A vendor argument that cuts against the rush to deploy agents
Infor published guidance this week making an argument that runs counter to the pressure most manufacturing CIOs are currently under: before deploying AI agents, fix the ERP data foundation those agents will depend on for every recommendation they generate. The company positions its CloudSuite Industrial Enterprise, paired with Velocity Suite, as the reference architecture for that foundation, built specifically around metal fabrication's industry-specific requirements rather than a generic manufacturing template retrofitted for the sector after the fact.
The argument is notable partly because it is not the easy sales pitch. Most AI vendors in this category lead with agent capability and speed to value, not with a message telling prospective customers to slow down and fix their data infrastructure first before anything else gets deployed. Infor's bet is that fabricators who skip the foundation step end up with agents that make confidently wrong recommendations, a worse outcome for the vendor's reputation over time than a longer, more deliberate sales cycle up front. It is also a bet that customers burned by earlier AI pilots that failed quietly will now value that candor more than a faster time-to-demo.
The four-step sequence Infor says cannot be skipped
Infor's recommended path runs in a specific order: first establish connected, trusted data infrastructure across the operation, then use process mining to identify where workflow problems actually sit rather than guessing based on anecdote, then introduce automation against those identified problems, and only then expand AI agents on top of the resulting foundation. Infor's own language on the data step is direct: connected, trusted data serves as the basis for applying AI to business processes, a prerequisite rather than an optional add-on layered in later.
Skipping straight to agent deployment, in Infor's framing, does not eliminate the underlying data problems, it moves them downstream into the AI layer where they become harder to diagnose and more expensive to unwind later. A fabricator with fragmented inventory data will not get better inventory decisions from an AI agent. Instead they get an agent that produces plausible-sounding recommendations built on the same fragmented data, arguably a worse failure mode because the output looks authoritative even when it is wrong, and because a confident wrong answer is harder for a busy planner to catch than an obvious data gap.
Why metal fabrication specifically makes this hard to fake
Infor's case leans on numbers specific to the sector: raw materials represent roughly 60% of total costs in metal fabrication, which means material pricing accuracy and dimensional inventory tracking function as the core determinant of whether a given job is even profitable, not a peripheral ERP feature nice to have someday. An AI agent recommending pricing or scheduling decisions without accurate access to those figures is working from the wrong foundation on the single largest cost line in the business, which makes the stakes of getting the data layer right considerably higher than in most other manufacturing subsectors.
Fragmented legacy systems, common across metal fabrication given the industry's long equipment lifecycles and history of point-solution purchasing, create barriers to both modernization and AI adoption simultaneously. Infor's Agentic Orchestrator, built to maintain traceability across automated actions, is positioned as the governance layer that keeps agent decisions auditable once the foundation work is done, addressing the accountability question that tends to surface only after something has already gone wrong on the shop floor.
The technical debt argument CIOs should take seriously
Infor's most useful framing describes how AI changes the cost of technical debt, a point that extends well beyond ERP feature comparisons. Legacy integrations and workarounds that were tolerable as maintenance annoyances become active constraints once an AI agent's output depends on the data flowing through them cleanly. A workaround nobody prioritized fixing for three years suddenly matters because the agent built on top of it inherits every one of its quirks and blind spots without knowing they exist.
That reframing is worth applying outside metal fabrication specifically, across any industry where AI agent deployment is now the forcing function for cleanup work that lacked an obvious business case before. Any CIO who has deferred a data cleanup project for lack of a clear return should revisit that backlog now that agent deployment supplies the business case directly. Integrations and data gaps that were previously someone else's problem to eventually fix become the CIO's problem the day an agent starts making decisions on top of them, and by then the cost of fixing the underlying data is usually higher than it would have been a year earlier.
What this means for your own AI sequencing
Infor is a vendor selling ERP software, so its advice to fix the ERP before deploying AI carries an obvious self-interest, and CIOs should read it with that context in mind rather than at face value. But the underlying sequencing argument, data foundation before process mining before automation before agents, holds regardless of which vendor happens to be making it, and it matches the pattern we have seen succeed at other manufacturers this year.
The practical test for your own organization is simple: before approving the next AI agent pilot, ask whether the data it will depend on has been audited for the specific gaps that matter to your industry, not manufacturing in general terms. If that audit has not happened, the pilot is likely to produce results that look impressive in a demo and fall apart against real operational data, a worse outcome than accepting the delay of doing the foundation work properly first.
