Natural Stupidity vs. Artificial Intelligence
Transcript of the talk given by Joannes Vermorel at SCTech 2024 on October 9th (2024). Symposium topic: ‘Supply Chain Intelligence - Artificial or Not’. The event was organized by the Interational Supply Chain Education Alliance (ISCEA).
AI has been the buzzword for 2024. Vendors, software companies and consultants alike are now making grandiose claims about all the benefits that can be expected from artificial intelligence. Lokad, my company, is one of those vendors, a software vendor that specializes in the predictive optimization of supply chain. However, my goal for today will be to show that the quasi-totality of those AI initiatives will fail and result in net losses for companies. This contrarian take is essential: why? Because, well, your AI initiative will fail, yes it will despite what my competitors are selling you, and when it does, you will remember that there was a guy with a funny French accent that told you so; and maybe, next time, you will pay attention to what his company, Lokad, is actually advocating as far as supply chain and AI are concerned.
In this age of emergent artificial intelligence, let’s have a look at its nemesis, natural stupidity. Einstein famously said, “Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe.” Einstein was right: the importance of natural stupidity cannot be overstated, and unlike AI, it has been around pretty much forever. Thus, it can be safely assumed that it will still be with us a few decades from now.
My proposition today will be as simple as it will be offensive: natural stupidity is the number one hurdle that will prevent you from succeeding with your AI-driven supply chain undertaking.
Now, this proposition will, most likely, be perceived as outrageously arrogant. To my defense, I will say that it doesn’t make it any less true, and that arrogance is also the second biggest national sport in France, right after bureaucracy. More seriously, I will be the first to admit that it is a bitter pill to swallow, being painfully aware of my own limitations, but remaining in denial of the problem isn’t going to solve anything.
Let’s first clarify an important distinction between two classes of enterprise software: the systems of records and the systems of intelligence. When it comes to supply chain, the systems of records are the bookkeepers of the flow of physical goods. Those systems record the products, the purchase orders, the manufacturing orders, the sales orders, the stock levels, etc. The systems of records take care of all the mundane data entries, and automate all the clerical tasks, like totaling inventories. They are nothing more than cheaper, more reliable versions of the old paper trails. ERP, CRM, PIM, PLM, WMS … all those maximally opaque acronyms refer to systems of records. Systems of records are categorically non-intelligent. They are not stupid either: intelligence doesn’t even apply to them. Thus, systems of records will be entirely left out of the present discussion.
Today, I will exclusively focus on systems of intelligence. A system of intelligence is a class of enterprise software that is intended to mechanize a decision-making process. The more intelligent the piece of software, the better the decisions. For example, antispam filters have been discreet but ubiquitous systems of intelligence that have been deciding, for the last two decades, on your behalf, what messages were worth your attention.
In supply chain, the decisions of interest are: When to buy? How much to buy? How much to produce? How much to allocate? Etc. Optimizing the flow of physical goods requires roughly a dozen major classes of decisions to be taken daily. Intelligence is demonstrated by generating profitable decisions. Conversely, stupidity is demonstrated by failing to generate profitable decisions. Thus, whenever artificial intelligence is mentioned, it must be understood as a component of a system of intelligence.
My proposition is that systems of intelligence dedicated to supply chain, which have been sold to enterprises since the late 1970s, have been an uninterrupted stream of dismal failures. Lokad, my company, put an end to this uninterrupted stream of failures in the early 2010s. Worldwide, there are a few other similar exceptions, however Lokad, like those exceptions, are beyond the scope of this talk. I invite the audience to have a look at the Lokad.com website, and the Lokad TV YouTube channel if you want to know more. Back to the topic at hand, the fact that artificial intelligence, delivered in the form of Large Language Models (LLMs), happens to be available is not going to be doing anything about this trend of supply chain failures. Indeed, as we will see, those failures were not caused by a lack of technological instruments, and thus, adding more instruments to the pile is not going to change anything, unless we address the underlying problem first. Unfortunately, it is not an easy task, as the underlying problem happens to be natural stupidity.
