50.239 --> 00:34:51.679 You know, now we're we get them, right? 00:34:52.800 --> 00:34:57.199 And I'm like, come on, dude, that is so obviously AI generated. 00:34:57.360 --> 00:34:59.039 It's like you're literally insulting me. 00:34:59.199 --> 00:35:04.320 We've had a relationship, you know, and actually, you know, humans racking and stacking this stuff. 00:35:04.400 --> 00:35:07.440 And so I get offended, right, when I see this stuff. 00:35:07.679 --> 00:35:11.599 So you you know the suppliers are gonna kind of they can see through a lot of it. 00:35:11.679 --> 00:35:20.800 So you better do it right around to your point, enabling those buyers to be better about how they campaign these things around the decision support and all that. 00:35:20.960 --> 00:35:23.280 Yeah, it's gonna vary based on the nature of the official. 00:35:23.519 --> 00:35:24.320 Yeah, it's interesting. 00:35:24.480 --> 00:35:32.480 I mean, I feel in a way that the sort of automated sourcing is kind of like a form of virtual e auction. 00:35:32.719 --> 00:35:51.519 You know, it's like, you know, someone described it as a haggle bot, but you know, I think that there is, you know, you can sort of if you have a bunch of bots negotiating in RFP or RFQ situations and they're talking to each other, you're eventually basically using information from all of these to run effectively what becomes an auction. 00:35:51.679 --> 00:35:57.519 So I'm kind of curious, like, you know, have you seen the, you know, I feel like there are flavors that come and go. 00:35:57.599 -->

00:36:12.400 And there was a period of time where eauction was like hugely popular, tons of people were doing it, you know, whether it was free markets, commerce one, uh, Ariba, there were so many auction platforms, and it felt like it went the way of the of the dodo for a while. 00:36:12.559 --> 00:36:16.800 Now I'm hearing a little bit more about eAuctions than I did five years ago. 00:36:17.039 --> 00:36:19.039 What's the supplier perspective on that? 00:36:19.199 --> 00:36:21.599 Is that something that suppliers are open to? 00:36:21.760 --> 00:36:23.760 Why do you think e auctions went away? 00:36:23.920 --> 00:36:27.119 And why do you think that they're sort of starting to creep back in? 00:36:27.280 --> 00:36:31.280 And is that a is that a small trend or is that do you think going to become a mega trend? 00:36:31.519 --> 00:36:34.400 Well, you've also packed in six or seven questions right there again. 00:36:34.480 --> 00:36:42.480 Um, Smarch, when we go go back in time and uh my my uh my my compadre, Jason uh Bush came from free markets. 00:36:42.639 --> 00:36:54.320 And uh we, you know, when I was at AMR Research, which became Gardener, you know, we analyzed free markets, you know, when they started getting into the software business versus just kind of running the reverse auctions as as smoothies for companies, right? 00:36:54.480 --> 00:37:02.079 So I would just say that there was that dot-com era where the executives were really enamored with downward price auctions. 00:37:02.239 --> 00:37:04.719 It's just a great boardroom sell, right? 00:37:04.880 --> 00:37:17.519 You just watch that price going down, and free markets was great at like hooking the CFOs and say they brought them into the warroom and the whole thing, and you know, you just look at the money real time that you're saving, right? 00:37:17.599 --> 00:37:20.639 So there's a gamification, so there's that piece of it, right? 00:37:20.719 --> 00:37:29.280 And so if there is some margin to uh, you know, if there's some fat, yeah, you can certainly do that, but there's also the winner's curse, right? 00:37:29.360 --> 00:37:34.400 You're gonna get the fact that the the win that people are gonna the bidders get caught up in it as well, right? 00:37:34.480 --> 00:37:43.280 And they end up kind of stepping their bounds, or they're like, well, I'm gonna just do it to win the business, and then I'll creep up my costs later after I kind of you know win the business. 00:37:43.440 --> 00:38:00.960 So, but at the end of the day, that technique of just saying of the buyer saying, here is the lot that I'm gonna present to you, here's a form fit function, everything is held constant, and the only variable that I'm really deciding on is price, and I'm gonna do that in a downwards, synchronous, you know, real-time event. 00:38:01.199 --> 00:38:29.280 Just think about how much value space you're destroying by collapsing that down into a little approach without actually listening to what the market can offer you by starting with innovation, opening this up, you know, and then getting into things like the techniques around common trail optimization, which is really scenario planning and opening up that market basket, and then actually going and figuring out the best event, and then whether it's Dutch or you know, Japanese or whatever the uh event flavor is. 00:38:29.519 --> 00:38:36.639 So anytime you're doing picking like a format like that, a bidding format, it should be what out, right? 