00:00:00 Introduction
00:02:19 Has AI replaced S&OP and IBP?
00:05:15 Why traditional IBP creates little value
00:14:14 What counts as a real decision?
00:16:15 Which decisions should remain human?
00:20:23 What does an autonomous supply chain look like?
00:22:53 How should autonomy handle major shocks?
00:25:50 From consensus to value optimization
00:30:16 Scarcity and economic optimization
00:36:26 Consultant AI jargon: real or invented?
00:44:09 Can AI orchestrate useful scenarios?
00:51:07 Are LLMs and digital twins enough?
00:56:02 Why a paradigm shift matters
00:58:36 Conclusion
Summary
Conor Doherty interviews Joannes Vermorel (CEO and Founder of Lokad) about whether AI has truly made Integrated Business Planning obsolete. They examine why traditional S&OP and IBP often remain disconnected from economic decisions, what genuine supply chain autonomy requires, and where LLMs can—and cannot—help. The episode also tests real vendor claims against invented AI jargon before contrasting digital twins and agentic tools with Lokad’s quantitative, economically driven approach.
Full Transcript
Conor Doherty: Well, Joannes and I have just finished the ‘Autonomous IBP: Is There Any IBP Left?’ discussion. We covered a lot of ground. We discussed what would actually make a supply chain autonomous in terms of its decision-making software. Are LLMs sufficient to do that? They might be necessary, but are they sufficient? Maybe they’re not even necessary. And fundamentally, we had a good laugh discussing [bleep] or real where I read Joannes some consultant and vendor claims and asked him to guess it. Is this [bleep] or real? So, yeah, stick around, enjoy the show. Some laughs ahead of you. And as always, if you want to learn more, you want to talk to Joannes and me, reach out to us on LinkedIn or send us an email at contact@lokad.com. So, Joannes, we’re here to discuss an article that I found earlier this week. I sent it to you. We had a bit of a laugh as we’ll get into, but that article was from EY, no offense, and it was called Autonomous IBP, the future of AI-driven Integrated Business Planning. Now, we’re not here just to talk about that. The discussion that we had led me to jump down something of a rabbit hole where I was just consuming quite a lot of consultant and vendor material and they all sort of converged on a similar theme that being that agentic AI and autonomous decision-making systems are no longer the future. They are in fact already here. Now we’ve been saying that for years obviously we’ll get we’ll get into that. We’re not going to bury the lead. We’ll get into that later. But the whole point is industry-wide now and I’m talking about not only EY, SAS, SAP, Kinaxis, Blue Yonder, o9, Gartner, McKinsey, I can go on. They’re either explicitly or implicitly supporting the idea that traditional planning concepts like an like S&OP and particularly obviously IBP, they’ve sort of had their day, but we’ve all moved past that now. AI has replaced that. So my first question to you is listening to that and bearing in mind everything you’ve always thought and written about has AI in 2026 effectively replaced traditional planning systems like S&OP and IBP?
Joannes Vermorel: No because you could do that effectively and Lokad has been doing that for a decade without what people call AI. So that’s the crux is that the automation of those supply chain changes do not require large language models. Yeah, that’s what most people are talking about and I think that’s the whole point that those things are missing is that you absolutely don’t need that. So you can automate all of that and it has been possible for more than a decade without LLM. LLMs let you do a lot of things you know faster and nicer. That is true. But the problem is that you have to understand that the sort of technologies that you need to deliver the complete end-to-end automation doesn’t require LLMs at all. LLM just becomes like the cherry on the cake to make that even faster and easier. But if you do not accept the sort of paradigm that comes along with the automation, you’re never going to be able to layer on top this those I would say LLM productivity benefits and that’s here the problem is that I think in this EY report they are taking the problem from completely the wrong angle. They think, “Oh, first LLM, let’s have the thing do everything.” And then at the very end and then they essentially they think of the LLM as a virtual employee that would just automate the things the way an employee is doing and I think that is complete nonsense. This will not really scale LLMs requires I mean for any kind of modest task we’re talking of hundreds of thousands and millions of tokens. I mean we are talking of stuff that is going to be expensive and it’s not so you a company that has a supply chain we are talking of tens of thousands of decisions to be revisited daily and if not millions actually so this thing is not going to be done with an LLM it’s you will require an a numerical recipe which can be co-authored with an LLM as a coding agent excellent but you see the premise is you need to think of this numerical recipe that will just automate and then we can talk about co-authorship between some something like a supply chain scientist so the human who carries the responsibility for this automation plus a coding agent that would just boost the productivity of this person
Conor Doherty: Well before I get into some of the claims not only from the article because and I will show receipts because I don’t want people to think that we’re misrepresenting information here. But before we get into that, a key point that you just made, however you arrived at the conclusion. So the conclusion is okay traditional planning systems like S&OP and then IBP, the successor. However you arrive at the conclusion that might be outdated. What do you think has caused that shift now? Because you’re saying, “Oh, well, you could have done this more than a decade ago without LLMs, but now all of a sudden the market, and I mean some of the biggest players in the space are saying, actually, no, no, no. Yeah, it is too late for that now.”
