We wrote about data governance in ERP systems back in 2024, and the core argument hasn’t aged: an ERP is only as good as the master data feeding it, and most manufacturers treat that data as somebody else’s problem until it breaks a production run. What’s changed since then is the stakes. Manufacturers have spent the last two years pointing AI at the same ERP systems — for demand forecasting, supply chain planning, process optimization — and the governance gaps that used to just slow down a monthly report are now feeding decisions an algorithm makes automatically, at a pace no one is manually checking.
AI Adoption Outran the Data It Runs On
Rootstock Software’s 2026 State of Manufacturing Technology Survey, fielded by Researchscape among 520 mid- to large-sized manufacturers across North America, Europe, and Asia, found that 94% now use some form of AI, and 73% believe they’re on par with or ahead of their peers in adoption. The categories seeing the sharpest growth are the ones most dependent on clean ERP data: predictive AI adoption climbed 12 points to 48%, supply chain planning AI investment jumped 19 points to 35%, and process optimization AI rose 11 points to 36%. Those are exactly the use cases that fail quietly when the underlying part numbers, bills of material, or inventory records are wrong — the forecast still runs, it just runs on bad inputs.
COOs Are Naming the Gap Themselves
McKinsey’s COO100 survey, fielded June–July 2025 among 101 chief operating officers at manufacturers with $1 billion or more in revenue, put a number on where that breaks down. Forty-six percent of COOs report meaningful limitations in their data or IT/OT systems — 19% point specifically to outdated infrastructure, and 18% name poor data quality itself as the bottleneck. A quarter of respondents said they struggle to build AI applications that are reusable across facilities, which is a data architecture problem as much as a technical one: a model trained on one plant’s master data doesn’t transfer cleanly to a second plant running on a different naming convention or a different item master. Rootstock’s survey found a similar signal from the operations side, with 49% of respondents prioritizing platform consolidation specifically to reduce data silos, and roughly a third citing talent gaps and cross-department collaboration as barriers to getting more value from the systems they already have.
Governance Is Also a Risk Control, Not Just a Quality Issue
It’s worth separating the AI-project failure conversation from the data-quality conversation, because they’re related but not identical. Gartner has predicted that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls as the leading causes — not data quality specifically. But risk controls and data governance sit close together in practice: an AI agent making purchasing or scheduling decisions off ERP data needs the same kind of guardrails — who owns a record, what triggers a review, what’s auditable after the fact — that a mature data governance program already provides. Manufacturers that never built those guardrails for their master data are, in effect, deploying agentic AI without the risk controls Gartner is pointing at.
What Governance Actually Looks Like in an ERP Context
The 2024 version of this argument focused on the basics: a defined system of record for each data domain, validation rules at the point of entry, and a named owner for each master data type — item master, BOM, vendor, customer. That still holds, and it’s still where most manufacturers are behind. What 2026 adds is urgency around a second layer: change control and auditability for the records an AI system is allowed to touch. If a forecasting model or a purchasing agent can update a reorder point or flag a supplier exception on its own, someone needs to be able to answer, after the fact, which record changed, when, on what basis, and who’s accountable if it was wrong. That’s not a new discipline — it’s the same change-management thinking manufacturers already apply to the physical production line, applied to the data line instead.
Our Verdict
The gap between AI adoption and data readiness in manufacturing isn’t a forecast anymore — it’s something COOs are naming directly, at 46% and rising. Ninety-four percent AI usage against 18% of leaders naming poor data quality as an active bottleneck means a lot of manufacturers are running production AI on ERP data they wouldn’t sign off on for a board report. Fixing that isn’t glamorous work — it’s system-of-record ownership, validation at the point of entry, and change control on top — but it’s the difference between AI that compounds good ERP data into better decisions and AI that compounds bad data into worse ones, automatically and at scale.
If you want a clear read on how ready your ERP data actually is before you point more AI at it, that’s a conversation we have with manufacturing clients regularly.