A recent industry analysis on AI and manufacturing ERP systems makes a point worth sitting with: manufacturers are pulling back on the very software investment AI depends on. Only a third of manufacturers surveyed plan to invest in operational systems this year, a sharp drop from the year before — even as AI raises the bar on what those systems need to deliver. As one data leader put it, companies with fragmented systems don’t get worse AI. They get “confident wrong answers faster.”
That line stuck with me, because I don’t think the real risk is a fragmented system. I think it’s a complete one — a system that does its core job well enough that you stop asking what it’s missing.
Our MES is genuinely good at what it’s built for: capturing production data. Machine states, run times, quality holds, inventory moves — accurate, structured, trustworthy. That’s the “What.” And for a long time, it’s tempting to assume that if the “What” is solid, an AI agent layered directly on top of it will get you the rest of the way.
It won’t. It can’t. Not because the AI is weak, but because the “So What” and “Now What” almost never live inside the system that recorded the “What” in the first place. Why a changeover ran long, whether a quality issue is a pattern or a fluke, what a supplier said last week that explains this week’s shortage — that context lives in conversations, documents, and decisions scattered across the rest of how a company actually works. An AI agent that only sees the production system can answer questions faster. It cannot tell you what to do about the answer, because it was never shown the rest of the picture.
Measuring the Wrong Thing
This matters because it changes how you should judge the tool. A platform-native AI agent should be measured on time saved per query — a real, legitimate, but bounded number. It’s an efficiency gain on work that was already possible. A cross-system AI workflow — one that can see production data alongside the conversations and context that surround it — should be measured on something else entirely: how many of its answers actually led to a changed decision.
Grade a “What” tool by “Now What” standards and it will always look disappointing, because it was never built to clear that bar. Grade a “Now What” workflow by query-speed standards and you’ll undersell the thing that’s actually moving the business. The two layers aren’t competing for the same job. Mixing up which one you’re evaluating is how a genuinely good system gets blamed for a scope problem that was never its to solve.
So What? Now What?
So what: if your company is investing in AI at the platform level and the people level at the same time, are those two investments actually reinforcing each other, or quietly working around each other?
Now what: pick one system in your operation that’s excellent at capturing fact. Before asking it to also hand you the decision, ask what portion of that decision’s real context lives somewhere else entirely — and whether your AI approach can actually reach it.
Source: Manufacturing Dive, “How Advances in AI Are Affecting Manufacturing ERP Systems”
Leave a comment