Joannes Vermorel: Yeah. I mean, you see, the thing is S&OP and IBP as a rule of thumb for the quasi totality of companies produce exactly nothing of substance. Nothing. So, it’s a pure bureaucratic undertaking. Pure. So essentially it’s people having meetings and producing Excel spreadsheets en masse and floating around Excel spreadsheets which are completely and rightfully ignored by anybody who has any resource to allocate for anything in the company. So and people realize that if you just replace the employees doing this sort of bureaucratic nonsense by AIs nothing bad happens and it’s cheaper. Yes, absolutely. And the reason if you if you replace bureaucratic nonsense made by humans by the exact same bureaucratic nonsense made by AI, you know what? It’s just going to be cheaper and because it’s nonsense either way, the fact that your LLM might hallucinate as much as you want, it doesn’t matter because anyway, the stuff that was originally produced by human was irrelevant anyway. And I think that’s the sort of point is that people I think many people especially people who have been working on the S&OP and IBP process I mean the typical situation that they don’t realize how disconnected this process is from actual allocation of economic resources in their company. So there will be tons of reports tons of projections being produced meetings happening things will be discussed at length etc. People go in circles and at the end of the day it will be other people in the organization that actually allocate resources and they will just do something completely different and that’s it and that’s it and we had this charade going on for a long time for many companies decades and now people realize that well it can be automated and yes and it works by the way it works incredibly well because again it whatever flaws your LLM might have it’s inconsequential because what you’re doing is inconsequential so that’s you know it’s faster inconsequential yes it’s whatever you know you can hallucinate just hallucinate those excel spreadsheets and that’s it that’s it that will not make a dent in your in your business again I wouldn’t I would never reveal the names and again this is an anecdote so people can believe it or not but we do know some senior S&OP executives who work within nominally the S&OP framework but a lot of it is workarounds that sort of violate the entire concept of consent. It’s like on paper, yeah, yeah, we sit down, we have a meeting, we decide, but then in the background, one of our data scientists who’s a whiz with Excel kind of does a bootleg version of Lokad. And again, I know I’m thinking of three examples. We both know them. We’ve had discussions with them. Now, again, I’m not saying that’s representative of everybody, but I’m saying at a certain level, and these are very, very big companies, maintaining the S&OP or IBP framework becomes difficult. And in fact, that brings me to the literally the opening of the article and I want your reaction to it because and I am reading from again this is EY’s own article title autonomous IBP the future of AI-driven integrated business planning. We’ll put up a screenshot. The opening is autonomous IBP is an AI-driven approach to integrated business planning that enables real time continuous decision- making across supply chain finance and operations. They add at the core of traditional IBP is a key assumption. Decision-making can be effectively orchestrated within a monthly cycle. That assumption no longer holds. Now Joannes, if you if someone told me you wrote that, okay, obviously you wouldn’t use the words IBP, but if you that comment about traditional IBP, presumably you agree with that. Not really. You see again the people under the mistaken understanding that IBP has anything to do with decisions. It does not. It does not. IBP decides nothing. That’s my main criticism. It’s a it’s a self-contained bureaucratic undertaking that is completely immune to anything that happens in the world and that is completely disconnected from whatever trajectory your company’s in. You see IBP only focuses on forecasting. It does not for it does not you see what people call decision they would call oh we revise the forecast up or down as a decision in my book it’s not a decision I can decide that you know the sales next year for Lokad are $100 billion I decide that and I decide oh no it’s a little bit excessive let’s go back to something more reasonable I can change this so-called decision 20 times; it still make no difference or what is h actually happening at Lokad. So you see when you just you’re just steering the forecast up or down etc. You are essentially toying with numerical artifacts. Those numbers are not real. They have they have no consequence almost except making maybe your finance guy panic if you if you produce really crappy projections whatever. But you see that’s when we say that’s the problem is that I think those in this those papers they use the term decisions but they use it in a way that is completely different from what Lokad actually says loc what we say when we say a decision we mean an irreversible use of resources. I decide to move a piece of inventory that I have to for example to my factory and then I will consume this piece of inventory to produce something. It’s done. The inventory is gone. I cannot the piece of inventory is gone. I cannot get it back. The production did happen. It has been transformed. There is no way back. If I decide at a given price point that I sell one unit to a client as opposed to raise the price and hold it for longer, then once it is sold, it is sold. The client now has the stuff and it is their property. I cannot get the stuff back. You see, and thus the pricing decision is final. And that’s here the problem is that what they are they don’t they are very vague on what sort of decision they are talking about. And the problem is that when you are in the S&OP or IBP paradigm, we are mostly discussing numerical artifacts stuff that is just you know an abstraction and not real. If I say you have a budget or to spend well as long as this budget is not actually spent, it’s not real. I can move the budget up down revise five times a month. Again, the only thing that is real is what has actually been purchased within the constraint. The fact that the budget limit is going up or down or whatever. It is fundamentally something that is you know optics. It is it is it is a numerical artifact and that’s my main criticism is that they are using this the jargon are saying you can you can have autonomous and you can you can have those decisions as if as if they were stakes and my point is no that’s the main problem with S&OP IBP the paradigm itself is that fundamentally the decisions that the so-called decisions are like completely devoid of substance.
