Transcript 0:00 Right now, somewhere, software is running a full sourcing event and no human is in the loop. It's picking the suppliers, sending the RFQ, and awarding the business. 0:12 In the last twelve months, that stopped being a demo and became a reality in organizations embracing agentic sourcing. 0:20 However, for every vendor telling you there's AI agents that can do this, there's a procurement team that ran the pilot and then went back to doing it by hand. So which one is true? 0:30 Where does autonomous sourcing work, and where does it fall apart? And more importantly, why? That's today's episode, the twenty twenty-six state of autonomous sourcing. Let's get into it. 0:41 [upbeat music] Computers have lots of memory, but no imagination. Man is a slow, sloppy and brilliant thinker. Machines, on the other hand, are fast, accurate and stupid. 0:54 AI will do what you tell it to do, but that may be very different from what you had in mind. Technology presumes there's just one right way to do things, and there never is. Computers are useless. 1:05 They can only give you answers. For a list of all the ways technology has failed to improve our quality of life, please press three. Welcome to ProcureTech Unpacked. 1:18 This is the show where procurement professionals come to understand the technology shaping our function. So a bit of context before we get into the show today. 1:27 I recorded the following interview on the road at SOAR twenty twenty-six, which was Fairmarket's yearly conference, and I have to say, they went all out. 1:36 They set up a proper recording area right there at the event, and my guest and I were both intimidated by the setting. So a big thank you to the Fairmarket team for setting this all up and having me. Let me be upfront. 1:48 Fairmarket is one of our content sponsors at Pure Procurement, but this episode was not sponsored. Nobody paid for this conversation. I just wanted an honest read on where the market is heading. 2:00 We use a few Fairmarket examples along the way, but most of what we get into is general information to the whole category, and so whatever tools you're running, it'll be useful for you. Recording this way is dangerous. 2:13 It makes me want to travel the world to have the ProcureTech conversations worth having. But that sounds very expensive, so do not get used to this setting if you're watching on YouTube. 2:24 Uh, and if you're listening, it should sound exactly the same as usual. Autonomous sourcing is a deep, specialized corner of the market, and I tried to quickly get to the heart of it with this interview. Enjoy. 2:36 Today, I'm sitting down with Tiago Melo, VP of Solutions Consultant at Fairmarket. 2:42 Uh, T-Tiago and I came up through similar paths in solution implementation, and so we tend to speak the same language, and that's a big part of why I wanted to interview him today. 2:52 Uh, autonomous sourcing is a category that has no shortage of big promises, and Tiago is someone who doesn't sell the hype. 2:58 He's honest about what technology can and cannot do, a-and he's also, uh, someone who doesn't have any time for snake oil, which is something I really appreciate. 3:07 And so when I need an honest take on autonomous sourcing, he's one of the people I call to get, uh, a signal on, on what's going on. 3:13 So I thought it was a worthwhile endeavor to meet with Tiago here at the SOAR two twenty twenty-six conference to get his take on the state of autonomous sourcing. Thanks so much for being here, Tiago. 3:23 It's great to be here, Joel. Yeah. So let's dive right in, uh, and start at the beginning. So when the market talks about autonomous sourcing, as I referred to in the intro, what does that mean? How do you define that? 3:35 Uh, how do you think about autonomous sourcing? Well, it actually depends a little bit on who's selling the idea of autonomous sourcing. Every vendor's gonna have a slightly different take of what it is. 3:45 We like to think of it as rather what the human does behind whatever solution it is that is being positioned as autonomous sourcing. 3:53 So, um, there's multiple stages or states of autonomous sourcing, arguably, and, uh, one of the analysts in the industry, Hackett, actually has a tiered level defining or describing in this, in a rather simple way, the states of autonomy and what constitutes autonomous sourcing. 4:11 Um, but autonomous sourcing is-- in our view, boils down to what can the technology do? 4:16 What can AI and the agent do with, for, and alongside the user that ultimately frees the user from some of the aspects, some of the tasks that the users used to perform in the technology that they are now relying on AI to do on their behalf. 4:35 Uh, would you say it's mostly like to, to enable procurement professionals who do sourcing or to make the business autonomous in how they, you know, in sourcing as well, right? 4:44 And, and taking procurement sort of out of the loop and making them autonomous. There's a path for both because procurement works alongside the business. I mean, procurement is part of the business. 4:54 So there is a degree of autonomous sourcing that is possible for procurement. I mean, Fairmarket obviously exists based on that premise. 5:02 But what procurement does transpires on how the business does what the business does and also engages with procurement. 5:09 So invariably, there's gonna be a state of autonomous capabilities related to sourcing, uh, that are just as important and material for the business, for the end user, for the requisitioner, right? 5:22 Some of the terms that are used to refer that side of the enterprise. 5:25 My takeaway is, like, you need to make sure that, as you said at the very outset, right, whenever you're engaging with a vendor that's talking about autonomous, maybe not just in the sourcing space, like, you wanna make sure you're, you're using the same definition, 'cause I've seen 5:37 m-multiple different ones be used, right? A hundred percent. There are vendors that describe certain aspects of autonomous sourcing as autonomous. That's their label. 