Claudio Laterreur, CIO/CDO, Technology and Business C-Suite Executive, Transformation, Turnaround, & Growth, Advisory and Board Member.
Manufacturing stands at a pivotal moment. Organizations have invested billions in ERP modernization, Industry 4.0 initiatives, cloud platforms, advanced analytics and artificial intelligence. Yet, many AI initiatives continue to underdeliver. The problem is rarely the AI itself. In many cases, the issue lies in the data architecture behind it.
For more than three decades, manufacturers have built operational intelligence from the top down. ERP systems became the center of enterprise decision-making, supported by MES and reporting platforms. While these systems brought standardization and discipline, they were never designed to capture operational reality in real time. Instead, they record and summarize events after they occur.
Should organizational intelligence be built from where data is reported, or from where it is created?
Increasingly, manufacturers are looking to machine-generated data as an important source of operational insight.
Many production, quality and operational metrics ultimately trace back to activity on the shop floor. Machines, sensors, controllers, vision systems and process equipment generate much of the underlying operational data that downstream systems use to describe performance. ERP systems do not create that reality; they document interpretations of it, often hours, shifts or days later.
As AI becomes a strategic differentiator, this distinction becomes critical. AI is only as effective as the data it consumes. Models trained on delayed, aggregated, manually entered or incomplete information can produce outputs that reflect the limitations of the underlying data.
The Emerging AI Readiness Gap
Manufacturing operates across three interconnected technology domains:
• Product Technology (PT): Machines, PLCs, CNC controls, sensors and embedded systems that generate operational data.
• Operational Technology (OT): SCADA systems, historians, MES platforms and plant-floor control infrastructure.
• Information Technology (IT): ERP platforms, cloud environments, enterprise applications, analytics tools and AI solutions.
Historically, IT was the bottleneck. ERP implementations were complex and slow, but cloud computing and AI have dramatically accelerated enterprise capabilities. In contrast, many OT environments still rely on event-driven architectures and manually entered data.
The result is a widening gap between what AI can achieve and the information available to support it. The OT-IT gap has become an AI readiness gap.
Why Machine Truth Matters
Traditional manufacturing systems rely heavily on event-based reporting. Production counts are entered at shift end, scrap is recorded after inspection, downtime is logged manually, and energy consumption is often allocated rather than measured.
Machines provide a fundamentally different perspective. Instead of periodic transactions, they generate continuous streams of operational information, including cycle counts, vibration and temperature readings, current draw and many others.
This data provides a more continuous view of operations as they happen.
Machine-centric architectures can help identify quality drift, equipment degradation, throughput losses and inefficiencies closer to when they occur. The difference is similar to diagnosing a patient using historical insurance claims versus monitoring a live ECG.
The Cost Of Two Versions Of The Truth
Consider a hypothetical production line where machine-generated data differs from traditional ERP-driven reporting:
ERP View:
• 835 good parts produced
• 82.1% OEE
• $43.15 cost per part
• Production met plan
Machine View:
• 824 good parts produced
• 74.3% OEE
• $47.82 actual cost per part
• Production below target
The difference does not necessarily reflect inaccurate reporting or intentional manipulation. Operators enter production totals correctly, MES records events, ERP applies standard costing and reporting systems aggregate results as designed.
Yet, the resulting conclusions can be materially different.
Eleven defective parts appear as good inventory. Eighteen minutes of downtime disappear because they fall below reporting thresholds. Standard costs mask actual production expenses. Performance appears healthy even though the line is underperforming.
At enterprise scale, these differences become significant. If that $4.67 per-unit difference persisted across a line producing 1 million units annually, it would represent nearly $4.7 million in additional costs. Inflated OEE metrics can conceal thousands of lost machine hours. AI models trained on incomplete or inaccurate baselines may optimize against conditions that do not fully reflect current operations.
The issue may not be intentional distortion, but differences in how data is captured, processed and reported.
Moving From Approximation To Operational Truth
The solution is not to replace ERP. ERP remains essential for financial management, planning, procurement, compliance and enterprise coordination.
What changes is how organizations establish, validate and use operational data as a source of truth.
A manufacturing operations platform (MOP) can provide a layer for collecting and contextualizing machine data, allowing ERP and other enterprise systems to incorporate more timely operational information. This shifts manufacturing intelligence from periodic reporting to continuous awareness.
Organizations can implement the transition incrementally:
1. Connect machines and operational assets.
2. Contextualize machine data.
3. Apply AI and machine learning to datasets.
4. Explore closed-loop processes where validated insights can inform actions.
5. Where appropriate, progress toward more autonomous optimization across facilities and value streams.
Data Quality Becomes A Leadership Responsibility
Data quality is no longer just an IT concern.
For years, manufacturers accepted data inconsistencies, relying on manual reconciliations and analysts to compensate. AI changes that equation. Machine learning systems institutionalize the quality of their training data. Poor data becomes embedded in forecasts, recommendations, schedules and automated decisions.
Organizations require formal data ownership, governance, accountability and validation at the point of data creation. Leading manufacturers will treat data quality with the same rigor applied to financial controls, quality systems and safety programs.
Connecting Operations To Financial Performance
Machine-centric architectures can create a more timely operational balanced scorecard.
Historically, operational and financial metrics have existed in separate worlds. Operations manages OEE, cycle times, scrap and downtime, while finance focuses on cost, margins, inventory and profitability. A machine-centered architecture can help connect these operational and financial measures to a common set of underlying data.
Machine-level events can, with the appropriate data integration and modeling, be connected to measures such as cost per unit, inventory, labor efficiency and profitability by product, line or facility. Rather than relying solely on retrospective variance analysis, organizations can use more timely operational data to identify and respond to emerging changes. Operational and financial performance become different views of the same truth.
The Strategic Imperative
Manufacturers’ competitive position will increasingly depend not only on ERP and MES capabilities, but also on how effectively they connect operational data to enterprise decision-making. Organizations that establish reliable, timely operational data as a foundation for enterprise intelligence will be better positioned to apply AI and advanced analytics effectively.
Artificial intelligence, predictive analytics and increasingly autonomous operations all depend in part on trusted, well-governed data generated as close to the source as practical.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