Conor Doherty: Well, to be fair, within again that article and I and I will get to others, but just within that article, they do actually contrast what the traditional IBP framework is and then the proposed autonomous IBP framework. So traditional IBP framework a monthly sequential process has human-driven data collection and reconciliation has the features excuse me teams preparing for meetings as the primary decision forum focus on alignment and consensus decisions made at fixed intervals during meetings
Joannes Vermorel: But no decision you see that’s a that’s a my take again in the classic form there is no decisions whatsoever not in the economic
Conor Doherty: We have consensus
Joannes Vermorel: But not necessarily people will yes they will reach agreement on numerical artifact but they will not reach economic decision. You see that’s a very big difference. We can reach an agreement for example we can discuss together and decide that next year we think the world population will not reach 10 billion humans. We have reached an agreement. We have a number but it is it has no consequence on the business of Lokad or whatever you see and that’s I think the fallacy is that people say we have reached a decision no we haven’t reached a decision we have reached an agreement on a number about the future
Conor Doherty: But not what we will do with that information exactly and some the people who will do something may actually completely disregard this number yeah it’s we have an agreement on that
Joannes Vermorel: Oh yeah fine I’m still purchasing this and that and that or not purchasing this and that and that you your the rest is still continuing. I would say the other processes are still continuing without being I would say bound by this agreement.
Conor Doherty: Well, so again there are a couple of terms to introduce here. The idea of like above the line and below the line and that’s that language is sort of used in S&OP and IBP circles to describe higher level of priority and a lower level of priority and if we were just to take that language and say what decisions and when we and I mean in our meaning of the word decisions what decisions would be sort of immediately automated sort of below the line and what would remain above the line and above the line would be I don’t know enormous allocations of resources to let’s expand into a new territory. I mean, that’s presumably something that will be okay. We need people in the loop for that. Yes. Okay. Yes. But everything south of that is automatable. We don’t need people in the loop for that.
Joannes Vermorel: No, I mean, we also need to have people to u think how we even frame the economics. You know, what does quality of service even mean? No, that’s a question the that’s a question that has no I would say unambiguous answer. You need to think it through and see how the quantification will unfold. So you will in the end how you have to decide in the end in dollars what is the cost of this service? If you don’t do that you’ve not actually started the economic reasoning. So you’re not optimizing anything and thus you’re leaving a lot of money on the table. So you need to be able to say okay what is the cost of this service and the thing is that you can think of that in many different ways. You can decide that some segments are for some reasons much more important than other segments. You might think that there are some seasons maybe before Christmas that are much more critical time critical than others. You can decide that some segments of products are much more critical than others etc. So there is many dimensions and the AI can you know provide a lot of guidance on that but fundamentally it’s we are not there yet to have anything that where an LLM can provide you know reliable judgment calls for high-level stakes like that. So you will still need to have that and where it’s you see this above the line below the line it’s slightly different is that answering these questions will then in turn shape the numerical recipe that will then impact all the daily mundane stuff. You see, so what I’m saying is that if they were really thinking, you know, the next generation of IBP and S&OP, they would say people need to get together to think about how do we frame the problem? Yeah. How do we what are like the key factors that needs to be included in the calculation etc. How what are the sort of things that needs to be guess guesstimated the meta classes of information as and whether for example we need to invest or not to acquire some missing information because the you know from the get-go what you have is all the transactional information is readily available in your business systems. So that’s a that’s a given already. And now if you want to have thing to have data that is not in your business systems that can be done but that there is a cost. The question is that is are the expected benefits justifying those investments or not? Those would be interesting discussion where humans still very much need to be involved because realistically an AI will not be able to operate anywhere close to autonomy on such topics.
Conor Doherty: That’s part so the way we would use I guess autonomous supply and we’re focusing on supply chain. The way we would describe autonomous decision- making in supply chain I think is quite different to how the rest of the market typically does it. You’ve given some of the insight there which is you know focusing on the like quantifying at least discursively how much do I value Joannes as a customer and serving him one unit versus Conor as a customer and not serving him one unit etc. What is the how what do I what do I estimate as the financial impact of this service or disservice? That’s part of it. That’s sort of crafting or goes into crafting the decision-making algorithm. But what does an autonomous supply chain look like to you once that goes into action? Like because you have people in a company, what does it look like after that?