First, let me demonstrate the validity of my diagnostic. Back in 1979, Russell Ackoff, an American pioneer of operations research published “The Future of Operational Research is Past”. This fascinating paper explains, with great clarity, why all the techniques that emerged from operations research, which includes pretty much everything considered as the “core” of supply chain nowadays is failing, why the entire domain is flawed, and why it will keep failing for as long as the community, both academics and practitioners, doesn’t come to terms with this flawed paradigm. This is a visionary paper, and probably the one paper I wished I had read when I started Lokad back in 2008. Unfortunately, I only rediscovered this paper a decade later, when I had essentially arrived at the very same conclusions, almost 4 decades after the work of Russell Ackoff. How do we know that those systems of intelligence for supply chains are failing? That what passes for enterprise software dedicated to supply chain optimization is failing? Well, after having the chance, through my career, to talk to over two hundred supply chain directors on both sides of the Atlantic, I can state the following: all those enterprise software products have invariably resulted in the supply chain teams reverting to Excel spreadsheets.
There is no shortage of enterprise software promising to automate supply chain decision-making processes. This has been the one central promise of all software vendors dealing with supply chain optimization since the late 1970s, and yet after every single deployment, supply chain teams reverted to spreadsheets. You don’t need to take my word for it. If you happen to work in a sizeable company, say, half a billion euros in turnover and above, I can guarantee that there has been at least 1 failed attempt at implementing such a solution every decade since the 1990s. Yet, dear supply chain practitioners, you are still using spreadsheets, and it is not because your company is immature or because your colleagues are lazy. You are still using spreadsheets because those enterprise software vendors failed, because their systems of intelligence failed. My proposition - and I am not going to make many friends today - is that those failures must be attributed to natural stupidity, in fact, the very same flavor of stupidity that Russell Ackoff identified with such clarity back in 1979 after decades of contribution in the field of operations research.
To demonstrate this proposition, I will be surveying 4 objects that, in the context of supply chain, are provably stupid. Those 4 objects are: RFPs (request for proposals), time-series, safety stocks, and service levels. Any company using any of those 4 objects is setting itself for failure. It does not matter how much “AI” is thrown at the case. Natural stupidity cannot be vanquished by artificial intelligence.
Let’s start with requests for proposals. Selecting the right vendor is obviously critical, as there is clearly no shortage of utterly incompetent software vendors that are more than happy to collect millions of euros in fees for dismal technologies. Thus, unless your company has a very robust vendor selection process, you will almost certainly end up with an incompetent vendor. However, RFPs are not the way.
As a software vendor on the receiving end - we get several RFPs per week - I can testify that those documents are not only invariably stupid, but they are also downright insane. As a rule of thumb, an RFP includes hundreds of questions. Every question seems to be competing for the prize of the most irrelevant question of all time. For example, last week, one of the questions was: what are the fireproofing capabilities available for the storage room dedicated to your fax archive? We are in 2024; I haven’t used a fax in two decades. In fact, some of the younger people in this audience might not even know what a fax is.
However, even the supply chain questions are stupid. Why? Because the quasi-totality of questions are not questions at all, but strict requirements. Most “questions” look like: Is your software capable of letting users update seasonality profiles up to 36 months ahead? There are so many things wrong with this question, I don’t even know where to start.
Let’s step back and imagine yourself writing an RFP to buy a smartphone. You intuitively feel that the iPhone is the best smartphone out there. However, you start listing requirements in your RFP such as the size of the battery, the material for the screen, the exact settings that should be, or should not be, available to the end-user. What are the odds that your list of requirements end up ruling out the iPhone from the RFP? Assuming 100 questions or more, the odds are 100%. You will invariably end up excluding all the decent vendors. The only vendors that will play this stupid game are the ones so desperate, because their technology is so lacking, that they have no choice but to say YES to every misguided requirement put forward by the client.
Thus, RFPs are the first piece of natural stupidity that needs to go.
Then, we have time series. Oh, I don’t deny that time series are good for visualization purposes, but for supply chain optimization purposes, this is stupid. As a result, any solution, framework or technology, that puts time series front and center is guaranteed to fail; except this failure will be on you, because you should have realized that using time series was a stupid idea in the first place.