5:48 Uh, we actually see it slightly differently because there are different degrees of autonomous. There's the, the, the distinction between autonomous and au- and automated. 5:56 There's the, the distinction between a task that is performed by, for example, a procurement professional that is no longer performed, performed by the procurement professional, but by someone else. 6:07 Uh, so there are- Process changes that are allowed by the technology, uh, which are very valuable, but may not necessarily fall within the common definitions of what autonomous sourcing is. 6:20 I want to address the setting here 'cause, like, as we, we were sitting down for this, we were going, "Oh, this is so formal. I feel like I'm interviewing the President of, of the [laughs] United States." 6:29 So we're here at the Whitman Mansion, and, uh, one thing that has, um, 6:35 come to mind, right, as, as I've, I've watched you guys through this conference as well is, you know, you have a unique vantage point in terms of being able to see multiple organizations and how they engage with autonomous sourcing and the various definitions that as you just outlined. 6:49 So what, what do you feel is, is something that, you know, you uniquely see as a, as from that vantage point that if I'm a digital procurement lead or if I'm someone looking to automate sourcing, I'm not seeing at the, the individual organization level? 7:03 So our vantage point is, uh, rather interesting, uh, like we hope, uh, vendors would have, which is the perspective, um, across the organization, the customer, and not the tool, right? 7:17 We, we have a lot of analysts that focus on understanding, uh, technologies, understanding the capabilities, and they classify it, they rank it. They have, uh, all types of insights around it. 7:27 Uh, obviously we have our own perspective, but our vantage point, uh, is primarily around what the customer does, how the customer, uh, adjusts and adapts to autonomous sourcing in general. 7:40 So we start to see, uh, the very premise that the constant is the tool, the variable is the customer. Mm-hmm. 7:48 And you can have, as we often see, um, similar customers in vertical and enterprise in the way they operate, uh, handling autonomous sourcing, the deployment, the usage, the adjustment to that new reality in different ways and generate different results. 8:05 So our vantage point truly is of, uh, the, the technology being the constant, but then how the customer operates the technology, how they deploy the technology and embrace it being the main variable around, uh, that very premise. 8:22 Another thing that we tend to see quite a bit, and it's really interesting from a sales perspective, which is where myself and my team spend quite a lot of time on, is that 8:32 what customers argue and stress on an RFP for autonomous sourcing for day one, month one, year one very often isn't as relevant for year two, year three. 8:44 Because it is slightly less focused on the tool, the capabilities of autonomous sourcing and more around how they are changing to take advantage of those capabilities. So it's a really interesting vantage point. 8:57 Yeah, and I think in, in a couple sessions yesterday, like some of your clients were actually saying that, right? 9:02 "We, we went in with X vision, X business case, and within year two, year three, we had completely switched that around," right? 9:08 One of the examples was, "We went in with a centralized assumption that we were going to do all the sourcing from the procurement team, and as we were rolling out the technology, realized that it was something that we could give in a decentralized fashion, at least for some categories of spend, to the plant buyers or to the, you know, the local organizations to use directly," which traditionally hasn't been something that happens, right? 9:32 People go, "Ah, these complex tools, like we're going to keep them centralized." 9:35 [laughs] Um, and I, I would want to also dig into, you know, is that, is that a part of the reason why, uh, earlier this year you guys released, 9:46 I, I don't want to say re-platform, maybe that's not the right word, but like the adaptive sourcing, um, uh, principle and concept and, and also platform to, and UI to, uh, to back it up? Uh, it is. 9:58 It's actually one of main-- one of the main drivers. 10:00 Uh, we understand that businesses are enabled by technology, and obviously, uh, in this day and age, uh, agentic sourcing, AI-driven sourcing is a big catalyst for efficiency and effectiveness within the business. 10:15 Um, our technology was proven. Um, we, we essentially were one of the pioneers for tail spend management, and our customers have been encouraging us to expand those capabilities to different use cases. 10:31 Uh, so we understand that capabilities allow business processes, and we realized that where the original version of our, uh, technology was didn't allow for everything that we wanted to empower our customers in. 10:47 So we did, uh, perform a sizable, uh, engineering and product exercise to essentially define a new foundation for Fair Market that allows for not only different use cases, but essentially a refreshed view of how sourcing can be performed and how different stakeholders within the sourcing process, uh, can engage, uh, within the application. 11:14 When I say engage, I don't mean just interacting with the platform, but also take advantage of some of the autonomous capability that comes with agents. So in other words, 11:25 we focus on the idea that, uh, AI or agentic wash, which is a term that I wish I coined, it wasn't mine, but it's really good, uh, that wasn't gonna be one of the things that we're known for. Mm. Agent washing. 11:39 We focus, we focus on-- Exactly. We focused on, uh, making sure that the platform, uh, allowed for a true agentic experience for the most number of use cases possible. 