00:38:36.719 --> 00:38:40.960 And you should always start with how do I maximize the overall value pie? 00:38:41.199 --> 00:38:50.000 Because I did a study on this actually 20 years ago, and I found that the combinatorial auctions really outperformed the straight up reverse auctions by a fair amount. 00:38:50.079 --> 00:38:54.079 And this was limited to transportation, some direct, you know. 00:38:54.400 --> 00:38:55.920 What is a combinatorial auction? 00:38:56.000 --> 00:38:56.880 Like, what does that mean? 00:38:57.199 --> 00:39:04.480 What that means is you are per letting the suppliers bid on whatever set of combinations of bids that they want. 00:39:04.559 --> 00:39:08.320 You put out basically a big market basket, so say in direct, right? 00:39:08.480 --> 00:39:16.639 I'm gonna take all of the uh electronic components or whatever, and I'm gonna put it all into one basket and I'm gonna give it to a distributor, I'm gonna give it to a manufacturer. 00:39:16.719 --> 00:39:19.360 I'm I'm just this is what I need in total. 00:39:19.519 --> 00:39:30.320 And obviously, it's most popular, you know, not obviously, but like in transportation, where you're I'm like, here are all the transportation services that I need to connect, you know, to do volume from here to here. 00:39:30.480 --> 00:39:32.239 That is my market basket of requirements. 00:39:32.400 --> 00:39:38.320 And rather than be like, Well, I'm gonna take this lot and this lot, and I think I'm gonna bundle that and then I'm gonna reverse auction. 00:39:38.480 --> 00:39:39.679 So now I'm gonna have 50 reverse. 00:39:39.840 --> 00:39:41.440 And I'm thinking about how to do that. 00:39:41.599 --> 00:39:42.719 No, no, you don't do that. 00:39:42.800 --> 00:39:50.320 You actually start and just say, open it up, let everybody bid on what they can to optimize their cost structure because you're gonna like this is a win-win, right? 00:39:50.480 --> 00:39:53.199 I want you to win, I don't want you to just overbid. 00:39:53.360 --> 00:39:57.280 Because if you don't, the supplier doesn't know how much business they're gonna win, right? 00:39:57.360 --> 00:39:58.639 In in these things, right? 00:39:58.800 --> 00:40:00.719 Across multiple lots, right? 00:40:00.800 --> 00:40:04.239 And if it's a big one, they're gonna be like, Well, shit, it's winner take all. 00:40:04.320 --> 00:40:06.960 Pardon my French on this, it's it's winner take all. 00:40:07.119 --> 00:40:10.079 So I better say, Oh, yeah, I can do

that. 00:40:10.159 --> 00:40:16.159 And meanwhile, I'm scrambling to like find other people to help supply, and then I'll I'm just gonna mark and pass it on to you. 00:40:16.239 --> 00:40:17.199 So that's no good. 00:40:17.360 --> 00:40:39.920 So it's actually just letting the supply market speak to you by letting suppliers bid in multiple ways, and then you use basically combinatorial optimization to figure out what is the optimal allocations that I have of these items to those suppliers given these constraints to maximize my objective function of total cost service and whatever. 00:40:40.000 --> 00:40:41.280 And you play around with those things, right? 00:40:41.360 --> 00:40:42.320 It's scenario planning. 00:40:42.400 --> 00:40:50.719 So it's a very sophisticated way to approach sourcing, but it also is a complexity that I think is hard for many people to grok on both sides, right? 00:40:50.880 --> 00:40:56.559 If I'm a supplier, it's hard to think about like what is a market basket proposal that I'd be happy with, you know. 00:40:56.639 --> 00:41:01.440 And then on the procurement side, it how do I do those constraints in a way that's that's reasonable? 00:41:01.519 --> 00:41:05.760 Because I might look at it and say, hey, there's an incumbent and I have a slight preference for them. 00:41:05.840 --> 00:41:08.480 I think it's slightly less risky to have a US supplier. 00:41:08.639 --> 00:41:11.920 I think it's slightly preferred for me if they take these payment terms. 00:41:12.000 --> 00:41:14.559 So you have this really complex TCO that you have to consider. 00:41:14.960 --> 00:41:20.079 Isn't it better to have both parties be smarter about that and ask themselves those questions about what are my constraints? 00:41:20.159 --> 00:41:21.519 What is the sensitive analysis? 00:41:21.599 --> 00:41:22.800 How important is that to me? 00:41:22.960 --> 00:41:27.280 What is the profit margin that revenue tied to that component, you know? 00:41:27.360 --> 00:41:30.239 And it's like, I better get my cost models in order. 