Data governance: what it is and how to set it up
Data governance is the set of policies, roles and controls that decide who owns data, who can use it and whether it is accurate. For operational data, it works differently from business data because the data comes from physical equipment, continuously, and often sits outside the rules that IT teams already run.
What is data governance, and how is it different for operational data?
Data governance gives an organization one way to decide how its data is owned, accessed, protected and measured. Business data such as orders and invoices usually flows through enterprise systems with IT-managed controls. Operational data does not. Sensors, controllers and meters produce readings all day, and many of those readings never reach an ERP system at all. Capgemini notes that this data often falls outside enterprise resource planning activities and is not addressed by typical IT data governance.
The gap is also structural. ISA describes operations technology as hardware and software that monitors and controls physical devices and processes. Information technology focuses on storing and moving data. The same report contrasts component lives: IT components typically last 3 to 5 years, with multiple support sources, while OT components can last 15 to 20 years under single-vendor support. Rules written for a laptop refresh cycle do not fit a plant that runs the same controller for two decades.
Operational data therefore needs its own governance scope. Quality rules must reflect how sensors behave, and access rules must reflect the physical risk of disrupting a control system.
What are the core components of a data governance framework?
A data governance framework rests on three things: people, processes and tools. Springer describes the people part as roles, responsibilities, working groups and committees, with data stewardship and data ownership inside it. Processes cover the standards and policies that govern each data domain. Tools make those rules enforceable in the systems that hold the data.
The roles matter most at the start. Dataversity names a data governance council of senior leaders who set strategy and resolve conflicts. It assigns stewardship to whoever maintains data quality and compliance. Owners are accountable for a data domain, while stewards handle the daily work. Before writing any rule, name an owner and a steward for each equipment data domain, and name the council that will settle disputes between them.
Snowflake separates three kinds of document that are often confused. A policy states what is expected, such as who may see a protected dataset. A standard defines how that expectation applies, such as how a tag is classified. A control enforces or verifies it, such as role-based access control, encryption or audit logging. Keep these separate, so that every policy has a standard beneath it and a control that checks it.
How do you set up data governance for operational data step by step?
Start with one site and a small set of equipment. A full plant or fleet at once creates too many owners, tags and exceptions to manage. A single site lets you test roles, rules and tools before you expand. Choose a site where the equipment is well known and the data already flows to someone who cares about it.
Next, name an owner and a steward for each data domain in that scope, such as generator telemetry or pump pressure. Databricks defines data owners as senior stakeholders accountable for their domain. Stewards enforce policy, manage quality and handle access requests. Then build a catalog of the tags in scope, recording each tag's meaning, unit, source and owner. The catalog shows stewards what data exists and who answers for it.
Here is a worked example for one generator coolant temperature tag. The owner is the engineer responsible for the generator. The steward is the maintenance data lead. The catalog entry records the tag name, the unit in degrees Celsius, the source controller and the retention period. The first rules are these:
- Flag the missing-data marker, where the byte carries 0xFF, as no reading rather than a value.
- Flag any reading outside the physical range of the sensor as impossible.
- Flag a value that stays identical across many samples as a possible flatlined probe.
Expand the scope only after the steward has run these rules on real data for a period and confirmed they work.
How do you define quality rules for sensor and telemetry data?
Quality rules for sensor data should test four things: accuracy, completeness, timeliness and precision. Control Engineering asks whether readings reflect the actual process condition. It asks whether data is missing or intermittent, whether values are properly timestamped and synchronized, and whether samples vary only within a sensible range. Write one rule per property for each tag type, so a temperature probe and a flow meter each get the checks that suit them.
Several failure patterns recur in telemetry:
- A sentinel value is a placeholder, such as a fixed number that means "no reading".
- A dropout is a gap where the signal stops.
- A flatlined probe keeps reporting the same value while the process moves.
- Drift is a slow shift away from the true reading.
- An impossible reading falls outside what the equipment can physically produce.
Record provenance alongside each value. Control Engineering defines provenance as the documented history of a dataset, including its origin and any changes since creation. When a flagged value reaches an analyst or a model, provenance shows where it came from and what happened to it on the way.
What security, access and retention rules apply to operational data?
Extend the security management system you already run for IT to cover OT as well. ISAGCA explains that ISA/IEC 62443 asks organizations to coordinate OT security with an existing ISMS. Shared controls protect the enterprise, while OT-specific controls cover its unique needs. The same paper states that OT protection exists to mitigate the physical risks of computer control over a physical process, rather than to protect data for its own sake.
Access and transport rules should reflect that physical risk. Keep data in transit encrypted and send it outbound only, so that no inbound ports need to be open on site. Give each device its own key. Cache readings locally during connectivity outages, and clear the cache only after the cloud confirms it has stored the data.
Set retention per data domain, and write the period into the policy. Record the retention period in the catalog next to each tag, so anyone can see how long a value is kept.
What are common governance mistakes, and how do you know it is working?
Two mistakes recur. The first is leaving operational data without a named owner. When no one answers for a tag, quality problems stay unresolved and the catalog goes stale. The second is writing policies that no control enforces, so the rule exists on paper but nothing checks it.
To tell whether governance is working, track quality metrics for each data domain over time. Databricks describes monitoring quality metrics over time and alerting stewards when a threshold is breached. The useful measures are how often flagged values appear and how quickly stewards resolve them. A falling rate of unresolved breaches, and a catalog that stays current as equipment changes, show the program is holding.
For a team that wants a start with one site, SailPlan offers a demo to begin with one site and one goal.
SailPlan builds the machine-readable data model that makes every AI tool in your stack actually work. Request a demo to see it in action.