AI in Supply Chain Management: Where AI Actually Matters

This post has already been read 7755 times!

AI can have a big impact in the right areas of supply chain management but there are significant challenges and obstacles

Artificial Intelligence, and in particular, generative AI, is in the process of completely revolutionizing supply chains (as well as other fields). It is one of those dramatic step changes that come along very infrequently and change everything.

The ensuing disruption will produce a new generation of winners and losers.

However, all is not perfect in the land of AI. There are many pitfalls that can trap the unwary. In this article we explore both the immense benefits of AI as well as landmines that need to be dodged in order to be successful in the context of supply chain.

"What AI brings to the table is a dramatic increase in capabilities that broadly fall under the umbrella of human-level pattern recognition and analysis." – Ranjit Notani #AI #MachineLearning #SupplyChain Share on X

Prior to AI, the state of the art in supply chain was about sophisticated intra and cross-company process execution coupled with sophisticated algorithms for planning and optimization and the seamless integration of planning and execution. A crude simplification of these approaches is that these were based on precision and determinism.

What AI brings to the table is a dramatic increase in capabilities that broadly fall under the umbrella of “human-level pattern recognition and analysis”. While pattern recognition and analysis were present in the previous generation, today’s AI makes a quantum leap in that regard. So, in this sense the strengths of AI are very complementary to strengths of the previous generation of determinism and precision.

It is evident to us that what we really need is a marriage of both these approaches.

Where AI Can Make an Impact in Supply Chains

So, where can this kind of pattern recognition dramatically improve outcomes?

The obvious example is in the area of demand and other types of forecasting. These processes were already conceptualized as pattern recognition-based approaches, so it was a natural leap to apply AI to these processes. Indeed, the new cutting-edge AI forecasters are already outperforming (sometimes dramatically) the previous generation of forecasting. One reason for this superior performance is that Machine Learning (the branch of AI getting the most attention currently) learns from huge amounts of data and can detect causal patterns that a human or earlier algorithms could not detect. Improving forecast accuracy is one of the known process improvements that drive direct bottom line benefits to companies.

Other examples of processes that fall into this category (i.e., already sort of utilized pattern-based approaches) were processes like multi-echelon inventory optimization, ETA predictors, etc. These processes are seeing similar dramatic improvements with direct bottom line benefits.

Lurking below the surface of these “obvious” areas is a vast set of processes which were essentially pattern based but the patterns were too complex to process and mostly were the domain of human analysts, planners, etc.

Where AI Actually Matters in the Supply Chain: A look at the low-hanging fruit for AI and the major challenges and obstacles… Share on X

Some examples of these areas are policy setting. i.e., what policies should one set to drive improved outcomes. Modern supply chain systems (even highly templatized with best practices) come with a huge number of policies that can be set, but figuring out how to set these takes time, even with experienced human analysts and planners.

Another area that falls into this category is in the realm of problem prioritization or “attention focusing.” Modern supply chain systems enable real-time visibility and alerts across a multi-enterprise supply chain. This can produce a blizzard of issues demanding attention. The question is what should the analyst/planner be focusing on.

The traditional approach would be some sort of high/medium/low classification. But AI can now do a much better job of focusing the analyst’s attention on the subset of issues/alerts that are truly the most critical. It can look at patterns of the underlying supply chain as well the user’s context to produce much better prioritization of issues that require attention. This allows much better utilization of the user’s most scarce resource (time).

AI excels at problem prioritization. It can do a much better job of focusing the supply chain analyst's attention on the subset of issues/alerts that are truly the most critical to the business. Share on X

Beyond visibility, the next step up is in the realm of decision making. Supply chain has a notoriously “high dimensional” decision space. In plain language, there are so many interacting factors and everything seems to impact everything else that it becomes hard to solve problems effectively.

In the past some of this was “solved” by doing planning with mathematically pristine objective functions and constraints. While, still very useful, these approaches had the major problem of not dealing well with the rapidly-shifting realities of real supply chains (especially the closer one got to execution). Here things would get increasing chaotic and users would resort to the process euphemistically sometimes called “expediting” and “firefighting.” However, with the advent of AI, it can now look at the complex planning and execution picture and suggest real-time, context-sensitive, Smart Prescriptions™, which guide the user towards optimal outcomes.