00:41:30.320 --> 00:41:36.079 And for the suppliers, it's like what I'm doing to them is just like, here's the market basket, just bid on what you want, right? 00:41:36.239 --> 00:41:37.599 And if you want to bundle them, great. 00:41:37.760 --> 00:41:40.639 If you just want to give me separate bids for each one, no problem. 00:41:40.800 --> 00:41:42.159 We're gonna keep it simple, right? 00:41:42.320 --> 00:41:44.639 And it's whatever happens, you're a winner. 00:41:44.719 --> 00:41:45.840 Everybody's a winner, right? 00:41:46.000 --> 00:41:48.320 You get a participation trophy no matter what. 00:41:48.559 --> 00:41:51.519 Last question that I know I know we're over time, so I'll let you hop off. 00:41:51.679 --> 00:41:59.440 What is one company in procurement, not in procurement, tech company that you've seen a demo from recently that you're really impressed by? 00:41:59.519 --> 00:42:00.880 And it can't be light source. 00:42:01.119 --> 00:42:06.400 That's a question that I should probably uh recuse myself from that's also fine. 00:42:06.480 --> 00:42:12.639 I don't answer, but but let me just state what type of technology is is exciting me, Nelson. 00:42:12.719 --> 00:42:13.519 What is okay? 00:42:13.599 --> 00:42:14.960 That's that's also fine. 00:42:15.199 --> 00:42:15.440 Right. 00:42:15.679 --> 00:42:29.280 So the the technology that is exciting me most is the area around ontologies and data and data modeling and data management and knowledge graphs are a way to model that. 00:42:29.519 --> 00:42:35.599 And when I was talking about that whole area around neural and symbolic, I think the there is a whole in-between. 00:42:35.679 --> 00:42:39.280 I do this whole DPW presentation, or anybody's interested, just reach reach out to me. 00:42:39.360 --> 00:42:55.440 Um, it is about kind of taking the richness around neural modeling and LLMs and using the power of that where appropriate to interact with and to create deterministic tooling and to create next generation applications. 00:42:55.599 --> 00:43:02.480 And you see glints of it in data fabric, composable apps, you know, these kind of things around the analytic space. 00:43:02.719 --> 00:43:15.519 You see um a lot of it around just what's happening within MCP and uh Gentic around the abstraction and that ability to kind of hide complexity and manage it explicitly, versus, like, for example, just simple punch out. 00:43:15.599 --> 00:43:18.960 Oh, I go out to a supplier website and I can shop there and bring it back. 00:43:19.119 --> 00:43:28.159 Well, that is like sounds really easy, but one of the problems with it is that's just an XML protocol and you can pass all sorts of parameters and there's no semantic definition. 00:43:28.320 --> 00:43:46.320 So getting semantic structure and being able to bridge that gap between this wonderful world of the LLMs and agents and agentics to that deterministic tooling side and creating this next generation of applications that actually are using ontologies and knowledge management. 00:43:46.480 --> 00:43:49.360 I mean, the best example in the market to me is Palantir. 00:43:49.599 --> 00:44:09.519 Now, you can say whatever you want about and feel about whatever you want about Palantir, but their platform around modeling ontologies and all the tooling around that, and yeah, it's more in kind of a services metaphor, but if you're gonna copy an architecture, I know you guys have a lot of common attributes with you know them, that's a good place to start because they are so focused. 00:44:09.679 --> 00:44:20.719 And any company that is really focused on the data and the knowledge, because ultimately AI is about taking knowledge and tools and turning those into skills that you can apply. 00:44:20.880 --> 00:44:26.639 A knowledge model is as much neural as it is deterministic, it's around semantics as well as structure. 00:44:26.880 --> 00:44:33.360 So that technology area on this next gen of apps that brings neural and symbolic together, to me, that's the game changer. 00:44:33.440 --> 00:44:42.960 And if you start bringing in open source and other things, that is now gonna be a massive driver versus just hey, look at the power of the later of GPT-5. 00:44:43.360 --> 00:44:47.920 It's now a trillion times what my brain can versus a billion. 00:44:48.000 --> 00:44:49.760 That's not a problem anymore, right? 00:44:50.000 --> 00:44:51.119 Interesting, interesting. 00:44:51.280 --> 00:44:53.840 Well, this has been a really fascinating conversation, Pierre. 00:44:54.000 --> 00:44:57.920 Always so good to chat and uh have a great rest of your Thursday. 00:44:58.159 --> 00:45:00.960 Thanks, Spencer, and uh by the way, happy birthday. 00:45:01.119 --> 00:45:04.079 I hope you enjoy your cake that's waiting for you.