Designing Data Products To Scale Quality and OEE Across Manufacturing Operations

From Projects to Products

Project-based analytics typically solve a specific problem for a single site. The resulting dashboard may work well locally, but its underlying definitions and data mappings often cannot be transferred cleanly elsewhere.

“Most plants still calculate OEE and quality metrics locally, one line or one site at a time, which means every deployment starts from scratch, and every number means something slightly different depending on who built it,” says Lorenzo Veronesi, associate research director at IDC Manufacturing Insights.

A data product instead starts with a standardized data set that supports maintenance, scheduling, quality management and AI applications. Its schema, calculations and operating expectations are established once and improved centrally.

“This is why platforms built around a common, normalized data structure can scale faster and are better suited for AI,” Veronesi says.

For example, such platforms can generate more reliable AI-driven insight than those relying on custom tagging per machine or per site.

READ MORE: Discover how predictive analytics are helping manufacturers minimize downtime.

One Metric, One Meaning

A manufacturing data product includes more than clean information by establishing an agreed-upon meaning for the data, assigning an accountable owner and defining expectations for its freshness and availability.

“While a dashboard simply shows a number, a data product guarantees what that number means, how it was calculated and that it stays consistent wherever it is consumed,” Veronesi explains.

This requires a defined schema, an owner accountable for its quality and a service-level expectation for timeliness and availability.

Shared downtime taxonomies are especially important for OEE, Veronesi adds. A changeover, microstop or planned outage must carry the same definition across sites. Standards such as ISA-95 provide a common framework for equipment and operational hierarchies.

“It lets a corporate team compare plant performance, benchmark improvement initiatives and trust that any OEE figure derived in one site means the same thing in another,” Veronesi says.

Governance Builds Trust

Reusable data requires an owner to be accountable for accuracy and documentation, rather than leaving those responsibilities solely to a platform team or system architect.

Governance must also cover data freshness, failure handling, lineage and the semantic layer that supplies consistent terminology to every consuming application. Human validation remains important when automated inputs could become accepted as operational fact.

“There must be clear service-level expectations around data freshness and system failure modes, a single semantic layer so that terms and metrics are not reinterpreted differently by each application or AI model, and human-in-the-loop validation steps before automated data is treated as fact,” Veronesi says.

These controls become more consequential as AI agents retrieve manufacturing data and recommend actions, with consistent definitions reducing reinterpretation of raw information and the risk of unreliable results.

DISCOVER: How manufacturers are navigating the convergence of IT and OT.

Platforms Make It Scale

Modern data platforms provide the connective layer for operationalizing data products. They can normalize machine data, manufacturing execution system data and enterprise resource planning data into common models while recording where information originated and how it changed.

Lineage helps users determine whether a questionable result came from a sensor, transformation rule or source application. With centralized quality rules, a corrected calculation can flow to every authorized dashboard and AI system.

“This includes automated normalization of machine and ERP data into a common model, tracking of where each data point originated and how it was transformed, and a semantic or knowledge layer approach so that AI tools query defined, structured data rather than reinterpreting raw inputs each time,” Veronesi says.

The transition requires manufacturers to fund data products as lasting operational capabilities, with assigned owners and measurable reliability, rather than as temporary analytics projects.

It’s a foundation that can make expansion faster without sacrificing consistency as companies add plants, use cases and AI tools, Veronesi explains.

“Vendors are increasingly building this connective layer as core infrastructure, to enable their customers with a data vantage point that allows OEE and quality data sets to be reused reliably as manufacturers expand from one line or plant to many,” he says.