The most common reason AI pilots fail in manufacturing has nothing to do with the AI. It's that the operational data the model needs — production counts, cycle times, quality events, downtime records — never made it out of the plant floor in the first place.
Rockwell Automation's research puts a number on the problem: half of manufacturers still rely on legacy operational technology (OT) systems, and many of those systems connect to enterprise IT through fragmented, custom-built middleware that is expensive to maintain and unreliable as a data pipeline. The 2026 State of Smart Manufacturing report from Rockwell Automation identifies secure, integrated IT/OT architectures as foundational to scaling AI and advanced automation — not optional infrastructure, but a prerequisite.
For mid-market manufacturers, this is the decision that precedes every other digital transformation decision. You cannot forecast demand from production data your ERP has never seen. You cannot train a quality model on inspection records that live in a disconnected MES. The integration gap is where data readiness fails.
What Rockwell's Research Actually Says
Rockwell Automation's 2026 research describes a manufacturing landscape still divided between organizations that have achieved connected operations and those still managing siloed systems through point-to-point integrations. According to Rockwell's post on fragmented systems and orchestrated operations, the challenge of siloed solutions is compounded by underlying system complexity — many organizations still rely on legacy OT systems, and the middleware connecting those systems to enterprise IT is often costly, brittle, and custom-built.
The statistics cited — including figures on data utilization rates, AI augmentation percentages, and cyber incident rates — come from Rockwell's own survey methodology and should be treated as vendor-commissioned research rather than independent verification. The directional finding is consistent with what operations leaders at mid-market manufacturers report in practice: shop-floor data collection is fragmented, integration maintenance consumes IT budget, and the ERP rarely has a complete, real-time picture of what is happening on the production floor.
The operational implication is concrete. If your ERP's production planning module is working from yesterday's batch export rather than live machine data, your inventory accuracy, BOM costing, and fulfillment commitments are all running on a lag. That lag is manageable when demand is stable. It becomes a margin and delivery problem when it isn't.
Why This Matters for Manufacturing Operators
The ERP is only as useful as the data feeding it. For manufacturers running legacy OT systems — PLCs from the early 2000s, MES platforms that predate cloud architecture, SCADA systems that were never designed to talk to an enterprise database — the ERP is often receiving a curated, delayed, and incomplete version of what is actually happening in production.
This creates three compounding problems.
Production planning runs blind. When cycle time, downtime, and yield data arrive in the ERP hours or days after the fact, planners are scheduling against stale numbers. Capacity assumptions drift from reality. Customer commitments get made on inventory positions that no longer exist.
AI and analytics investments hit a wall. Demand forecasting, predictive maintenance, and quality analytics all require clean, standardized, timely operational data. If that data is locked in a disconnected MES or captured only on paper, no analytics platform can compensate. The model is only as good as what it can see.
Middleware becomes a hidden cost center. Custom integrations between legacy OT and enterprise IT require ongoing maintenance. Every firmware update, every ERP version upgrade, every new data field creates a potential break. At mid-market scale, where IT teams are lean, that maintenance burden crowds out higher-value work.
Where the Exposure Shows Up
The IT/OT gap surfaces in specific, recognizable ways. If any of the following describe your operation, the integration gap is already affecting your ERP's reliability as a system of record.
- Production data is batched, not real-time. Your ERP receives end-of-shift or end-of-day production summaries rather than event-level data. Inventory and WIP records are always slightly wrong.
- Quality events are recorded outside the ERP. Inspection results, nonconformances, and rework records live in a standalone quality system or spreadsheet and are reconciled manually — if at all.
- Downtime is tracked on the floor but not in the ERP. Maintenance and operations know which machines are running; finance and planning do not, until someone manually updates a record.
- Middleware is maintained by one person. If a single IT staff member or outside contractor is the only one who understands how shop-floor data gets to the ERP, that is a single point of failure — for operations and for any future transformation project.
Each of these is a symptom of the same root problem: the ERP and the shop floor are not operating from the same data.
What to Check Before Committing Budget
The audit that matters here is not a technology evaluation. It is an inventory of what you have, what it connects to, and what data actually reaches the ERP. Work through these six questions before committing budget to any analytics or AI initiative.
- 1. OT system inventory. Document every MES, SCADA, and PLC system by vendor, version, and deployment date. Note which systems are vendor-supported and which are end-of-life or running on unsupported configurations.
- 2. Data collection coverage. Map which machines, processes, and quality checkpoints are instrumented and which are manual or paper-based. Identify the gaps — these are the places where operational data disappears before it can reach the ERP.
- 3. Middleware and integration documentation. List every custom integration, API, or data transfer job connecting OT systems to the ERP. For each one, record the vendor or developer, the maintenance owner, the data latency, and the last time it was tested after a system update.
- 4. Data quality and completeness in the ERP. Pull a sample of production orders from the last 90 days and verify whether actual production counts, cycle times, quality events, and downtime records are present and accurate. If the ERP shows only planned quantities and standard costs, the integration gap is already affecting your financial reporting.
- 5. IT/OT network architecture. Confirm whether shop-floor systems are on an isolated network segment or integrated with enterprise IT. Identify whether that architecture was designed intentionally or evolved organically — and whether it creates security exposure as systems become more connected.
- 6. AI readiness baseline. Before evaluating any AI or analytics tool, confirm whether the operational data that tool would need is clean, standardized, and accessible. If the answer is no, the integration gap is the first project, not the AI tool.
What to Watch
SAP has stated publicly that operational data must flow into the ERP before AI can act on it — an architectural assumption that only holds if the shop-floor-to-ERP connection is reliable and current. Mid-market manufacturers without that connection are excluded from that architecture by default, not by choice.
Customer audits, regulatory compliance, and supply chain visibility requirements are adding pressure on manufacturers to demonstrate data governance and traceability. An ERP that cannot see real-time production data cannot produce reliable traceability records, regardless of how well the business-side data is managed.
Monitor ERP platform updates that expand native OT connectivity — particularly around MQTT, OPC-UA, and edge computing integrations that reduce dependence on custom middleware. The cost of standardized integration is declining relative to the ongoing maintenance burden of custom point-to-point connections, which shifts the build-versus-buy calculation for lean IT teams.
Bottom Line
The IT/OT integration gap is a data governance decision that determines whether your ERP is a reliable system of record or an expensive planning tool running on incomplete information.
Manufacturers who close this gap — by replacing legacy OT systems, standardizing data collection protocols, or investing in edge computing to bridge the connection — gain something more valuable than AI readiness. They gain an ERP that actually reflects what is happening in production. That is the foundation every downstream investment depends on.
The audit comes first. The AI investment comes after.