Joannes Vermorel: It looks like the essentially if you are truly if you achieve a satisfying degree of autonomy your supply chain can operate for three months without any almost any white-collar workers. If there is no spectacular shocks, yes, there is if there is nothing spectacular that is happening on your market, you can just remove all you can send all white-collar people on vacations for three months and the thing will be just running just fine. So that means that what needs to be purchased from your suppliers will be purchased. What needs to be produced and so your production schedule and everything will be scheduled and produced. The allocation of inventory inside your chain will happen again automatically. The if you face situations where you have like shortages, you will spread your limited capacity of stuff reasonably across your clients including if your clients are trying to game the system and saying, “Oh, I need 1 million units. Just give me the last 10 that you have.” No, sorry. We are just going to spread that slightly more fairly across I would say trusted u customers etc so and again that’s that is what is true autonomy is that at least for me is when all those mundane decisions are taken autonomously I would say in a completely unattended way
Conor Doherty: Well, my follow-up to that then is because you’ve said previously that fundamentally a shock is still just a I’m paraphrasing any shock even a systemic shock is simply more data that enters like numbers ones and zeros etc that enter into the that enter into transactional data and then your numerical recipe adapts to that. Well then by that logic what’s the difference between like minor disruptions and then systemic disruptions in terms of maintaining autonomy like what’s an exception within that framework.
Joannes Vermorel: No I think you see the thing is that when something really changes fundamentally in your market you cannot just tweak the numerical recipe. It needs to be I would say deeply re-engineered you know
Conor Doherty: Do you have any examples in mind of that?
Joannes Vermorel: I mean COVID presumably is one. I mean, for example, during COVID what happened is that we had clients that were where before the lockdowns 80% of their business was brick-and-mortar. Yeah. Yeah. After the lockdowns 80% was e-commerce. So are you do you are do you think your supply chain and everything that goes with brick-and-mortar first or e-commerce first and obviously the swap happened in like 12 months. So when you have something that’s where you went to e-commerce first while you were just you know brick-and-mortar-first the consequences are cascading all over the place. You cannot just say adjust a few constants here and there. No, the whole thing need to be completely rethought and you have to even re-imagine what purpose do you have for your traditional stores? Do you want to do showrooming? Do you want to have to let people do their pickups at the store with secure facilities, etc., etc. I mean, it’s plenty of questions. And that’s again that is the sort of change that a numerical recipe cannot accommodate by just tweaking a few constants or having you know an LLM just doing some modest follow-up just for numerical correctness. It goes way deeper. So what I’m saying is that for the day-to-day mundane stuff especially mundane chaos you know one supplier for example is getting a little bit late there is a spike on one type of products but every single year you have multiple product categories that spike it’s nothing out of the ordinary all of that should be dealt with you know by I would say by the numerical recipe that’s like the ambient noise But if you have something more fundamental where your market changes in nature then it’s something that fundamentally you need to have white-collar people potentially equipped with LLMs to just re-engineer the numerical recipes.
Conor Doherty: One of the things that was common throughout not only the EY article this now to sort of broaden the discussion a bit more. One of the things that was common was a move from the language of consensus towards the idea of value optimization. That is a phrase I saw over and over again. That specific pairing value optimization. Now my thought when reading that was and this is all in the concept in the context of like level up your S&OP, level up your IBP. Yes. And yeah optimize value and create great value optimization. But the thing is value as a term can mean a lot of alignment consensus everyone getting along that’s a value I value a frictionless environment yes money is also a value so in this context what is value optimization to you in autonomous supply chain
Joannes Vermorel: The, I mean, Lokad would approach that from economic perspective obviously here the problem is that both IBP and S&OP non-economic perspectives. They are not they are not they are a machine a process to generate an agreement on projections. You know that there are mechanism by which people will come to an agreement on numerical artifacts. That’s it. That’s it. And that’s why you see the problem is that you can say oh it would be nice if this thing was actually creating value. I agree but can it?
Conor Doherty: So just to jump in for people who might not have heard that when you say numerical artifacts you’re talking about like intermediate KPIs like oh we forecast a forecast
Joannes Vermorel: A forecast is a numerical artifact you know I can say next year we forecast we I forecast we are going to sell 1 million units of this. Okay. The next minute I say 1 million 100,000 and in one hour from now I change again I say 900,000 units. Well, I again it’s a projection. It’s not real. I can just say whatever I want. I can revise that as many times I want. Besides I would say, you know, it’s fundamentally that’s the thing with numerical artifacts is that it doesn’t change anything. Again, and numerical artifacts are all over the place. If I say for example, I look at a product and I say this safety stock, I am moving it at 99%. And then I think again I decrease the safety stock to 70%. And then again I can have this number you know fluctuate 20 times a day. It is inconsequential. But as soon as I pass a purchase order for 100 units to a given supplier then this is real because the money left the bank account of the company and it will have to be paid and the units very real are coming your way from this supplier and will take space in your warehouse. This is again real. So that’s really difference between numerical artifacts and decisions and here the people will say the when they say we want to do value optimization I mean I agree with your spirit yes I would even call it you know economic especially because again values I completely agree with your statement can be anything and the rest we are for example honesty is a value and yes and okay liability these real values. Yes. So here what we’re talking is really the economic perspective. So that means the long-term capacity for your company to generate the highest possible cash flows, you know, that’s what we’re talking about. So maximizing your profits but taking like the long view. So essentially you’re not sacrificing the future just to have a good quarter. That’s when we say economic optimization. That’s what we’re talking about. It’s something that is as opposed to let’s say a financial optimization that is really a little bit like the short-term arrangement so that you will boost your earnings in a definite period. The economic optimi optimization is really like the long-term view.
