Mapping the manufacturing data landscape: sources, systems, and silos

Seven systems, one physical batch of parts. Each system was bought at a different time, by a different department, often from a different vendor, with no shared data model. That is how silos are born, not through negligence but through years of reasonable, local decisions.

Where the value leaks out

Silos cost money in three specific ways.

1. Root-cause analysis takes days instead of minutes.

If the vibrating machine caused the dimensional defect, someone has to manually cross-reference the CMMS ticket timestamp with the historian's vibration trend and the QMS nonconformance log. If these systems don't share a common asset ID or timestamp format, that correlation can take an engineer a full day of spreadsheet work. Multiply by hundreds of defects a year.

2. Machine data never reaches the people who could act on it.

A PLC might generate thousands of data points per minute, but if the historian samples it down to one reading per minute and nobody feeds it into a predictive maintenance model, the plant is paying for sensors it doesn't use. Industry surveys (McKinsey, Manufacturing analytics research) have long estimated that a majority of industrial data generated on factory floors is never analyzed. Treat that as a directional estimate, not a precise figure, but the direction is well established.

3. Duplicate and conflicting master data.

The same machine might be called "CNC-04" in the CMMS, "Line3-Mill2" in the MES, and an anonymous tag number in the historian. Without a shared asset hierarchy, joining these datasets requires manual mapping tables that break every time someone renames an asset.

The core datasets that matter

Rather than trying to master every system, focus on five data categories that recur across the sector:

| Data category | Typical source | Update frequency | Primary use |

|---|---|---|---|

| Machine/process data | PLC, SCADA (Supervisory Control and Data Acquisition), historian | Sub-second to minute | Process control, predictive maintenance |

| Quality/inspection data | QMS, LIMS | Per batch/lot | Compliance, defect tracking |

| Maintenance data | CMMS | Per event | Asset reliability, spare parts planning |

| Production/order data | MES, ERP | Per shift/order | Throughput, costing, scheduling |

| Supply chain data | ERP, EDI (Electronic Data Interchange) feeds from suppliers | Daily/per shipment | Inventory, traceability |

Traceability across these five categories, being able to answer "which raw material lot ended up in which finished product, made on which machine, inspected by whom" is the single most valuable integration outcome in regulated manufacturing (aerospace, medical devices, automotive safety components).

Data qualityData qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → metrics that actually matter here

Generic data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → frameworks talk about accuracy, completeness, and timeliness. In manufacturing, translate these into specific, checkable metrics:

• Sensor uptime / data completeness rate: percentage of expected time-series readings actually captured. A historian tag with 60% completeness due to network dropouts will produce misleading trend analysis.
• Master data match rate: percentage of asset IDs that resolve identically across two or more systems (e.g., CMMS vs. historian). Below roughly 90 percent (a commonly cited practical threshold, treat as an estimate) cross-system analytics become unreliable.
• Time-to-reconcile: how long it takes a quality engineer to trace a defect back to its process conditions. This is a proxy metric for silo severity, not a formal industry benchmark, but tracking it internally reveals whether integration investments are working.
• First Pass Yield (FPY) data lag: the delay between production and when FPY (percentage of units made correctly without rework) is visible to the quality team. Real-time FPY dashboards are increasingly standard in automotive and electronics manufacturing.

A worked example: master data match rate

Suppose a plant has 200 registered assets in its CMMS. When cross-referenced against the historian's tag list, only 168 have an exact matching identifier.

Match rate = 168 / 200 = 84%

That 84 percent means roughly 1 in 6 assets cannot be automatically joined across systems, forcing manual lookups for any cross-system query involving those assets. Raising this above 95 percent is a realistic, common target cited by MES/historian integration vendors, and is a good first KPIKPIKey Performance Indicator, a measurable value that shows how effectively you're achieving a specific objective, tracked over time against a target.Voir la définition complète → (Key Performance IndicatorKey Performance IndicatorKey Performance Indicator, a measurable value that shows how effectively you're achieving a specific objective, tracked over time against a target.Voir la définition complète →) for any digital transformation initiative, well before investing in advanced analytics.

Governance: who owns the data model

Data qualityData qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → problems are usually organizational, not technical. A practical governance structure for manufacturing data includes:

• A common asset hierarchy, often modeled on the ISA-95 standard (an international standard for integrating enterprise and control systems), so "CNC-04" means the same thing in every system.
• A data steward per domain: someone in quality owns QMS data definitions, someone in maintenance owns CMMS conventions, and they meet regularly to align.
• A single source of truth per data type: ERP is the system of record for cost and inventory; the historian is the system of record for time-series process data. Conflicts get resolved by rule, not by whoever shouts loudest.

Frameworks like the NIST Manufacturing Extension Partnership resources offer free, practical guidance for small and mid-sized manufacturers starting this governance work without hiring a large data team.

🎬 [VIDEO: "How Industrial IoT and Data Historians Work Together" - youtube.com - search for recent AVEVA or Siemens explainer videos on historian architecture and OT/IT data integration for a visual walkthrough of how sensor data flows from PLC to analytics dashboard]

A simple technical snippet: joining silos with a common key

Even without a full integration platform, analysts often start by writing a join query across exported CSVs using a shared timestamp and asset ID:

SELECT
    h.asset_id,
    h.timestamp,
    h.vibration_reading,
    q.defect_code
FROM historian_export h
JOIN quality_export q
    ON h.asset_id = q.asset_id
    AND h.timestamp BETWEEN q.inspection_time - INTERVAL '2 hours'
                        AND q.inspection_time

This basic pattern, matching machine condition data to a quality event within a time window, is the manual precursor to what predictive maintenance platforms automate. Understanding it helps non-technical managers ask sharper questions of their data teams: "What's our asset ID match rate?" is often more useful than "Can we do AI?"

Key Takeaways

• A single work order touches many systems (PLC, historian, MES, LIMS, QMS, CMMS, ERP), and silos form naturally when these are bought and managed independently over time.
• Track concrete, checkable data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → metrics: sensor data completeness, master data match rate across systems, and time-to-reconcile a defect to its root cause.
• A master data match rate below roughly 90 to 95 percent (industry estimate, verify against your own systems) signals that cross-system analytics will be unreliable.
• Governance, especially a shared asset hierarchy (e.g., aligned to ISA-95) and clear data stewardshipdata stewardshipA business-side owner responsible for the quality, consistency and appropriate use of data in their domain.Voir la définition complète →, fixes more silo problems than new analytics tools do.
• Before investing in predictive maintenance or AI, confirm the underlying data can actually be joined across systems: that foundational step determines whether advanced analytics will work at all.