From Unknowns to Business Impact: Why Capturing Data Quality Annotations Matters in PI System Operations

A PI user can notice a data problem before the system can detect it automatically. The observation has little long-term value if it remains in a chat message, notebook, or memory.

Annotations and issue records turn local knowledge into shared operational context.

Capture what the user knows

A useful data-quality annotation can record:

• Affected tag, attribute, analysis, or display

• Time range

• Observed condition

• Operational context

• Suspected cause

• Owner

• Status

• Resolution or disposition

Keep the required fields small enough that engineers will use the process.

Separate observation from confirmed cause

A user may know that a value is wrong without knowing why.

Record the observation first. Mark suspected causes as unconfirmed until investigation provides evidence.

This prevents guesses from becoming permanent technical facts.

Connect the issue to the data object

An issue is more useful when it remains associated with the tag, AF object, asset, or display that it affects.

Future users can then see known limitations during troubleshooting or review.

Prioritize by risk

Use asset criticality, downstream usage, production importance, safety relevance, and age to prioritize the backlog.

Do not create a process in which every annotation becomes an urgent incident.

Close the loop

When the team resolves an issue, record the action and verification result.

If the condition is accepted, document why it is acceptable and when the decision should be reviewed again.

Use recurring findings to improve the system

Annotations can reveal patterns such as repeated instrument failures, poor tag ownership, recurring mapping errors, or specific templates that create quality problems.

Use these patterns to improve monitoring and governance.

The practical objective

A lightweight annotation process preserves operational knowledge that would otherwise disappear.

It helps teams move from "someone knows this value is bad" to a controlled record of what is wrong, who owns it, and what was done.