Supply Chain Software and ChatGPT: The Ultimate Squirrel Test

In the movie “Up” there is a great scene where the dog named Dug is having a conversation with the main character Carl, and Dug suddenly stops speaking, rapidly swivels his head, and… “Squirrel!”.??

Why is the scene so memorable and funny???Maybe we can relate to real life when we are working on something and out of the blue “Squirrel!” we get distracted and completely lose focus. Happens to all of us.

This is what is happening today in the field of Supply Chain software, especially artificial intelligence (A.I.) applications. Like Dug, it seems the supply chain community was having a reasonably productive conversation about how A.I. could advance our profession, and then… ChatGPT!??The mother of all technology Squirrels.

No doubt, ChatGPT, a form of Generative A.I. called a Large Language Model (LLM), is a powerful technology and will disrupt text-based professions including sales, marketing, and academics. Other LLMs are available from Google (Bard) and Meta (Llama). However, the applications of LLMs to supply chains are dubious.??There are many ways A.I. is advancing supply chain management by analyzing patterns in graphs and numbers, but text-based LLMs are not the answer.

We should swivel our heads back to the productive conversation and focus on A.I. use cases that help supply chain professionals.

Forgive the basic primer on A.I., but there is a lot of confusion in the market.??There are two fundamental types of A.I. – Predictive and Generative. They both have tremendous potential to have a positive influence on many professions including supply chain management. Predictive A.I. analyzes patterns in data and provides matching labels for those data based on patterns it has learned - e.g., this is a picture of a cat (image), this paragraph of text has a positive sentiment (text), this forecast matches the likely future demand for a product (numbers).??In contrast, Generative A.I. creates new content (not matches of existing content) from a prompt. One simple way to understand this new technology is that Generative A.I. is Predictive A.I. running in reverse (no DMs from Data Scientists who may argue fine points please). So instead of uploading a picture of a cat and the A.I. labeling it as “cat”.??You prompt “cat” and it draws a picture of a cat. Or instead of uploading a poem and the A.I. labeling it “inspirational poem about the sky” you prompt “inspirational poem about the sky” and it writes a poem.

At risk of stating the obvious, supply chains can best be represented as graphs (locations of plants, distribution centers, etc.) and numbers (supply data, demand data, constraints, etc.) rather than text and images. I guess you could use LLMs to write a limerick about your supply chain and it may be amusing, but it won’t satisfy the CFO. So clearly, in supply chains, Predictive and Generative A.I. algorithms should focus on graphs and numbers rather than text and images. Note: Ask ChatGPT to use numeric data to predict the world human population in the year 2050. You will get something like “I am a language model and don’t have the ability to generate numeric forecasts”.??Or ask ChatGPT who the President of the United States is and it will tell you the last data refresh it received was in September 2021. It has no idea what has happened in the world in the last two years.

For supply chain management:?

Predictive and Generative A.I. – yes please.??

LLMs like ChatGPT, Bard, and Llama II – no thanks.?

Let’s double click on useful applications of Predictive and Generative A.I. within supply chains.

Predictive A.I. to Forecast Demand, Supply, and Inventory Imbalance Risks?

For decades, supply chain professionals have been hitting their heads against a wall trying to get better forecasts of supply and demand. Typical error rates for demand forecasts exceed 30% and supply forecasts typically are even worse when measuring timing and quantities delivered.??No matter how good your planning process is, if you start with bad forecasts, your plan will be wildly inefficient. Garbage in garbage out.?

Predictive A.I. allows supply chain professionals to create forecasts of demand and supply that are dramatically more accurate and informative than legacy Advanced Planning Systems (APS).??

The trick is to use a data driven approach and automatically apply the right combination of A.I. algorithms, statistical forecasts, and human input at each level of the supply chain hierarchy: products (SKU, brand category), timing (weeks 1-16 from sale), and geography (distribution center, region, global region, global).??Given the right technology architecture, we can generate individual forecasts for each combination of product, geography, and time – millions of different forecasts.??No one size fits all – because it doesn’t. After creating these millions of predictions, the software reconciles these forecasts so they make sense together.??This approach yields dramatic reductions in error rates and sets up successful sales & operations planning (S&OP) and sales & operations execution (S&OE) processes.?

An added and consequential benefit of an A.I. led approach is the outputs of these forecasts are not single numbers like those from APS. Instead, the algorithms output the probabilities of different forecasts. We all feel the limitations caused by legacy APS ignoring uncertainty and failing to give planners a feel for risks.??APS systems give forecasts and present them as 100% certain. Everyone knows the numbers are wildly wrong and the systems have no ability to calculate nor present planners with the shape of risks they are taking. The new A.I. driven approach empowers planners with a deep understanding of where risks loom and help users make better bets to achieve their inventory, fill rate, and financial goals.?

In short, using A.I. analytics you get better numbers and the unprecedented ability to understand and manage risks in those numbers.?

Generative A.I. to Optimize Materials and Inventory Allocations

Generative A.I. rightfully has rocked the A.I. world.??It is a breakthrough and can be both jaw-dropping and scary. ChatGPT has been the most visible example of Generative A.I.. And rightfully it has gained much attention for its ability to generate text. But in supply chains???Squirrel!

The most impactful approach to Generative A.I. within supply chains is to input the topography of your supply chain as a graph. The nodes in the graph are physical locations of your supply chain sources and destinations.??These nodes then are connected by lanes representing the transport options among nodes.??Every node and lane is loaded with data on constraints and historical performance under different conditions.??Then you add in the demand and supply forecasts and their corresponding risk profiles (the reconciled output from the Predictive A.I. algorithms).??So the opening scene is set. Once all these data are loaded into the graph, you ask the Generative A.I. to create a series of inventory movements to optimize your goal (e.g., profits, on time in full supply, inventory efficiency, waste reduction). The Generative A.I. algorithm then produces what we may see as a movie of the steps over the next 12 weeks (or whatever time frame you ask) to achieve that goal. Change allocations, de-expedite, expedite, change transport lanes, etc..?

If you don’t like the output, you can change your prompt and ask it to optimize in some other way. (Somehow “more cowbell” pops to mind) You also can change the constraints or network characteristics to see the effect. “I want less risk on revenue so take more risk on inventory”.??Or, “I want less risks on margins, so take more risks on expedites and service levels”. These risks can be tailored by customer, product, geography, or time.

So, using Generative A.I. you get very large scale simultaneous optimization of all supply, demand, inventory positions, and operational costs across every node and lane in your supply chain network. And you get to understand and control risks to match your goals.??Powerful stuff.

Wrapping all of these A.I. powered capabilities together you get the next big step change in supply chain management – Generative Probabilistic Planning. More accuracy, more risk control, more understanding. You set the desired outcomes you want your supply chain to produce.?

Final thought: as temping as it is to have ChatGPT edit this article, let’s ignore the squirrel and get back to helping the supply chain profession move forward.

Stephen Pratt, Founder & CEO, Noodle.ai