Indeed, time series, as a mathematical model, are simply not able to convey the information that we need. Time series are unidimensional, and for supply chain, this is just not sufficient. For example, let’s consider a company that has been selling 100 units, give or take, every week for years. The demand seems extremely stable. Now, let’s consider two variants of this situation. In the first situation, the company has 1000 clients, where each client buys 1 unit every 10 weeks. In the second situation, the company has 1 unique client that buys 100 units a week. What are the odds that sales can go down to zero next week and stay at zero forever? Well, in the first situation, those odds are very low. There are 1000 clients after all. Losing all of them will most likely take time. However, in the second situation, it only takes 1 client to change its mind to lose 100% of the sales. Those two situations are nothing alike, and yet, they have the same time series.
This example demonstrates why time series are not adequate to represent the past. Dozens of other examples can be found. I am leaving that as an exercise for the audience.
However, time series fare no better if we are looking at the future. Your time series is stating “here is what will happen”, “here is the one future”. However, the future depends on decisions that have not yet been made. Time series are entirely oblivious to this, they treat the future as completely symmetrical from the past. However, supply chains aren’t physics. We can’t look at the future demand for a product as if it were the movement of the planet Mars. The future is radically distinct from the past because we can change the future.
However, we can only change the future if we are willing to change it, that is, if we are willing to stop using time series. Again, in supply chain, no amount of artificial intelligence can undo the natural stupidity associated with time series.
Let’s now turn our attention upon safety stocks. This is one of the pillars of the modern supply chain theory. There is hardly any supply chain textbook of the quantitative variety that does not cover safety stocks. Safety stocks are also the sort of basic features that any supposedly decent supply chain optimization software is expected to feature. Yet, my proposition is – and it shouldn’t be too much of a surprise at this point – that safety stocks are stupid.
The reason is extremely simple: safety stock formulas are the correct answer to the wrong question. Let’s consider a supply chain that invariably involves thousands of SKUs (stock keeping units). If you have €1 to invest in inventory, then the question is “what is the unit of inventory across all SKUs that will maximize my profits?” It does not make sense to frame the question as “Should this one SKU in particular get 1 extra unit in stock?”. All SKUs are in competition for the same scarce resource: the cash of the company. Treating all SKUs in isolation is like assuming the amount of cash available to the company is infinite.
Moreover, the very definition of economics is the science that studies the allocation of scarce resources that have alternative uses. The very concept of safety stocks contradicts the elementary economics, in fact, it contradicts the very definition of economics.
Thus, considering the magnitude of the error that safety stocks represent, it is wholly inadequate to qualify this error as being misguided, now we have a much better expression for this class of errors: being stupid.
Finally, let’s consider service levels. If I could gain 1€ every single time a company was leveraging a service level to inflict economic damages upon itself, I would be a billionaire by now. In supply chain, the service level is the probability of a given SKU of not facing a stock-out during the next inventory cycle. As a piece of descriptive statistics, service levels are fine, they are neither intelligent nor stupid, like any other simple statistical indicator.
The stupidity only manifests itself when people start to assume that the service level - this percentage - is somewhat correlated in any shape or form with customer satisfaction or the profitability of the company. This is not the case. Let’s consider fashion: to make room for the next collection, the company must liquidate the previous collection. Getting the service levels down to zero is necessary to bring novelty, and keep customers satisfied.
Conversely, let’s consider aviation. A typical jetliner involves about 300,000 distinct parts. Tens of thousands of parts need to be inspected and replaced routinely. If a single NO-GO part happens to be unavailable, then the aircraft gets grounded causing hundreds of thousands of euros of economic damage per day. Having 99% service levels on all your SKUs means nothing: only the probability of not getting aircraft grounded matters. It’s the weakest link that defines the number of AOGs (aircraft on ground) incidents you will get. The average service level is wholly irrelevant.
Similar problems in every vertical, not just fashion and aviation. Once more, I am leaving this as an exercise to this audience. Elementary logic dictates that only fools would chase service levels, and yet, here we are, with most companies doing exactly that, and their managers wondering whether they should invest in some AI-powered service level optimization technologies.
In conclusion, artificial intelligence will not save you from bogus supply chain theories. It will not save you from consultants whose only competence is showmanship and their capacity to instill confidence. It will not save you from software vendors who are more than willing to sell you whatever insanity happens to be fashionable today.
To succeed with artificial intelligence, natural stupidity must be defeated first. In the grand scheme of things, my company, Lokad, will not succeed; but, if we can make a dent in the problem, pointing out the most widely popular stupid ideas of supply chain, as I did today, then it’s already a step to victory.