11:51 And so, and if I, if I kind of state my, my understanding right, with an example is like pre-adaptive sourcing it was, you know, as at the outset of your project, you know, give it a scope, give it a amount of spend, target amount of spend for the, the contract, and then you determine what the steps are gonna be. 12:07 Maybe it's an RFI and followed by an RFQ, and you get to award at, at the end, and then you send the documents, and it's very workflow driven, very linear. 12:15 Whereas- The way I understand adaptive sourcing, the way you guys have termed it is, you know, things are gonna change. The context is gonna change as you're running an event. 12:24 Um, parameters are gonna change, and so if you're locked into a linear kind of workflow sourcing project, that's gonna limit you, and you're gonna brush up against those limitations. 12:35 Whereas in adaptive, well, you're able to, to switch things up as you're going along, and the platform supports it with how the, the data objects are, are structured. Is that a good way to think about it? 12:45 It's a good way to think about it, right? It-- And it's not only on the linear process flow standpoint, right? 12:51 As you saw, uh, Kevin introducing the idea of agents working collaboratively to define what's the right workflow. 13:00 Uh, we also are trying to deal with some of the problems that are relevant and prominent, uh, to be dealt with across businesses, which is, for example, um, bad data quality on intake, uh, uh, having challenging stakeholder engagement with the platform, right? 13:18 Meeting the stakeholders where they are beyond just giving them yet another tool that they have to log into and perform, again, steps in a linear fashion. 13:28 So, um, making essentially the technology more, uh, flexible, but also less central to where the work is performed and more of a skill that is 13:41 leaned on by whatever other tech stack and other processes that the businesses may have that would rely on sourcing. And I-I would term that, like, openness, right? 13:51 Like the-- making sure all the APIs are up to snuff, making sure that you have an MCP server, and we'll have an episode on that, uh, is up to snuff to access those APIs and that, to your point, I could probably get statuses on my events from Teams or Slack or wherever I'm at and potentially interact with it via email, right? 14:11 As a supplier, et cetera, or a stakeholder. And so to your point, like multimodal, uh, multimodal use of, of the software, not having people have to go into the user interface to interact with that process. 14:23 And then, and then making the process dynamic based on the changing context of the, the event at play. Exactly. I had a question around like, "Hey, is this, is this just automation dressed up with a new costume?" Right? 14:34 But I think we-- in that, in that discussion, we kind of illustrated like what, what does it mean to be agentic first? 14:40 What does it mean to, uh, to, to, to renew your platform structure to be able to make it available for agents or for, for, for the outside? So openness, adaptiveness. 14:50 Do you see any other things that are required in the, in the era of AI to be successful? The 14:57 appetite for change in processes, change management, and some degree of appetite for risk, because we are, with this technology, challenging some of the preconceived notions of how sourcing is done and who controls what in the processes. 15:12 So it's a different way of, of, uh, thinking, of, of, uh, sourcing. And as a result, um, a lot of what 15:22 leads customers to, to a good outcome isn't necessarily around what the capabilities of the tech-technology are, but how they embrace those, which has an impact on the governance, which has an impact on process, which brings different stakeholders, uh, into the mix, into the decision process. 15:40 Like for, for, uh, a long time, we didn't have to involve, for example, uh, legal or risk. 15:47 But now that we're dealing with agents in multiple different ways, uh, what-- which are operating, we hope, within the policies that the customers define, there's a different set of questions that are raised that go beyond procurement, again, uh, based on your implementation, that different, uh, uh, governance and controls have to be thought of and different stakeholders have to be part of these. 16:10 So that's interesting, and, and it leads me to think about, you know, pilot jail and the fact that there's a lot of organizations which are stuck, 16:18 you know, have, have delivered a pilot in a test environment or in a sandbox, giving interesting first results. 16:24 But to your point, right, it might be more complex to, to implement, uh, or to get everybody aligned on what you need to scale this to the enterprise level. 16:33 Uh, although I'm sure you have clients that have done that, so I'd be really interested to get your take on like why do people get stuck in pilot jail, specifically in autonomous sourcing, and what do they need to do to get out of there and to scale these types of solutions? 16:46 It's a really good question, and we actually have a session today with our friends from Liberty Blue talking about that very same topic, right? Why do some pilots fail to get their wings? 16:57 That's actually the, uh, the name of the session. There's a couple of simple principles behind why that is, in our view. Uh, pilots are structured to succeed from a technology evaluation standpoint. 17:10 Um, we generally see customers choosing the friendliest categories, the ones that require perhaps less change to test the new, uh, technology with the friendliest teams. 17:25 It's essentially a process that is designed to survive for weeks, for months, for a year, however the, the length of the, of the pilot is. 17:34 Um, getting the pilot to essentially move into the production, into the business as usual, it's not a technology question. It's a procedural question. 