Conor Doherty: Well, underpinning the economic perspective is the idea of scarcity. Yes. You because again you have a finite resources you have finite resources that could be money that could be the inventory that you already possess. So in terms of how an autonomous supply chain decision-making system approaches scarcity, imagine I have I have a finite I have one unit. We’ll make it super simple. I have one unit. I can send that to any number of any number I can distribute in any number of ways. My value could be that I want to preserve margin and that might go over there. I want to protect a relationship with a strategic client. That means it no longer goes over there. It goes over there. Any number of ways this can be sliced and diced. Yes. How does at a high level, how does an autonomous supply chain decision-making system approach that? Because the way we use that language is quite different to the way we’re about to look at from others. So, how would how would our autonomous decision-making system approach that?
Joannes Vermorel: So, the way Lokad approaches that is that we approach scarcity as you know something completely fundamental. I mean again the very definition of economics without scarcity exactly that’s literally the very definition it’s a science of the resources that are scarce and that have multiple uses. So that’s literally scarcity is almost tautological with economics. It’s you’re talking about the exact same thing and for us that means that you need to have essentially you need to think of the alternative uses and their value. So when you whenever you do an allocation you would not say I earn money the question is that is it the allocation that earns you the most when considering all the alternatives. So the counterfactual is critical that’s you cannot just say this is a profitable use of this resource therefore I use therefore I use it. It is it like the maximally profitable use among all the alternatives and that’s how you know that you’re good. Now this is and again IBP and S&OP being non-economic perspectives those questions do not even exist they are not addressed and they are not even on the table it’s the counterfactuals are not even on the table it’s it do not exist the paradigm does not acknowledge this sort of concepts and that’s why I say when people say value optimization I mean the problem is that you the paradigm that you’re using just will not let you do that and it doesn’t matter if you have an AI or not because the problem would be exactly the same for a human even an incredibly smart human you cannot optimize if you have the economic value if you have like if you operate in a non-economic paradigm so you will be stuck and you will obviously if you if you force an AI to do that it will just hallucinate something because those LLMs are you know fine-tuned to be extremely agreeable. So if you I if you tell your LLM please square this circle it will it will lack knowledge yes boss will do my best and just proceed even if what you’re requesting is technically you know not feasible. So if I were to if I were to distill that if you are presented with phraseology be it a product or a service and whatever it sounds it has agentic in the title it has decision-making software in the title it has value optimization in the title unless the value that is being optimized for is of an economic perspective you should be very wary that for supply chain for supply chain it should be for supply chain excuse me because again that’s the context we’re discussing Excuse me. There. Yes. There are plenty of other domains where you don’t have this problem. You know it’s if again we the what makes supply chain very specific is that we are talking about you know the flow of material goods. It’s it has like a very strong physicality to it. Everything that you that you will be doing with this flow of material goods will require investment. Transport costs money. Storage costs money, production costs money, transformation costs money. And so that and those operations are essentially discrete because most supply chains are discrete. So it’s like a pallet. It’s boxes, it’s units, etc. So all of that makes it like super amenable to purely quantitative reasoning. This is I’m not claiming that whatever I’m saying is equivalent applies for I would say domains that are that have completely different properties. You know here we are talking of something where the for example the scope the scope of decisions defined in the way we define them which is allocation of resources. It’s very narrow. It’s like there is maybe like 20 different classes of decisions and this is it and we’re done and then every decision is essentially comes in a quantitative framework which is allocate zero unit, one unit, two unit. So it’s not like an open problem where you have like endless options. It is obviously it is infinite in the sense of 0 1 2 3 till infinity but in fact it’s if you eliminate the numbers that are ridiculously irrelevant you are facing a finite set of options. It’s large but it’s finite and it’s very constrained and that’s why it has those properties where I say you need to have this approach that quantifies the change.
Conor Doherty: All right. Well, I just asked you essentially about buzzwords. Yes. And that brings me happily to what I hope will be a recurring segment that I’m calling [bleep] or real. So based on all the reading that I did, I collated a list of agentic AI or autonomous AI supply chain products, capabilities, and descriptions from within the industry. Now, these are real terms used in the industry. However, I’ve also slipped in a few that I have completely made up. Now, I know that you are a student of comedy, particularly British comedy. So, you might be thinking, “Oh, I know where this is going. Conor has invented them all. Haha, that’s the joke. I give you my word. Some of what I will read to you is completely real. Now, I should also, just for our own protection, say that this segment reflects our subjective appreciation of comedy. It is not a comment on the efficacy of anyone else’s products. This is this is for the laughs. So, Joannes, again, [bleep] or real decision-based agentic supply chain orchestration?
