Ep 01: The Agentic Procurement Diaries: Why This Is a Once-in-a-Few-Centuries Revolution

0:00Hi there, my name is Philip Eidson and welcome to the Agentic Procurement Diaries. It's a show where we explore what procurement could become in an age of Agentic AI. The Agentic Procurement Diaries is presented in partnership with Xycus, a leader in Agentic AI for autonomous procurement. If you are a CPO navigating this shift, this show is for you. The truth is that nobody has all the answers yet, not even the most forward-thinking leaders. We're all learning, experimenting, and figuring out what this technology could mean for our teams and our function. In each episode, you'll hear from CPOs about what they're experimenting with, what they're learning, the challenges they're encountering, and how they're preparing their organizations for what comes next.

0:42These are conversations about the journey told by the people taking it.

0:50Hey everybody, and today Art of Procurement is kicking off a brand new podcast series. It's called the agentic procurement diaries and it's presented in partnership with Xycus now it's going to be published within the art of procurement podcast feed and this show is going to allow you to hear how CPOs are thinking about and advocating for their team's agentic AI journey but in their own words so here to talk about the project to kick us off in this series its objectives and how we collectively believe it's going to help the procurement community I'm I'm delighted to welcome Atish Dedia. Atish is founder and CEO of Zyka.

1:25So welcome to the show. Hi, Phil. Thanks for getting me on the show over there. And we're very excited about this. I think we are in, I believe, and you'll see, hear that word again and again. We are in the midst of a revolution. And I think in a revolution, we need to, as a vendor, we also want to invest a lot on education of the community in terms of what's really changing, how it's changing. And I think Phil has a fabulous community over there. And that's the best place. It's only procurement people. So we really want to get the message out. And then in the series you're going to have, Phil is going to talk to multiple leaders in procurement about how they view agentic AI over there.

2:05So this is our bit over there of educating the community. Thanks. And we're so big on peers helping peers and taking experiences from each other because at the end of the day, that's what helps you really drive, have confidence to make decisions to derive changes when you're hearing how your peers have done it as well. And so that's a big part of this series is bringing those procurement leader experiences and making them as available and accessible to everybody as possible. Now, before we go into some of the details of the series, I do have to ask you kind of my stock question I ask all the time at the beginning of a podcast.

2:46And that's, did you find procurement or did procurement find you? Definitely procurement found us over there. So rewind back to early 2000, 2001, 2002. We had a technology at that time and it was machine learning, correct? So we had machine learning technology for aggregating internet data and classifying it, correct? So it was that time a Bayesian classification model and so on. We had no clue about procurement, but we were pitching it to a VC and we said, take all the information on the internet and then classify it you know the way yahoo and others would do and the vc asked can you do the same thing for procurement catalogs right okay i didn't know anything about procurement we went back but why not um so we went we looked at it we you know worked a poc over there with one of the customers and then the journey began correct so procurement definitely found us but it's it was really the absolute right time because when we entered over there you know the age of digitization of procurement was really starting correct i think um in early 2000 it really started over there and we've been through the entire journey correct and now it's another inflection point which is coming yeah yeah that was about the time i started in procurement as well and um you know it was uh e-auctions was kind of a starting point for us in the technology that we used and um and obviously you know things grew very slowly so it grew slowly and then started to explode in terms of you know the impact that technology and digitization can have on the profession and and now we're at this such a such an exciting time i think for most of us i know there's a lot of fears as well but i think it's a really exciting time for like what comes next because the possibilities are probably beyond anything that we can imagine you know we we sit within this four four walls of what we think procurement is because that's what we've been taught procurement can be and now those walls are disappearing and now we're not figuring out okay what can we really be when we use the technology that's available to us.

4:43No, I totally agree. I think what happened with the digitization wave which happened is it finally got procurement to the table, correct? At that point, procurement was viewed as really tactical and then you had, in fact, not all organizations even had a chief procurement officer, their VP of procurement and so on and they kind of migrated to that. But this is going to be again very different over there, correct? I think it's then, again, it's going to change how CPUs are viewed and they'll be viewed even more strategically because of AI, and I'll come to that later. Yeah, so we talked a little bit about the last 20 years.