17:46 Can the customer make the necessary changes within their processes, within their teams, uh, to operate with the new business as usual, um, that ultimately takes advantage of the tool? 17:59 As I and a lot of customers would, would attest to this, I tell customers all the time that if you buy a new tool and operate the old way, all that you have is a very expensive paperweight. Yeah. 18:10 You won't be able to take advantage of it. 18:12 And pilots, while they're great to validate the technology- They often don't put emphasis on validating the necessary, uh, change management process changes, governance changes that are necessary for the new normal of the organization. 18:28 I find that's a great insight, right? Like, if you're gonna be p-testing the technology, how can you also test the new org structure, the new roles and responsibilities, right? 18:37 Without changing all the HR role, role descriptions and roles and responsibilities, even though you're operating with a friendly-- Well, hopefully you're operating with a friendly business unit or a friendly, uh, category that's simpler to manage, but I think that's a great insight, right? 18:51 It's like, as you're designing pilots or IT is forcing you to run a pilot, how do you also do the, the business transformation pilot at the same time to set yourself up for success once it's time to deploy this thing to the rest of the organization, and already know where you're gonna have those change management hiccups, right? 19:07 Where you're gonna have to put more, more effort around supplier enablement or potentially, you know, des-decentralized model where you have to retrain and, and get people to rethink about how they execute their work. 19:19 Exactly. Very cool. Um, and so we've talked about, you know, the possibilities. I'd love to shift gears a little bit to get your sense on, you know, agentic, the era of AI has this awesome possibility in front of us. 19:32 If we hear the news, it's, it could be quite catastrophic as well. Uh, but on the ground here, like, where is, is, is AI strong? But where, m-more interestingly, does it have still weaknesses? 19:44 Or i-is-- are there things that you think are, yeah, are, are Achilles' heels or things to watch out for? The, the very premise of SOAR this year is to answer that. 19:55 Uh, and AI is getting stronger from a, a sourcing enablement standpoint. Uh, historically, traditionally, AI performs really, really well when you want to 20:07 competitively source something that you have good data for, that you have, uh, suppliers for, and you have an environment where you can essentially be competitive. 20:17 So a lot of folks describe that as a commoditized, um, buying environment. Traditionally, uh, autonomous sourcing operates really well. 20:27 Some of the areas that we see have a little bit more attrition on taking off is when, for example, defining the actual requirements of what you need to source is the work that needs to be done. 20:40 Um, there are advances being, being made by multiple, uh, providers, not just us, on that front that push the customer forward in that task. 20:51 But it may not be, um, an area where AI is quite there yet at the same level of the commoditized, right, the competitive bidding event. You have situations where, 21:03 um, you have relationship-based, uh, interactions with suppliers. 21:10 We, we, we often refer to it as when you are twenty percent of the business of the supplier, and a lot of the, the, the work that the supplier does for you is specific creative work, for example, where it's not-- it isn't as simple as, "Let's go and find more suppliers that do this," because you have a relationship base. 21:30 The supplier almost embedded itself within your supply chain, and there's a very good reason for that. And sourcing really isn't the lever to, uh, to, uh, essentially- Absolutely... influence those, those relationships. 21:43 And then you have more novel use cases, um, uh, like, um, 21:48 uh, like very specific, uh, consolidated, um, or regulated spend, where controlling the spend, uh, and, and, uh, classifying what good looks like is a process that takes months, that they-- that is done over quarters, and sourcing really isn't the lever there. 22:09 Mm. Um, it is a different approach to how that spend is managed. So the, the technology is enabling some of those areas where historically, um, it may not be as good at, but they're still working. 22:22 Yeah, and then from what you're saying, right, what I'm, I'm hearing is, like, it's not necessarily based on the technology, it's more based on having clear requirements, right? 22:31 Like, having stakeholders align, like, having a, a clear objective of what you're wanting to maximize with your, your sourcing event, right? 22:38 Which, uh, then an LLM, which is a statistical-based, like, next-word guesser, if I put it to-- like, it's... And, and integrating that, uh, into the, the, the process. 22:48 Like, you're gonna be able to facilitate certain things or help draft stuff from a sourcing standpoint, but we still need to be clear on what we're trying to achieve. Is that a fair way of characterizing it? It is. 23:00 We actually get asked that question virtually every single, uh, customer engagement. They, they talk to Fairmark and they ask, "Hey, where will your tool perform well?" They even give us their spend file. 23:12 We, we comb through it. We know what good looks like based on the billions of dollars that we have going through the platform every year. So we understand what works, but there's only so much we see. 23:23 We always require context from the customer, and sometimes when we're seeing transactions that are traditionally slotted into tail spend because they seem small, because they seem almost irrelevant in the grand scheme of things, but it just so happens to be the one supplier in North America that sells that good or service. 23:42 So it is then that perspective of, is this sourceable at all, or do we want to step back and realize that there's multiple avenues on how, uh, the business process needs to be performed? 23:55 I think context is the, the, the word that's super important, right? 23:58 Is like, as you run events through a tool and as you capture that context of, "Hey, this supplier, we sourced it once, but there's only one supplier, it's single source, like, it's super critical to our business." 