Joannes Vermorel: Decision-based agentic orchestration. Wow, what a jargon.
Conor Doherty: Let me just parse it. Yeah, I’ll say it again. Decision-based agentic supply chain orchestration, [bleep] or real?
Joannes Vermorel: I would tend to think it’s probably real. It’s
Conor Doherty: Is that your final answer? I mean it’s
Joannes Vermorel: It feels like the sort of over-the-top [bleep] jargon that you would find that enterprise software vendor would produce but yes I it feels it feels it feels possible.
Conor Doherty: Yes it is real. Again, no judgment. Kinaxis describes their Maestro agents as advancing the customer journey from conversational AI to decision-based agentic supply chain orchestration. No judgments. Yes. No judgments. Yes. Round two, autonomous scenario orchestration. [bleep] or real? I mean, this is this is much less, you know, this zany. It’s not as zany as the first one. Yes, they will get zany. I’m not going to lie to you.
Joannes Vermorel: That’s plausible. And the thing is that it would even be plausible to do it. Conor Doherty: That’s why it’s tricky. That’s why I picked this. That’s why I picked this.
Joannes Vermorel: So, I don’t know if somebody’s claiming to do that, but clearly I would say with an LLM and you want to rearrange your scenarios and whatnot, this is this is something where the LLM would actually be quite good at. Okay. So I would say I would say probably real but I don’t know if a client is sorry an actor is actually using this terminology. Conor Doherty: It’s real EY lists autonomous scenario orchestration as one of their 10 autonomous IBP capabilities.
Joannes Vermorel: Yeah, you can do that with ChatGPT.
Conor Doherty: Yep. Okay. Round three. Some of Some of these are real. Zero-touch agentic value orchestration. [bleep] or real? Zero-touch agentic value orchestration.
Joannes Vermorel: Oh, that’s again you have like people in marketing that are just perverse if it’s real. I mean again it’s this sort of hyper-pedantic phrasing where I don’t know if people are trying to do the sort of buzzword bingo like oh yes with colleagues we say that we wanted to tick all the things. I would I would I would think it’s real. I would I mean it’s obviously complete [bleep] of content objective level at the objective level but is can it be found? I would say I would say probably yes.
Conor Doherty: Your [bleep] detector is good. It is [bleep] I made that one up. However, in my defense it is heavily inspired by an article from McKinsey where they describe no touch fully automated planning. But that was from 2018. No touch fully automated planning. So, Mackenzie’s been talking about it for a while. Next one. Inventory in sorry, inventory investment allocation agent. [bleep] or real?
Joannes Vermorel: Frankly, it’s for me it’s it sounds it sounds real. I think it’s especially the actors who are just plugging an LLM on their CRUD piece of software. It would work. You know, it’s just you just you again, I know that some actors are doing that. It’s you take a CRUD software. Yes. So, create read delete and you just have an LLM that will just manually go through like a human would do. It would work. It’s just an abysmal use of LLM tokens. It’s highly inefficient. It’s going to be also very inefficient in terms of economics. But yeah, I think it’s exactly the sort of nonsense that many of our peers would actually do. So I would say real.
Conor Doherty: Okay. Because of what you’ve just said, I’m going to say I swear to God and we’ll will sign a legal document confirming this. You did not know that I was going to read that. So he did not know these answers. It is real and it’s part of SAP’s planning assistant. Oh yeah. Which leverages two agents, exception management agent and inventory investment allocation agent. Yes. But you did not know that. That was you were just speculating.
Joannes Vermorel: And again it confirms the pattern. You take CRUD software. Yes. So SAP ERP. Yes. Your own CRUD and you take the exception management which is the most the thing that requires the most manpower but the reality is that if you had designed your software correctly you would not even need this module in the first place and now you’re adding an agent to deal with the noise that your system is generating. It’s pure nonsense but yes I think yes it’s really fit the sort of line of thinking of many companies.
Conor Doherty: The next two are my favorite. So join us. Agentic tradeoff adjudication with financial-constraint harmonization. Is that [bleep] or real?
Joannes Vermorel: Ah, it feels like I don’t know. It feels like [bleep] Over the top. Over the top for marketers. I don’t know if someone dared to have to have that.
Conor Doherty: No, it’s [bleep] It’s word salad. I made that one up. I completely made that one up.
Joannes Vermorel: Okay, I was thinking it’s over the top even though but again you know the bar is quite high.
Conor Doherty: So well the last entry in this round is agile adaptive autonomous planning and execution and I will tell you if it’s not real I would I would I’m pretty sure that somebody could come up with that.
