Supply Chains Were Always Probabilistic—AI Is Finally Catching Up

This voice experience is generated by AI. Learn more.This voice experience is generated by AI. Learn more.Ravi Panja is CEO of Soulax.

gettyThree trucking carrier CEOs recently told me they use about 20 percent of the telematics data their fleets produce. Twenty years of wiring trailers, trucks and warehouses, and four-fifths of what those sensors capture is collected but never used.

Picture a control-center dashboard that shows a truck sitting idle at a distribution center. The delivery window is closing and fuel costs are climbing. The system flags a delay but can’t explain it: Is it the weather? A maintenance issue? Maybe it’s something more ordinary like a missed call, a breakdown somewhere in the chain that never made it into the system.​

Now multiply that moment across thousands of shipments, dozens of partners and systems that don’t talk to each other. That’s what a supply chain looks like: a chaotic moving target that never settles into a straight line.​

Enterprise resource planning (ERP) tools gave us structure when things were more centralized. Then planning engines came in and turned supply and demand into math. Workflow tools automated sequences and predictive analytics identified data patterns. All that helped, but it never held together once things got messy, because supply chains stretch across geographies, companies and incentives that don’t always align. Plans change and information is delayed.​

So we built around it with more systems and dashboards. ​Today, every major platform—ERPs, planning suites, now TMS vendors—ships AI agents inside its own application for things like planning, sourcing, dispatch and fulfillment. Each one makes its own system smarter, but think about what a dispatcher needs at 11 PM when a trailer alarm fires. The alert lives in telematics, the load value in the transportation system, the driver’s hours in compliance software, the last service flag in maintenance, and the customer’s rejection policy in an email. No single application owns that decision. A person assembles it, by phone, in minutes. An agent inside any of those systems can’t see the other four.

What’s different now is what we’re willing to operationalize. I saw this last year at a global manufacturer, months after a merger. Two ERPs, more than a dozen planning instances, a data lake that held most of what leadership needed—and 30 spreadsheets per business unit holding the rest. When a senior operator asked about network capacity, the answer existed; it just lived in seven places and assembling it took the planning team weeks. The board wanted it by Wednesday. ​

The instinct is to fix the data first. But in thirty years, nobody has. Today, with AI, you can take those signals as they are, hold them together long enough to make sense of them and act while things are moving.​

Consider a refrigerated freight operator who has 500 trucks and 800 trailers, every one of them wired. On a Sunday night, his phone rings: a trailer carrying $200,000 of cheese is showing temperature drift in Pennsylvania. His dispatcher saw the alert two hours ago but didn’t escalate immediately, because the system throws hundreds of alerts a day and most of them are nuisance​. He has fifteen years of data and three minutes to decide. The data will not help him in the time he has.​

That gap between what the operation knows and what it can utilize is the whole problem. The technology has to work in motion; It has to handle ambiguity without falling apart as soon as something deviates.​

At a beverage bottler running direct-store delivery, thousands of pallets and crates leave the yard every morning and disappear the moment they cross a store’s back door. The operations lead can see how many are in the field; she cannot see which store is sitting on a month’s worth or which route is bleeding assets. A reader on the pallet jack registers each asset in and out—the data is finally there. What was missing was anything that could reason across it while the route was running.​

The systems don’t go away. They remain the reliable, rule-bound layer that most of an operation needs to be. But they stop being the place you go to figure things out. What’s new isn’t any single capability. It’s holding rule-bound systems, probabilistic reasoning and human judgment in one decision loop—precision where the operation must be exact, judgment where uncertainty lives. The dispatcher stops navigating five screens to make one decision. Instead, the system navigates them and brings her the decision.​

There’s a lot of talk about automation replacing people. In supply chains, that’s not where the problem ever was. In fact, the commoditization of AI might be the best thing to happen to humans. ​

The hard part has always been figuring out what’s actually happening. What slows things down is chasing information, reconciling systems, piecing together a picture that never shows up fully. Take that friction away and the judgment is still human; there’s just more room for it.​

We used to write software and deliver it as a service. Today our services need to behave like software: running continuously, embedded in operations, not waiting in a dashboard to be opened. Tomorrow our agents will reason and learn the way experienced operators do. Supply chains are where this is showing up first because they were always the hardest place to make software work. If it holds here—under this much fragmentation and uncertainty—it won’t stay contained.​

Every operator I’ve described was sitting on years of data they couldn’t use. The harder question is why the answer is not another data-cleanup program. We talk about data as the new currency. But coal is not diamonds until it is mined and shaped, and most enterprise data is closer to coal. What it takes to refine it, and why that’s a different problem than cleaning it up, is where this goes next.

​Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?