24:10 I think you gave me an example of, like, vanilla offline, right? Of, like, a specific type of vanilla, one supplier in a geographic region and, and a whole part of the business that hinges on that vanilla. 24:21 Are you do-- guys doing anything to capture that context and then to help inform the next person that comes along if the category change-- category manager changes or whatever else happens, right? 24:31 To say, "Hey, next time we try to run an event on this or we run something adjacent to it, here's some really helpful context that might help you craft a strategy." 24:40 So in our view, that's actually one of the hallmarks that makes an, an agentic or an autonomous sourcing system really autonomous, not automated, is the ability to learn from feedback. 24:52 Now, a lot of, uh, a lot of customers, uh, know very early in the journey that we're gonna need historical transaction data to teach the system. 25:00 That's a starting point, but that's a starting point only to solve what our engineering team calls the cold start. Mm. What does essentially the tool know about how your business performs? 25:11 Um, as activity rolls out in the tool, as stakeholders engage with the agents, that the agents essentially learn what's normal, learn what's applicable, and frankly, learn what not to do. Mm. 25:25 So that the very, uh, idea of recommendations is based on your, uh, business processes and principles. 25:33 So it's not just about setting up the policies, but also, to your point, understanding some of the nuances from certain categories that are gonna require a different way for that spend to be handled, which is also one of the big reasons we're launching new agents. 25:50 Uh, we're announcing them here at SOAR. For example, strategy agents that based on whatever it is that is being requested, what's the right way to deal with that? And it's not just how, right? 26:04 How are we gonna open a multi-round, uh, event? Maybe we have an option if we really want to focus on aspects like pricing. But also, should this be sourced at all? 26:15 Or is the system saying, "Based on what we've seen so far for this category within your organization, this is, if anything, a single-source transaction." Mm. "This is relationship based. 26:26 Your lever isn't on trying to get, uh, a lower price through cont-competitive bidding because you're not competitive." And, and here's the, the documentation to back that up, right? 26:36 Past events, uh, policies, whatever else, right? We try to make it as simple as possible because the last thing we want to do is to give twenty pages of documentation- Yeah... as to why that is, right? 26:48 Why is the model reaching out to that conclusion? It's really complex. We're aware of that. 26:52 We're trying to make it simple, uh, for the folks behind the screen to essentially teach the AI, but then again, take advantage of what the AI has already learned about them. 27:01 Yeah, it's-- but, but-- uh, and the way you're describing it, I see it very much as a collaboration, right? We don't want to solely depend on, on the AI. 27:08 Um, and that's further reinforced by a point that I think one of the clients that was presenting made, which resonated deeply with me, which is, you know, w-- there's always been this prerequisite of like high-quality data to implement a sourcing tool, right? 27:22 Oh, you need, you need really high-quality vendors, category strategies, policy, all of this. And so it's, it's often a roadblock that people put in front of themselves, right? 27:30 To say, "Oh, well, we're not ready yet because we need to do a lot of work on, on data quality before we can." 27:35 But what this client said is, "No, actually we did it the opposite way, where we used, we, we used a cold start, 27:41 uh, in the platform, and we just started accumulating that context and that data, and that is what helped us-- It's actually helping us improve the data quality over time just by using the tool." 27:52 Is that something you see, generally speaking, with this new paradigm? I've been working in the, uh, procure tech space for a few years now. Yeah. 28:02 [laughs] And if there's one thing that I realized is that there's no such thing as perfect data. Organizations strive to have perfect data or good data. 28:10 What we're noticing, especially in the world where we have agents that can do certain things for us in that regard, is that good enough, frankly, is good enough. Mm. Um, you may not have all the aspects. 28:20 You-- maybe you, you perhaps don't have the best intake mechanism or the nicest descriptions or all of your supplier contacts, or maybe your taxonomy isn't really there. 28:31 But doing a two-year project to improve all of that generally slows things down. Mm. 28:37 And that's what that customer yesterday mentioned is, "We're aware of the gaps, but we're actually using the e-sourcing process that's being aided by the agents to actually help improve the data." 28:47 So the journey is focused on the idea that looking back and looking what needed to be changed that ultimately gets changed, ends up improving the customer's data quality. But, um, w-we often get that question, right? 29:01 What do I need to deploy, uh, a solution like Fairmarket and an agentic sourcing solution? None of it is AI. It's all about data quality. It's all about, uh, processes. 29:14 It's all about the ability to perform change management, to be aligned with initiatives that allow you to take advantage of the, the solution. As a result of that, your data quality will improve over time. Mm. 29:25 That's generally the best way, and our customers are testing it. It's the best way to improve your data quality, is to be honest about the process and run it. I feel it's like just in time as well, right? 29:35 Like it's event by event, and you're gonna get a higher quality decision for each event, but then the, the, the aggregate result is better data from which to structure the next cycle of, of sourcing events or even, you know, your pricing or whatever else, right? 