Joannes Vermorel: Yes, it feels like very much the sort of copywriting style of some of our peers. Yes, I would I and yeah, it’s just the line of thinking. Yes, I tend to think it’s real.
Conor Doherty: Okay, final answer. Yes. Agile adaptive autonomous planning and execution or APEX for short from March 2026 o9.
Joannes Vermorel: Oh yes, they launched APEX, the agile, adaptive autonomous planning and execution capability to accelerate enterprise value creation in the AI era.
Conor Doherty: That was the official release from March 2026. I think you got most of those right actually. I’ll have to go back. I was caught up in the moment. I think you got like four out of five, five out of six or something like that. The thing is now the thing is and this is why I picked these ones. Like for example, when you talked about autonomous scenario orchestration, one of the alleged 10 autonomous IBP capabilities. Okay. Whatever we might think about that in this context. Surely you’d be on board with that conceptually within an economic perspective. You’d be all on board for
Joannes Vermorel: I do think that this would be an actual legitimate use of AI. And the way I would think of it is have the agent every day read the news that’s why I say ch you know it’s like let the agent read the news every day and update 10 scenarios of interest taking into account you know evolution of geopolitics that’s will the situation on the Strait of Hormuz be you know sorted out and by which date and these sort of I do think that when it comes to just have an LLM consume the news and give you probably like a dozen of different outlooks on your market. Yeah, this is this is a very valid use and especially if you were using in the past, you know, business analyst to do that can be completely automated. Now the value of that will still be a little bit low. Let’s make no mistake you know that you can have you know those 12 scenarios but it’s just perspective and it’s the sort of things where it’s more like food for thought for the upper management you know it’s not something that is it’s something that is nice in order to I would say refresh your understanding of some situation you may not keep an eye every single day on what is happening in this region of the world is there something you know the LLM could update oh the European Union is getting a little bit testy on some regulatory ideas for this and that maybe we should include one scenario which includes like a massive regulatory disruption for this and that you know that’s the sort of things where yes absolutely LLM would be would be quite useful I do not but where I would I would be very skeptical is that I’m very skeptical that a company like EY or classic enterprise software vendors EY is not an enterprise vendor but let’s say that a classic enterprise actor yes would deliver anything that is not immediately that is not just ChatGPT repackage you know that’s my take is that yeah it’s a valid thing but you can also have like just a ChatGPT setup with teams and a shared folder and ChatGPT will just do that for you and it’s fine especially now that they have like scheduled tasks it will just do that no need to go for a fancy enterprise package the consumer product ChatGPT would be would be just fine for that
Conor Doherty: Well, to be fair, I didn’t read you the EY explanation of autonomous scenario orchestration you’ve heard those words and you’ve gone in a in a different direction we use the were different. You’ve gone in a different direction because the way they use it is explanation AI simulates thousands of tradeoffs across cost, service, and margin in which scenarios are dynamically triggered not manually generated. That sounds antithetical to what you’ve just described in terms of an LLM’s capabilities.
Joannes Vermorel: Ah yes here basically they are saying oh we can we can have the LLM just play with the digital twins. Yeah. And I say it’s completely useless because the problem with your digital twins is assessment whether this is a correct simulation of the reality and your LLM doesn’t know that so it’s just going to again that’s full that’s this is this is in terms of method this is very wrong you are the problem is that if an LLM starts operating a digital twin it will the digital twin will generate figures, but are those figures real? LLM will have no freaking clue and it will give you conclusions that are supposedly grounded in numerical simulation, but in practice completely bogus. That’s, you know, that’s the number one value of humans. The way I see it in this I would say post LLM world is that you still need to have humans to make really sure that the things that you’re doing are numerically are really grounded in the reality and that’s one of the big problems with digital twins is that the grounding is extremely weak and shallow. So, plugging an agent and I can really see why they want to plug an agent. It’s because there is like the problem with digital twins is that you can resimulate and retweak everything and it will give you endless numbers about your supply chain. But none of it is real and an agent is just going to spin that endlessly. It’s not going to ground. It’s not going to produce any kind of tangible proof. You should not confuse I would say an agent being able to orchestrate let’s say an actual experiment with conclusions and thus ground its reasoning into something with an agent playing with digital twin where it’s just going to be a lot of nonsense. Lot of nonsense. And by the way, it would be the exact same nonsense if you were to say I take a very smart graduate of MIT and I say you have three months to play with this supply chain twin and give me and produce an interesting report. You could do that. You could do that in real life with an actual graduate from MIT. This I can guarantee you that this graduate will almost invariably produce things a report that is very clever-looking and complete nonsense and complete nonsense. Again, the problem being that you if you want to you need to have a mechanism to challenge the very fabric of the digital twin. That’s how you get something that is that is actionable for your company.
Conor Doherty: Well, one thing that unites all of the vendor and consultant material that I pursue peruse was the idea of you your company is going in the direction of decision intelligence or decision-making intelligence systems or systems of intelligence. I mean that’s our language but it all sort of orbited around this language. Now you’re making the argument that digital twins and LLM are not sufficient to get you there. Then what is?