All of the aforementioned applications of AI have the advantage that they also continuously learn based on a tight feedback loop between action and outcome. This in turn means that even when there are “phase shifts” in supply chain characteristics, the AI picks it up.

The next level up in human-level pattern-based analysis and action comes from processing human languages. This is where the Large Language Models (LLMs) like ChatGPT are playing a significant role. On the one hand these models enable chat, but if one zooms out a bit, what they fundamentally do is open the world of textual data to be incorporated into supply chain decision making. It turns out that there is a gigantic volume of useful textual information out there, but these could never really be processed at scale. LLM’s change that.

"Supply chain risk is a critically underemployed capability and is needed now more than ever. With AI we can now conduct supply chain risk management at scale." – Ranjit Notani #AI #SupplyChain Share on X

For example, supply chain risk data is primarily found in textual sources and for the first time these can be processed, semantically analyzed, classified and applied. This is allowing supply chain risk management to finally be done at scale in a systematic manner. If the last few years have taught us anything, supply chain risk is a critically underemployed capability and is needed now more than ever. Along with the sudden availability of risk-information-at-scale, risk-enabling the “deterministic” side (multi-echelon constrained planning and execution) of this is also needed and being done.

Other examples of textual information creating breakthroughs are in the area of quality management. Finally, a huge part of the transactional supply chain is still documentation oriented, and all these processes can now be fully integrated.

So, far we have discussed some examples of mostly supply chain functional transformation. However, there is an equally dramatic transformation happening on the information technology (IT) side.

For example, one of the major drivers of time and cost to deploy modern supply chains is the time and expense of integration. However, AI is now enabling “semantic integration” (interrelating diverse sources of information and data), dramatically reducing the time to do these integrations. Another major building block is clean multi-echelon master data with clean cross-referencing and naming. Again, AI and its semantic mapping abilities are bringing dramatic improvements to this process.

The Challenges of Deploying AI in Supply Chain

With all of these revolutionary changes happening it is important to not overlook some important caveats.

The Black Box Problem

One of the major issues is that AI is something of a black box and produces great results most of the time. However, because the way results are actually produced are poorly understood (compared to say combinatorial optimization algorithms), it is hard to anticipate and deal with some failure modes.

At least in the near term, the whole user experience of supply chain applications needs to be amended to allow humans to naturally vet decisions, especially high stakes ones. Then, as users get comfortable with the decisions, these applications should easily allow users to “promote” the AI to a fully-autonomous mode.

Data Quality & Volume

Another major caveat of AI, especially the Machine Learning sub type where most of the active advancements are, is that they are highly dependent on large volumes of high-quality data. No data, no AI. This means that unlike traditional algorithms which will sort of function in all contexts, an AI may do very well on a problem where there is lots of data and fail on a problem where there is less or poor data.

There are two solutions to this problem. The first is that any modern AI-enabled supply chain system should be able to ingest AI models trained on data that may not even be in their system. This requires a standards-based approach to dealing with AI models. Both de-facto and de-jure standards are a critical element of this.

The second solution to this problem is based on the notion of foundational AI and so-called transfer learning. With this approach AI’s can be trained on large datasets to solve problems generally and then via transfer learning with relatively small additional data be “fine-tuned” to solve a particular problem where the data may be much sparser.

Where AI Actually Matters in the Supply Chain: A look at the low-hanging fruit for AI and the major challenges and obstacles… Share on X

Unleashing the Effectiveness of AI in Supply Chain Requires Significant Change

In summary, AI is dramatically changing every aspect of supply chains leading to dramatically improved outcomes to companies that most effectively adopt them. However, it requires a comprehensive change to how modern systems are built.

An effective merger of traditional combinatorial optimization with AI to play to the strengths of both.
A rethinking of the entire user experience to move to a more natural chat-based interface, as well as deal with the issues of AI’s being somewhat of a black box.
Bringing AI into the IT underpinnings of supply chain systems to dramatically improve the time to value.
Building supply chains that take into account the caveats of AI.

At the end of the day, we are on the path to creating true, creative supply chain AI assistants that will make human users much more productive allowing them to focus on the most difficult problems. These supply chain AI assistants will be imbued with knowledge of the supply chain in general and the user’s supply chain in particular.

If you’d like to dive deeper into this topic, I highly recommend reading 8 Keys to Success with AI in Supply Chain Management. It will guide you through some of the critical factors needed to properly deploy AI in the supply chain.