There are two clear camps in the world of AI: 1) it will lead to a cambrian explosion of new businesses that could have never existed nor could we have ever dreamed of prior, and 2) it will mostly benefit incumbents who have existing data and distribution, further deepening their moats.
I think both are true, but require the overlay of time as a lens.
Let’s look at the world of supply chain. From what I’ve seen amongst supply chain tools, I think AI will likely be overhyped and disappointing in the short term (next 3-4 years), but has the potential to unseat major incumbents in the long term.
Supply chain is a vast market and there are are myriad of decades old verticalized software solutions for every conceivable workflow as I outlined in Supply Chain API. Vertical supply chain software can be extremely sticky. Once software like a WMS, TMS, or IMS becomes core to the operations of a company, the majority of operational data flows through the system, and it handles several key workflows sufficiently, it is unlikely that it will switch to a new provider. Software solutions can be expensive, but they are also customized to the specific workflows of a given company, and that is hard to simply rip out.
As a result, the vertical supply chain software incumbents today have an advantage that will only be expanded with the deployment of AI. Companies like O9, Kinaxis, E2Open, Shipstation, and Project44, and Deposco already have unique workflows and data capabilities, large customer base, and services based ecosystem to have a significant moat and distribution advantage. These companies will be able to use AI for back office jobs that further deepen that moat and make the user experience even better. All of these solutions should be able to layer on the low hanging AI fruit of OCR, RFP writing, email drafting, and summarizing data. Even though such capabilities will functionally be a layer on top of OpenAI, that’s okay. It doesn’t have to be anything more because their existing distribution and data advantage is where the core value exists.
So where do you look for those applications of AI that have the chance to unseat incumbents? In supply chain, it will be the products that sell supply chain work rather than a more efficient existing process. This dynamic has been well codified in Sarah Tavel’s “Sell Work Not Software.” In the world of supply chain, the O9s and E2Opens of the world sell extremely powerful, complex supply chain planning software on a per seat basis. They make the work of a supply chain team easier by centralizing important data (after 6-9 months of data wrangling) and providing rails for key workflows and analysis.
A paradigm shift in supply chain planning an operations would be one where a product would actually do the work that an O9 or E2Open theoretically enables a supply chain team to do. This means ingesting all sources of data, connecting the dots between these disparate sources, which are often both structured and unstructured, and then providing users with supply chain decision recommendations and insights on an ongoing basis. Said another way, if platforms like O9 help companies run their S&OP processes (typically a rolling 4 week process), a product that actually does the work is one that surfaces the recommended outcome of the S&OP process.
In supply chain, data analysis and decision making can often be very complex, and unlike our current experiences using tools like ChatGPT, Anthropic, and Perplexity, the information presented has to be right 100% of the time. Answering the question of “where in my operations can I most effectively increase margin” sounds simple, but requires extensive scenario analysis. Or let’s take a more tactical question: “If I have to change the label of Product X within 3 months, when should I discontinue the old label so as to minimize inventory loss and ensure a timely launch?” This sounds basic but is actually really complex, because it involves understanding the cost of your label, tracking inventory for that label and projecting inventory for that label based on projected demand for all of the SKUs that use that label, and then projecting a potential inventory write down based on any number of hypothetical launch dates.
AI is not yet at the stage where it can handle queries like those outlined above, but when it does, it will be transformational, and a true 10x improvement from the last 10 years of supply chain software. Why won’t the incumbents get there first? Because the AI capability required to handle an entire S&OP process is fundamentally at odds with the business model an O9 or E2Open has developed for the last 15-20 years, and would require significant product cannibalization.
The incumbents are built to optimize complex, multi variant supply chain planning workflows. The AI-native supply chain winner will be built around delivering the work that those complex workflows were developed to produce.