29:50 Uh, okay, cool. Interesting. Let's shift gears here. So we've been talking about AI agents. Uh, I feel like it's, it's all I ever talk about these days [laughs] anymore. 29:57 Uh, what's, what's your perspective on where, you know, ownership, accountability, uh, and, uh, governance need to be, uh, in this agentic world, right? 30:07 Because that's the big fear is like, oh, we don't know what this agent does, how it's configured, what, you know, what it's gonna do. It's a statistical model, so it's not gonna reproduce the same result every time. 30:17 So how do you think about those, those key topics around, you know, governance, uh, and, uh, and accountability? Interesting you ask that question. 30:25 Uh, one of the polls that we got yesterday with the audience is, uh, does the audience believe that, um, an agentic solution can run fully autonomously, or should it essentially ask for validation, verification at every step? 30:41 Um, we actually got a surprising amount of mixed results in the room. Uh, that goes exactly with that is, um, the idea of accountability with AI. The accountability doesn't shift. It stays with you, the customer. 30:56 But rather than, for example, approving four hundred events, a-awarding to the right supplier four hundred events, if you are able to define the policy that the agents should follow, then you're still in control through the policy on what the AI is doing four hundred times for you. 31:16 Uh, AI, the agent, um, is not accountable. They are simply performing the execution, following the directions, following the processes that you, the customer, still define. So the accountability doesn't change. 31:31 What changes is the nature of the execution. And is it, uh, is it about using-- 'cause I've, I've been hinting at this fact that, like, an LLM is a statistical model, right? 31:40 So it can, it can give you different answers to the same question when you ask it two different times, right? And so we call that non-deterministic, right? 31:48 And so how are you-- If I'm defining a policy, how are we ensuring that as we go through the process of sourcing something, we are making deterministic checkpoints or auditability to make sure that on those four hundred runs, it's not, uh, I don't know, three hundred, three hundred and ninety times that it's okay, but it's really four hundred times, right? 32:09 How do you think about that? So th- we use the same standard, uh, when, when discussing with customers and even when developing the solution, we use the same standard ourselves on how to build, uh, the LLM. 32:20 Essentially, if, if an LLM or the agent rather, because the agents lean on LLM, but not only. Yes. Um, when the agents are performing a task, whatever task they perform, can that outcome be auditable? 32:34 Would a category manager who's done the exact same thing be able to justify what was done? 32:39 So the auditability is not so much as understanding what the LLM did, but why it did it, what policies it followed, what was the version of that policy that day, because customers do change the policies over time. 32:54 It is a-- it's an exponentially growing set of tasks that the customers are gonna do is evolve their policies so that the agents perform differently. 33:04 Um, it's essentially how that, that standard is imposed within our product and engineering team is will procurement, will category managers be able to look at our tool and be able to answer, "Is this a data problem? 33:19 Is this a policy problem, or is this an LLM problem?" And when we say, "It's the LLM issue," that's an easy answer, and it isn't always the right answer. So we need to be able to say, for example, "It's not, 33:34 uh, an audit log, what the model did. Can you explain the model?" That's fine, but that, that is not a standard for an audit. 33:40 It's to understand the reason behind the tasks that were performed, what was followed, what escape mechanisms were there that were or were not followed, and as a result, ensure that whatever processes y-you're running with agentic sourcing are still compliant with the underlying policies of your organization. 33:57 And I, I just wanna, uh, pick up on something you said around, like, evolving policies. Like, I've been with, uh, organizations over the years where the policy is, like, an immovable object, right? 34:08 Like, it was signed in nineteen ninety-six, and, like, it hasn't been changed since then. So I-I've written at length, and it's just a trick for viewers, right? 34:15 Which is if your policy is like that, right, and it just seems like this big thing that's standing in your way, it's like do the battle once again to go up to the board to get it changed. 34:23 But put a change control procedure in the policy itself so that you are able to make it a living document and/or you point to other types of documents like a standard or a directive or whatever else you wanna call it. 34:36 But give yourself a framework to do what you're saying, Tiago, which is make that policy a living document, evolve it over time, yearly basis at, at the maximum, right? 34:46 So I-- when you said that, I feel, like, really passionate about that because I've been in situations like, "Well, we're gonna design this system really badly because the policy, you know, forces us to, to do so." 34:55 So please don't do that. Um, and in the age of AI, I think that this, the, the teams that have the speed of change for policies, business rules, et cetera, to your earlier points, will, will make more, 35:07 uh, more headway on that, uh, on that front. And you asked me a few minutes ago around what AI strong or not strong. Um, we get that question asked all the time, including policies. 35:20 When we're, we're, uh, talking to a new customer and they have policies. For example, they have a value-based policy that says anything under a certain dollar amount 35:31 should go through a tool like Fair Market, should get three bids, and anything above that does not go through the same process. 35:40 Uh, but then we, when we start looking at their spend, we realize it's not value-based, it's shape-based. Mm. Definitely example, small transaction. 