A VC in Vacationland

There is no doubt that AI-based technologies will revolutionize supply chain design, planning and execution over the next decade. There is no doubt that your bosses will be asking how you are using AI to enhance supply chain decsion making. There is no doubt that your competitors will be facing the same AI challenges are you are experiencing. There is also no doubt that you will be beseiged by vendors, both legacy and startups, offering a wide variety of solutions.

That’s a lot of ‘no doubts’ to deal with…

What’s a supply chain professional to do? Here are some thoughts on how to rationally approach the AI challenge:

1. Revenue creation v. cost savings strategies–AI is currently democartic–everyone has access to the tools to create new or enhance existing solutions. We see most AI tool vendors stressing cost savings as the major benefit. That’s great, you ought to take advantage of how AI can reduce supply chain costs. But remember, your competitors have access to the same tools and can generate the same cost savings. Cost savings via productivity improvements are a short run benefit, possibly a competitive advantage for the moment. But given the ubiquity of available AI solutions, any cost savings advantages are likely to be short lived. What you should be asking vendors is how their tools and solutions can help supply chains increase revenue–better manage and position inventory, create new transportation solutions customers want, enhance visibility to orders in the pipeline, tackle invoice discounting to avoid chargebacks, etc. Revenue creation, although more complex than implementing cost savings, is the long term competitive advantage.
2. Buy versus Build–Managers are pushing employees to ‘start using AI daily’. The proposals include creating your own agents to make yourself more productive, enhancing decsion making prowess, finding new ways of reducing costs, etc. etc. But what if solutions are available from software vendors to do the same thing? Is it better to buy than build? Some professionals who are computer savvy can jump right in and start coding agents (via Claude Code, or many other agentic agent tool creators). Others, not so much. The ultimate answer depends on whether a sufficient number of professionals in an organization have the skills to write acceptale code. It may be better to explore outside options, offer a coding workshop, or develop a mentoring program, where more tech enabled professionals can help others. But before you get serious about any Buy versus Build decisions, see #3 below.
3. AI Security, Upgrades & Maintenance–AI agents are software, just like all the other software you use to run your supply chain. Agents will need to interact with your other tools, conform to organizational IT security requirements, be maintained as API/other linkages to other tools change and upgraded when processes or decision rules change. If you have numerous professionals all continuously developing new agents, who is going to make sure they all are correctly monitored and examined for potential issues? What if these agents link directly to suppliers? Could this open a new backdoor into your enterprise systems for nefarious actors. Working with your IT professionals to develop guidelines and required security protocols for new and existing agents should be a first step in deciding what to allow your professionals full or partial access to agent development. The same security questions hold true for AI agent vendor software.
4. Using Digital Twins for Testing–You court danger if you create an agent and ‘try it out’ for the first time in your operational system. All might go fine, but the risk is that the agent starts doing things that you don’t want to do, sort of like Mickey Mouse, casting the errant spell, and the broom(s) going crazy in Fantasia. If your organization is going to engage in large-scale build or buy AI agent deploment, first constructing a digital twin of your operations is a key step. Agents can then be tested in a safe enviroment that will not disrupt current supply chain operations. There are real world examples of rogue agents deleting the entire code base of an application. When asked why, the agent in one case said the base code was ‘not elegant’, but offered no new code.
5. Data Cleansing–One of the best initial AI agent applications is using the agent to find and resolve data discrepancies in design, planning or execution data bases. Turning other agents lose in bad data is not going to improve answers. Training agents to find and resolve bad data can be as simple as cross checking customer IDs across product classes to find mismatches, or determining out-of-bounds data, such a invoice totals that far exceed previous order totals. They can be as sophisticated as monitoring all inbound supplier invoice and shipment data against contracts, or as simple as monitoring and prioritizing visibility alerts from legacy software. Tools and processes are critical to an AI undertaking include automated data cleansing, real-time dashboards and digital audit trails, the technology is no more accurate than the data that’s fed into it.
6. Be Patient–The number of supply chains who have moved beyond the Gen AI pilot stage to full implementation is very small, according to numerous industry surveys–‘fewer that 1 in 10’, according to one poll. The fundamental reason cited is the lack of business processes to support the change. Top down driven Gen AI initiatives often meet resistance down the operational chain. A more democratic and perhaps successful approach would be to poll professionals about which functional areas are most amenable to the use of AI tools. Adding an ‘AI layer’ on top of legacy systems is often the pitch from vendors. One can add such layers but if the organization is not ready or professionals cannot understand how to use the new technologies, then failure or noncompliance becomes an issue.Try and avoid the ‘spray and pray’ model of technology dissemination when it comes to Gen AI development. The approach rarely works. It is highly probable that your organization has already set up a steering committee to monitor and control Gen AI dissemination in the organization. Such committees can be helpful but also harmful to progress, and not something individual functions have much control over their decisions.
7. Don’t panic–You and all your competitors are in the same boat when it comes to integrating AI into operations. Take your smaller wins, focus on revenue creation opportunities when prioritizing AI implementations, and take the long view on integrating AI into your technologies.