5:16We've had automation, digitization around analytics, machine learning, as you said, is kind of where you started. More recently, the last three or four years, now generative AI. I wonder if you could talk about what changes when we introduced the idea of autonomous agents, and perhaps even a little bit before that, because I think it's important from a definition perspective. the difference between, you know, as we throw around terms like generative AI and autonomous agents? Sure, sure. Okay, good question over there. But let me classify into two buckets. And on bucket left is automation or digitization, right?

5:52What the software which exists as of today over there, correct? And which was primarily more standardization of processes, centralization of data, and then analyzing and digesting those data, correct? And then obviously you had forms and workflows over there. So what the automation really did over there, it didn't automate, but it standardized. So it made sure that people follow the same processes, the data is visible to everyone. So visibility became an outcome over there of automation. So it definitely had its value over what was happening earlier. Earlier contracts used to be in drawers and now contracts would be visible with all the clauses and so on.

6:34So that was one phase. Now, the second phase I'll talk about is AI. It started off machine learning. It's like I mentioned, it's been a few decades since machine learning is there. And machine learning was good at classifying data and analyzing data over there. But technically, Gen AI is a subset of machine learning. Now again, that terminology, but what Gen AI did was it could generate new content. I think it, so earlier you had DeepMind, correct, which was in UK and they really changed how neural networks work. So it can generate data. Previously, it was doing a task over there. Now it is generating information based on whatever it knows.

7:14So Gen AI, which started with the chat GPT era, everything after that, agentic AI is nothing but a thin layer on top of the Gen AI. So it obviously uses the power of the LLM, but you need to harness the power of the LLMs, which is where you have agentic AI. Now people harness the power of AI in two ways. First of all, you had an agent. So we also had the journey a few years back, we said we have multiple agents sprinkled across our source to pay product. You can summarize a contract, you can generate an RFT. So that is an agent. Now, when you talk about agentic, an agentic flow, you do not start with a point, you start with the end outcome.

7:59What's the business outcome you really want to have? So I need to buy something. Now, it kind of works backwards and lines up all the different agents which are required, which is how real life work happens. You would identify identify what has to be bought over there, you identify the suppliers, you will benchmark it over there, you will create an RFP. But agentic AI does that. So agents collaborate, they decide which path really to take but towards a final outcome. So agentic is a layer on top of the powerful LLMs, the frontier modules. And on top of that, you have agents doing multiple steps in collaboration towards a final outcome.

8:39And that's what is agentic. And people often confuse agentic with just a point agent and point agent is not agentic over there, correct. So an agentic flow as an analogy, plan my entire vacation, it understands your requirements, your family's needs and what are the timings you are available and then it searches and it identifies and it books and it's ready, your vacation package is ready, correct. So it's very similar to that as opposed to tell me what's a good hotel in this location, correct. That's a point agent yeah so i think with point agent a lot of agent sprawl which is happening but because of that people are missing the big roi of ai so you have to start with an outcome what's the end outcome and then work backwards to that and so when we talk more about agentic it's the um the coming together of all of the maybe point agents but doing different tasks but what makes it truly agentic is a system that brings all those together so that to you, it's kind of seamless.

9:39You don't know that there's multiple point agents going and doing tasks because you're just seeing kind of the outcome and it's all these point agents working together and sharing information with each other to drive a certain outcome. Yeah, I think that's very important. Agents talking to other agents, correct? So they'll exchange information in a certain way over there, correct? Which the other agent can understand And it can also pick which agents are really required to get to the final outcome. But for you, I rightly said, Phil, correct, it is transparent over that. It will come out as a senior flow.

10:12You don't see the agents really handing off one to the other at the back. Why is this cycle different for procurement leaders? As compared to? Yeah, when we talk about, so as we think about digitization, you know, the different waves of digitization of procurement. And, you know, as we've gone from just having access to technology to from an efficiency play to, you know, what a genetic may now enable for a procurement leader and how that differs from previous waves, if you will, of digitization. So I think if you look at the waves, correct, which have happened in my lifetime, you had, you know, obviously, you know, the mobile or the cell phone, and then you had, you know, cloud, and then you had, you know, before that, desktop computers, and so on.