35:49 Yeah, it may follow the policy of anything that's under this threshold should operate in a certain manner, but it's not a good fit. 35:58 And conversely, you may have the five million dollar transaction that is just as simple and commoditized as something that is relatively low value that fits the policy. So it is normal 36:12 for our engagements to essentially also lean on challenging the policies because customers realize very quickly it's a shape, uh, question, not a value or threshold question, which generally is what their policies lean on- Yeah... 36:28 prior to, uh, agentic sourcing coming in. Yeah, absolutely. And it, it also brings me to think about, like, 36:35 you know, the, the, the change in roles and re- and responsibilities because as I was hearing you talk there around validating the, the-- what's coming out of AI and ensuring that's aligned with strategy, like, to me, that's- It's an editing capability. 36:48 It's a managerial capability, right? And so-- And it's a business analysis capability of like, did-- You know, what's the process? Did we follow the steps? Should we change the steps and the rules here? 36:58 And so for me, that's like the big shift in agentic AI, at least from a, a personal skills level, is how, as a top performer or, you know, a senior category manager, even a junior one, how do I develop these managerial skills that aren't necessarily for people, but that are gonna be for the agent, uh, the agents that I'm, I'm running or operating with as a, as a professional, right? 37:20 And so I think that's-- I don't know if you wanna comment on that. It's just a, a thought here. [laughs] You, you were touching on essentially what changes at the human level within certainly procurement- Yeah... 37:28 on an organization. And there are changes, absolutely. Um, bluntly, the tactical buyer, the role that it performs is changing. Yeah. Because the tactical buyer has historically been focused on execution, right? 37:44 Clicking the right buttons at the right time to, uh, perform the right decisions. As fast as possible. [laughs] As fast as possible, right? With enough capacity to deal with all of this. 37:53 I mean, we're oversimplifying it here, but historically, that's an apt way of describing what the tactical buyer does. 38:00 Because you are essentially, uh, enabling the agentic sourcing to do a lot of that, that work for you, then the role of the tactical buyer is invariably changing, for example, to agentic supervision- Mm-hmm... 38:13 to policy wr- uh, writing and policy monitoring it. So you are not so much, uh, in, in a role of execution, but you are now in a role of supervision. 38:24 What, for example, Kevin refers to as human in the loop, making sure that the system can perform whatever it needs to perform and only call the human to action when the human is needed, essentially for judgment. Yeah. 38:37 Right? Category managers maybe don't have 38:41 that sharp of a change, but some of their execution in the past, like those large events, that was part of their craft as well, is creating essentially the right, uh, competitive bidding package, if that's the case, or evol-- or, or the overall engagement with their partner vendors. 38:59 Uh, some of that can be as well performed and automated through agentic sourcing. 39:04 So maybe their role doesn't see, uh, that dramatic of a shift like the tactical buyer, but they will certainly be impacted, we think, for the better, 'cause we're essentially injecting efficiency into the way they perform. 39:16 A-and my head just goes towards like, well, if I don't-- if I have time to, uh, you know, to be more efficient with certain parts of my job, then I need to shift that to like building up the tactical buyer competency to be a business analyst, to be a person that looks at policy changes and rule changes. 39:32 Uh, so it's all very interesting, uh, on, on that front. Uh, I'd love to end on, uh, the, the, the eternal debate of like, you know, end-to-end platform, end-to-end suite versus specialized tools. 39:47 Um, Fair Market being a specialized tool, right? Like, what is your perspective on, uh, big suites building their own agents while you guys also start building your agents? 39:58 How do you see that shaking out or fitting together or, or how would you comment on that? So my position is a privileged one because I've actually been on both sides. 40:06 I've been on the specialist side, I've been on the suite side, and the way we see it, you need both. Uh, the platform 40:15 solves very well for your foundation, your underlying processes, your data, your policies that are encoded on how the platform works. The specialists, they're specialists for a reason. 40:26 We do see suites moving more and more into agentic and building those capabilities. 40:32 But while they are doing that, so are the specialists who have a very narrow, a very focused point of view on a set of problems that they're solving. 40:40 And we find that, um, on average, a specialist solves a problem in a different, better way than suites historically do. Mm-hmm. So it's, it's-- we can, we can almost, uh, use, uh, an analogy within the automotive market. 40:57 When Tesla launches electric cars and now a lot of the other manufacturers are moving into the electrification to catch up. Well, while they're catching up, maybe the pioneers are pushing the boundaries forward. 41:08 That's the same view with, uh, AI-enabled sourcing applications, is we're pushing the boundaries forward, uh, that ultimately allows the customers to think a bit differently, which invariably takes the customers down to, honestly, the real battleground, which is integration. 41:24 Mm-hmm. Can we have an integration that justifies its keep, uh, so that we can have a specialist within our process that is enabled by the suite? AI is helping us there too. 41:38 Now, with the advent of MCPs, we're making integrations far less complex, far less costly, and it makes more sense to continue to lean on your underlying foundation in your, uh, STP, to use that, that term, and then lean on specialists for what they do really, really well, and then together continue to push the boundaries forward. 