Joannes Vermorel: No, I mean you have to understand that what they are graduating toward is a very shallow cheap incremental take on inadequate technologies. You see the way they approach the problem is they take a technology that is essentially crude. So create, read update, delete. It’s a system of records in disguise. And then they think of the most incremental thing that they can do to duct tape an LLM on top and that’s exactly what they’re doing. And then once they have done that, they say success, mission achieved. We have agents. And I say no, no, no, no, no, no, no, no, no, no, no, no. It’s you can’t just you know take a horse put an exhaust pipe on the side and say damn we have a car there is metal an exhaust pipe it’s no it’s the thing is that if you want to move toward autonomy what I’m saying is that there are things that are simpler that do not require LLM and but that requires paradigms change okay you see that’s the thing is that the transformation that you need is much simpler. It’s much more fundamental and easier. And LLMs, they’re great. It’s a productivity boost, but they do not change. If you don’t do the paradigm switch to say, am I thinking I’m approaching the case from an economic perspective or a non-economic perspective, as long as you remain in the non-economic perspective, you’re stuck. You will not do any kind of optimization because optimization means that you can optimize for dollars but you operate in a in a in a paradigm where dollars do not even exist. S&OP the dollars do not even exist. So there is no optimization to be to be obtained and so and what I see with those vendors who are pushing that is that literally they just bolt on top of system that are complete legacy in term of paradigm and they just they just bolt an LLM on top make no sense make no sense it’s if you really want to take advantage of this LLM revolution ution I would say absolutely but you need to rethink the way you approach the problem. It’s and by the way it is cheap it is fast. I’m not recommending for something where you need to spend millions in consulting fees. The economic perspective is in fact the perspective that is already adopted by your teams. M that’s the sort of things that what I’m advocating for is already what is happening in your company against IBP and against S&OP because at the end of the day your company if your company is making profits people are already doing this doing those economic analyses except it’s a second-class citizen it’s like not recognized as such but at the end of the day there is someone who says okay do we purchase something from this supplier Yes or no? Someone is actually answering this question and this person is making an economic reasoning. Someone is already doing an arbitrage arbitration between we have 10 units left the customers. We cannot serve everybody. Who do we serve? Yeah. But it’s ad hoc. It’s ad hoc. It’s ad hoc. It’s ad hoc. What I’m saying is that the true revolution is just to say we are going to formalize that so that we can robotize and we need to formalize the forces that come in those arbitrations and then you can use the LLMs to even speed up the process and do a lot of things that are very interesting. But fundamentally you need to put yourself in a in a you need to equip yourself with a paradigm that is quantitative that is economic and that focuses on actual decisions. So allocation of resources and not numerical artifacts as long as you’re operating in a plane of numerical artifacts nothing will be real. You know it doesn’t matter LLM or not it’s it will be just whatever.
Conor Doherty: Well, that was by way of a of a concluding remark I was just about to say to you. So earlier like the first question was you has AI killed S&OP and IBP and you said no because I mean technically you don’t need AI to do that you could have done it 10 years ago but based on the discussion that we’ve had it sounds as if the inclusion or the bolting on as you said of LLMs and Agentic AI into these processes is essentially just more of the same. So nothing is going to die within the sorry within the framework of people who already use it.
Joannes Vermorel: Yes. Exactly. I mean it’s again it’s and the worst part is that people will feel productivity benefits but just for bureaucratic tasks you know it’s a it’s not the economic revolution that you’re that you yes again it’s you have this those bureaucratic processes it’s you see and I think here Elon Musk was saying something that is very interesting when he was doing his vice business say you should never optimize what should be just deleted in your organization And he said many times we optimize stuff where if we were approaching the problem differently this thing would not even exist anymore. So this process doesn’t have to be improved optimized you know etc. It’s if you approach it right, it just disappears altogether. And that’s the problem that I have with again if we step back on the philosophy that is adopted for most of those actors is that they adopt a very incremental perspective where it’s there is never anything that is subtractive you know it’s what can you remove and instead that’s a key question and I think for the audience when you want to think of the LLM revolution is that if you can’t just purge your organization from stuff you’re approaching the problem wrong you know it’s you should really think it and I re you really use the approach the word strong word purge I really mean it’s something where you delete it is less and it’s better the same remark about how do you what do you replace a tumor with exactly what do you replace at wave exactly that’s the exact the exact line of thinking
Conor Doherty: Right well thank you very much for your time I don’t have any more questions but it was a pleasure as always and to everyone for watching. If you want to continue the conversation, if you want to pitch us questions about autonomous IBP or the economic perspective, you can reach out to us on LinkedIn, continue the conversation there. Or if you want to learn more about Lokad’s economic perspective, you can shoot us an email at contact@lokad.com. And with that, we’ll see you next time. Get back to work.