10:59I think AI is very different in scale compared to any of those. And to put it in perspective, what I feel, like what we feel at Zykes, this is, you know, it's once in a few centuries kind of revolution which really happens. So the closest analogy is the industrial revolution. Industrial revolution, you know, you had, you know, once we had the machines coming in, you had, you know, the assembly lines coming in, it completely transformed the economy and people were doing different kind of work after that, people were consuming different kind of stuff. So it remade economies, it remade industries and so on and what the industrial revolution did in copying what the human muscle really does, is what Gen AI is doing by doing what the brain really does, a lot of it, where it can reason a lot over there.

11:58So in terms of what it can do, you can take a person out of college with two years experience and make him as good a category manager as someone with 40 years experience managing that category if you have the right agentic flow so it's going to it's going to you know dramatically change over there so i think you know i always feel that uh people are significantly underestimating in terms of you know how ai will really remake not just procurement but it'll remake the whole world also um and do you find that to be the case you know when you think about underestimating and also what do we overestimate you have so different conversations with procurement leaders every day um you know just kind of doubling down on some of those things first of all that that that perhaps the people you talk to are underestimating the impact of ai on from but then on the flip side are the things that they overestimate what they think ai can do or should i say agentic ai can do sure sure so first i'll talk about you know overestimating over there like it's it's the usual you cannot put technology on a broken process we had you know we had agentic ai for intake and a customer put it on a process which says you need to get an approval for everything above$500.

13:06So AI is going to do what you want it to do. So if you have a broken process, so you have to not just put AI as a band-aid on top of it. You need to rethink the entire process in terms of how it can really work. So in fact, AI can possibly decide over there what should be the thresholds based on how it has been purchased in the past, what's the category and so on. It can be very dynamic over there, but our procurement leader willing to take that risk so one is you know you cannot put technology on a broken process bad data ai can still handle that over there but a broken process is something which uh you know you cannot hide from yeah you cannot hide from a bad process exactly exactly so i think the process is very important because people don't rethink the process in fact if you ask ai it will redesign the process for you right it will redesign it will tell you this is what the process should be based on your data which you have.

14:00Now coming to the other part over there, what they are underestimating and I think I was talking to someone and he mentioned that humans are creatures of habit, they change but they take time to change over there. So I think people are underestimating how much AI can do as of today and I'll just give an example. At Zycus over the last two years using AI, we have increased our efficiency by almost 45 to 50%. Which means we have half the team doing significantly more work, more products, more sales and so on. Now we are developing products, writing code, implementing difficult, complex enterprise software.

14:48If we can get that efficiency, there is no reason procurement cannot get that efficiency. I'm just using efficiency as one parameter. Obviously, it'll free up people to do a lot more strategic work which is there. So I think people are underestimating in terms of how AI will significantly remake the entire operating model. What people are doing today, they will not be doing tomorrow. People are not going to do accounts payable, people are not going to do tactical buying. People are not even going to do supplier risk assessments. They will be out working over there with the suppliers, building relationship with the stakeholders, making judgment calls about the future which AI by the way cannot do.

15:31AI cannot for its life over there predict what's going to really happen in the future. And humans have that instinct, that judgment by talking to people and they really can do that over there. So I think people are underestimating what kind of efficiency gains you get and how will it really remake the operating model. But I think it's a normal cycle. You have some companies that are leading. You have a bell curve over there. So I think over the next 10 years, I see that huge change happening within procurement. It's interesting you just kind of put a time frame on it. So I was going to put you on the spot and ask you for a time frame because I feel like at this point, it's not a technology problem for procurement leaders.

16:14It's a change management problem. The technology exists and technology is continuing to exist, which will enable you to do a lot of things that you couldn't do two or three years ago. And so it's as much a change issue. I think one of the fears that I hear a lot is control, losing control to an agent. and uh you know i'd love your perspective on um you know how do we govern agentic ai or how can we be thinking about governing so that so that you know the agentic um uh ai the agents and the system is doing the things that we wanted to do is doing the outcomes that we want but there's trust that it's not going to do more than that or that you know we still have that human and whether you know, where in that process that human comes.

17:05Do you have any perspectives on how we can be thinking about that to help us overcome some of those fears and drive the change faster? Sure, sure, sure. I think there'll be two layers in where the guardrails, as I call it, will come in, correct? There may be three layers, correct? One is in terms of, you know, the software. The second is in terms of policy guardrails for AI agents. And the third is, you know, the human in the loop, correct? But the first, I'd say one thing we have to realize is that obviously procurement is very close to finance, correct? You know, you're posting into a financial system, you know, purchase orders, invoices, correct?