42:02 And I feel, just to, to add a, a point on that, it's-- you talked about the shape of sourcing events and, and the impact that that has on how you wanna address it. 42:09 I would say the same thing around the shape of businesses, right? Like the businesses where sourcing is gonna give them a massive advantage if they have really deep capabilities. 42:18 And so adding that to their, to their stack is gonna give them an outsized return, right? And so-- And maybe there's shapes of businesses where that isn't the case. 42:26 It's more SRM or it's more some other part of the value chain where a specialist solution is gonna earn its keep, to use your, your expression. 42:34 And so I feel like it's, you know, it aligns with what you're saying, but it's also, it's gonna depend on where the value levers in your business as a procurement organization. Let's take a couple minutes here. 42:43 I think we've, we've-- you've been more than generous with your time around and expertise around autonomous sourcing. 42:48 I'd love to, to, to di-dive in selfishly at the end of these int-interviews with, uh, you know, the man behind the, uh, behind the ideas, so to speak. 42:57 Uh, how, how did you end up working in, in this corner of procurement, right? I think you're originally from Portugal. Now you, you, um, you're out of the States. So what's, what's the story? 43:07 In a, in a simple sentence, I didn't know this space was even a thing. I almost stumbled across this space over the years. So, uh, I actually left Portugal in my, uh, last year in college, never came back. 43:19 So that's one, one of those transplant stories. Uh, left Portugal, um, ended up initially in, uh, in the management consulting house and then moved, uh, shortly after into procurement technology. 43:31 Uh, within procurement technology, I started, uh, to essentially f- falling in love with the idea that technology does empower business. Technology is more than its own. 43:43 There's a purpose behind it, and procurement became an area that I'm fascinated with simply because it's such a broad area, uh, with so many problems to solve that I'm not aware of a single person who could answer, uh, all aspects of it. 43:57 So I ended up working in, um, in specialists. I ended up, uh, as well working in suite, uh, providers and in different parts of, of the procurement process. 44:08 So I wasn't aware procurement, procure tech, these types of problems were a thing, stumbled across it and decided, "Hey, I really like this. I'm gonna stick around for a while." Yeah. That's funny. 44:19 Yeah, 'cause I always tell-- I always joke with people like, "I always have the same conversation over and over again, but in different contexts, different industries, different size of businesses," and that's what keeps it really fresh and interesting because the answer is always different based on those contextual elements. 44:32 So very cool. And I know you're-- you alluded to it with your example, right? But you're a car guy. So what, what are you, what are you working on these days? What kind of crazy things? 44:39 I know you, you, you do a bit of racing as well, I believe. I do. I do a little bit of, uh, track day racing, some autocross. Not the most diligent driver. Um, Fair Market does not pay for speeding tickets, FYI. 44:52 Uh, but I do tend to spend a lot of my, uh, free time tinkering with cars, playing with cars. I enjoy driving. I'm, I'm known to do 12-hour drives simply for the sake of driving because I enjoy it so much. 45:04 Oh, you're one of those Sunday drive guy- "Let's go out for a Sunday drive." Hundred percent. Everyone should have a convertible in their life. It's one of my mottos. [laughs] I love it. I love it. 45:12 And then lastly, like if you have advice for, for younger people watching or for people who are, uh, you know, looking to, to procurement or are starting out in procurement and looking to have a fulfilling career, do you have any-- How do you think about that for your own career? 45:26 One of my guiding principles throughout my career has, has always been this: Uh, put yourself in a situation where you will be wrong in public. It takes you out of your comfort zone. 45:37 It forces you to learn, to grow, and to adapt. And being a solution consultant at heart, um, uh, that's, that's my day to day. I sit between the customer and their problems and the product. There's no place to hide. 45:51 [laughs] So you are, you are focused on understanding the problems, and you're focusing on solving the problems with the help of technology. And sometimes that means that you don't quite understand it. 46:01 You don't quite know, uh, what the next step, what the answer is gonna be for the next question. But that's okay because that essentially puts you in a position of growth. Yeah. Very cool. 46:11 And I feel like it's such a, a good way to wrap up because it's also, I feel, the ethos of, of Fair Market as a brand in general, which is like, you know, we're working really hard on developing, you know, leading frontier capabilities in the autonomous sourcing space. 46:24 But we're also very humble in the fact that, you know, we're, we'll find-- We don't have all the answers. We'll help you find them. We'll discover them together in your context. 46:32 And, um, and I'm very glad we got to sit down because I feel you're the, the embodiment of that, and it's great advice for, for our listeners. So thanks so much, uh, Tiago, for, for this interview. 46:41 And, uh, thanks for inviting me and hosting me at SOAR here in this beautiful venue, and I hope we'll get to chat again soon. Thank you so much. [upbeat music] That's a wrap on this episode of ProcureTech Unpacked. 46:52 If this one resonated, subscribe wherever you get your podcasts and sign up for the ProcureTech newsletter for weekly insights between episodes. 47:00 If something we covered sparks 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.
