Watch: Moving From AI Experimentation to Real-World Execution

Companies must move from technically impressive AI use cases to ones that are operationally durable, financially justified and ready to scale, says Descartes general manager Dan Cicerchi.

Large language models not only enable unstructured automation, they lead to non-technical people doing semi-productive things, Cicerchi says. “I think that creates a lot of what they call AI slop where you play around — it's kind of fun, but there's no enduring value or outcome.” That’s why with AI it’s critical that companies are outcome-focused and the business side is in alignment. “I think when leaders start considering AI use cases for production investment,” Cicerchi says, “they've got to walk through that scenario and give that careful thought.”

Any technology project is concerned with governance, security and repeatability, but those things are more important with AI because the technology can enable outcomes one isn’t prepared for.

“Certainly, in supply chain we deal with a lot of regulations, so we need to make sure that we can apply guardrails to the solutions,” Cicerchi says. “I think focusing on that allows you to then ultimately move from an R&D project into a productionalized solution or set of use cases that go to market.”

Companies get “stuck” in their AI journey when there hasn't been a clearly defined outcome driven by business value, Cicerchi says. He uses estimations to illustrate his point. "We spend 10% of our team's time every week doing estimations. [What if] we take all of what it took to estimate the last 10 years, take the data and train a model to do that, remove the friction and time, and deliver faster? I think focusing on those value-based outcomes enables you to move from an R&D project or a fun IT project to a highly valuable impact business problem.”