17:38And you cannot even get a few cents wrong over there, correct? It has to be absolutely accurate. So what you have is at the bottom, there has to be a deterministic layer. So for example, at Zycus, a current source to put software now becomes a system of record. So you have the system of record, you have all the rules over there, you have the delegation of authority, the workflow, So all of that stays in that system of record over there. So that has to be really deterministic. It gates everything. So effectively, it blocks AI from doing something really wrong. The second thing is you have, when we build, so our layers which we have is we have a, say, SoStupe software, then you have a Merlin agentic AI on top.

18:15So in the agentic AI layer, you can put for every agentic flow, you can put guardrails and policies, correct? Do this, don't do this, and so on, correct? So you can put a lot of guardrails within that order. So that also will keep the agent within certain boundary over there. So for example, if we get supplier responses over there for risk, do you want to only use a supplier response or do you want to use external? External is fine. But if you're getting a price from a supplier, you don't want AI to really round it off or make any changes over there. So those are the guardrails which you put. Now, the third is we talk about autonomous agentic flows.

18:56But every flow can be with human in the loop at any point. So you can decide, okay, this is something which is very tactical for me to buy. And I've seen AI do this again and again, go ahead and do it autonomously. In the other case, you'll say, okay, this is the first time I'm doing this kind of services, RFP for, let's say, cybersecurity services, just involve me over there in the RFP. And when we're doing the final evaluation over there. So on agentic sourcing, you can decide at what point the human in the loop comes in over there. So I think what we do is we put guardrails, one at the system level, which is a system of record.

19:31Second is that the agentic thing, and then allowing that control to the human to decide whether it should run autonomously or not run autonomously. As part of that control, how much transparency does a human have into what an agent did, you know kind of audit trail or and i think it's as much of a to go back retroactively and understand okay these are the steps that were followed and then that helps drive the trust to know that okay those steps that were taken by an agent are the same steps that i would have taken as a human for example sure sure so i think um what you have is one is you have the audit trail as i mentioned in the system of record correct so everything is recorded over there you know what workflow it went through and so on.

20:14And for every agent also, we have an audit trail over there. So the audit trail says that this is what the agent really did over there. So that audit trail also helps people in understanding what is done. What cannot be done is you cannot tell why an agent did something. So why did the agent recommend this particular supplier over there? can you know that often explainability of ai is not so easy over there but yeah at least what ai gave you know it can definitely say okay i you know the supplier told me that you know this is the price and i counted over there that you know the other supplier is giving at a lower price we need to give it a lower price over there i need to get you know early payment discounts for it and so on so you can see the entire trail of that in terms of what the agent really did over there correct But you still need to have that audit trail in your system of records.

21:06If you try to build an agentic AI without a system of record, you will fail because AI will go wrong in certain cases if you do not have that layer before the ERP, which is really guarding everything. Yeah. Interesting. So I want to pivot just the last few minutes here to talk a little bit more about the series and lead into the rest of the series. And the question that some listeners may be thinking is, why is a technology company like Zycus supporting and making possible what's a pretty big undertaking in bringing these stories together that deliberately isn't going to focus on technology itself?

21:43It's more about the journey. Yeah, I think first, as I mentioned, we strongly believe this is a revolution. So if it's a revolution, as a revolutionary, you need to really rally everyone around it. So we need to educate everyone about what is possible and what is not. Now, obviously, we are a vendor. So we're going to pitch over there what we have in Agentee AI, which is why I feel we're working with the art of procurement so that you get practitioners. And they don't talk about technology at all, but they would talk about how do they view AI? What are the risks of AI they see? What are the benefits?

22:17What are the challenges? What are the processes of trying out AI within that organization over there? So I think one is obviously education, because we have seen when we talk to CPOs, they're very, very curious. In fact, some CPOs have formed groups among themselves to say, okay, what are you doing about AI? So people are very curious about how everybody knows that something is going to happen. Everybody knows that it's going to change. So I think people want to hear from other practitioners in terms of what is it they're trying, what has worked, what has failed, and how do they go about it and so on.

22:55Yeah, I want to double down on that a little bit. And anyone who's listened to the podcast recently probably heard me talk about this a lot. But the most, the procurement leadership that I see who are most likely to succeed are those that are the most curious and those that recognize in this current situation, they don't have all the answers. And so everybody is kind of on this journey of discovery. And as much as we can bring together, everybody from across our community to be on that journey together, the better it is for everybody. Because, you know, as I alluded to earlier, I think earlier, certainly in my procurement career, and as I mentioned, it's been 20, 25 years plus now, procurement was a certain thing.

23:36And it was just all about how do we optimize this certain thing that we do. And that's what all the conversation was around. But now if we kind of continue to think like that, that really constrains our thinking. And it constrains, you know, how we think about the value proposition, how we think about the business and operating model that surrounds it, how we think about where we invest, how we think about skill development and making sure the right skills are in place for when we do have agents doing a lot of the work for us. And those are all big unknowns at the moment. So we're grateful for the opportunity to bring together some of these stories so that we can help help peers or help procurement leaders learn from their peers.

24:14I think that's really important. That's definitely very important. I think people need to know from others, you know, what's really, how are they going about it? Because again, like you said, it will change the operating model. How do people visualize what's going to change when practitioners come and say, this is what we visualize it, what's really going to change. But these are the steps we have taken now. And that's very important. Yeah. And I'm sure you have a lot of these conversations in private, you know, with CPOs as you're talking both to clients and to those that you are getting to know.

24:47Are there any themes that you would pull out of those conversations or, you know, kind of where you see right now the mindset of a forward thinking procurement leader based on all these different conversations that you have? That's a good point. And by the way, we used AI for analyzing that. What we've done is in all the conversations which everyone has across the organization with prospects and customers, and we put it in the data lake and then we have AI analyzing it. So what we identified is there are like a bell curve. There are three segments of procurement organization or procurement leaders.

25:26One is AI hungry. The second one is AI curious. Yeah. And the third one is AI resistant. Yeah. And in fact, the AI resistant over the last couple of years has shrunk to almost 15 to 20%. It's only in regulated industries like healthcare and so on. But now coming to AI hungry, what we have seen the AI hungry ones is they are looking to do something in AI. And it often stems from the board or the CEO. So the board has said you have to do about something about AI. What are you doing? and the CEO then asks the team or creates a fund over there in terms of let's try out something on the AI. So the AI hungry are the ones who are, they can do this frontline explorers over there.

26:13So they're trying out stuff. So they have the management, the approval, the funds and the backing over there to try on AI, whether they're doing it internally, whether they're buying something from outside but they are the ones who are really experimenting and trying with ai right and then you have the ai curious so the ai curious they know that ai is going to change something over there but they are often waiting to figure out what's the roi so obviously there's money to be invested over there so the first category is given money saying do something with ai we have you know gotten this uh initiatives over there so the ai curious are looking at what could be the roi for ai so a recurring theme which is happening both in the ai hungry and ai curious is what is the ai areas we can focus on which can give us good roi yeah so there is a shift now which has happened because obviously uh ai economics also come into play you're going to spend on the llms and so on so what are those areas so we have an access over there correct which is saying what is you know the impact and the practicality correct for ai so i think people are trying to figure out what are the areas they can use ai for which is you know a high impact you know lower risk and so on over there so that's a theme which we are seeing you know uh across is ai hungry and ai curious and ai resistant i think they will change you know they always have a bell curve over time that's a 10-year period i talk about it where people get converted that yes it's applicable for me also correct so but that that's a very thin slice as of yeah that's that's fascinating I think at some point in the future, I'd love to unpack what those who are, you know, AI hungry and AI curious are doing.

28:00And we can draw a lot of lessons, I'm sure, from some of those observations. Now, as we wrap up, I do have one last question. And this is a question that we're going to ask throughout the series. The question is, if we were having, you and I were speaking three years from now, what do you think that procurement organizations, What's one thing that you think procurement organizations will be doing that feels today like it's almost unimaginable? Yeah. So let me tell you what people will not be doing three years from now, which is unimaginable. So people will not be processing invoices. They will not be processing tactical bias.

28:41They would not be really hiring category managers with decades of experience over there. Correct. All that will be delegated to AI. So, you know, AI will do a lot of the tactical work for you over there. Correct. Now the challenge over there is, you know, if a lot of the work which you're doing currently is going to go to AI, correct, your team members would be freed up. Now either you have a smaller team or you leverage the same team for different kind of work. So I think that's where the challenge would really lie. Correct. So once you free people up from tactical work, correct, where are they going to spend that time like are they going to spend time on new product introduction you know working early in the cycle like you mentioned correct maybe the value proposition of procurement to the rest of the organization changes correct it's not you know you know i'm going to you know manage supply risk or i'm going to you know source for you and so on correct it could really significantly change over there now so i definitely see that a lot of the work will be offloaded to agents and i talked about that percentage to start with 10 percent 20 30 percent it'll go up to 50 percent maybe in three years over there correct now you have 50 percent more bandwidth how are you going to leverage that correct so i think that's what the change so that other part i don't know it depends on organization to organization how you know people use human creativity judgment you know intuition over there correct in relationship building correct how are they really going to change on that so i think it's it's people will have to start thinking in terms of how the operating model changes and what's your value proposition to the enterprise.

30:16Yeah, I'm actually, I love to hear that because for me, the biggest question of today is what's next, you know, and we have to think about, you know, there's obviously the journey to get there, which is the journey that all these organizations are on, as you say, some are more ahead than others, but then you have to prepare now for what's going to come later, because otherwise you'll get to the later and then be looking around thinking, okay, now what's next? Now what do we do? And if you're thinking about it, then it's too late because all those efficiencies that you have taken out of the system will result, in my opinion, in job losses because you are now having agentic systems doing the work.

30:58But if you prepare for that as you go on the journey, now you have this open capacity and if you're ready for it, that's where you can really drive exponential ROI. But you have to be preparing for that before you get to the point where you're able to go and use that bandwidth. No, very true. And I'd like to make another point because what happens is you mentioned people are experimenting and trying out stuff over there. But people are able to do that first. We get a competitive advantage. That happens in every wave. Yeah. And one of the things I saw on your podcast also, you often talk about what failed over there.

31:34Now, especially for AI, it's very important for companies to fail. And I'll mention why it's important. There was a book, I forget, by a Harvard professor, correct? What are the right ways to fail over there? If you fail doing some simple stuff or collaboration, it's a wrong kind of failure. But if you fail in exploring a new area, correct? So Christopher Columbus also had a failure and had a success after that. So AI is new. So even at Zycus, we have had failures as we went on the agent TKI. Two years ago, we said we'll build a lot of point agents over there. We got it out. We didn't see the ROI from customers.

32:14We pivoted very quickly over there. But if you had not done that, we have not been at version five where we are currently over there. So the same thing is true for the organizations in this AI. And the good thing about AI is you don't have to buy a huge software over there. You can start with one flow, one agent over there. Just start with that and you learn from that in terms of what's happening over there. So I think that adventurous spirit is what we see over there of that AI hungry companies really doing it. And we'll see more and more companies getting to that for sure. Well, I'm excited to see how we continue to kind of evolve the conversation, bring lots of stories in through the Egentic Procurement Diaries podcast.

32:51For anyone who's listened to today, just a heads up, we will be publishing the podcast likely every couple of weeks. We're here in the Art of Procurement feed as you've listened today, also on the Art of Procurement website. So you can always check it out at artofprocurement.com, see the latest episode. And we'll kind of see what we uncover and explore and then share a lot of those findings from CPOs in their own words around how they're approaching and thinking about the use of agentic AI. So thank you so much. Actually, I really appreciate you joining me on the show today. And I look forward to speaking to you again soon.

33:28Yeah, thanks, Phil. It was great talking to you. And we also look forward to what our CPU is thinking about how they go about the journey over there. Thank you. Thank you.

33:57listen to your podcasts. Since 2015, we've built the world's largest free resource for procurement professionals looking to elevate their impact. Our resources span podcasts like this, videos, blog posts, papers, and events. To join us on the inside and to ensure you never miss an episode, a webinar, an event, or a post, please do subscribe to our weekly newsletter This Week in Procurement. You can do that at artofprocurement.com slash subscribe. That's artofprocurement.com slash subscribe. Thank you for joining us today and I look forward to seeing you again soon.