Before publishing a PI dashboard in Oman, write the metric contract

The technical connection from AVEVA PI to Power BI is only half the dashboard. The harder question is whether a field engineer, planner and manager will all understand the same number in the same way. If a production figure changes when a report is refreshed, a tag is reclassified or a downtime rule is altered, the dashboard needs an agreed explanation—not another visual.

For an Omani operator, the useful first deliverable is a metric contract: a short record that turns one operational question into a governed data product. It makes the dashboard reviewable by operations and maintainable by the team that owns the data path.

What belongs in a metric contract?

• Decision: name the operational question and who uses the result—for example, whether a unit met its daily availability target.
• Definition: state the formula, unit, time zone, reporting cut-off, exclusions and treatment of missing, bad-quality or late data.
• Lineage: list the PI asset or tag groups, the aggregation and transformation step, the destination table and the Power BI measure that presents the result.
• Freshness: set the refresh cadence, the acceptable delay, the visible last-refresh timestamp and the action when a refresh fails.
• Ownership: assign a business owner who approves the meaning and a technical owner who changes the pipeline; record the test that must pass before either definition changes.

Keep the historian, pipeline and semantic model in their proper jobs

AVEVA PI is the operational record. A pipeline should shape the small, time-bounded and contextual dataset the decision needs. The Power BI semantic model should hold the shared measures, names and relationships that report authors reuse. That separation prevents every report from quietly rebuilding its own calculation against a raw export.

Microsoft's guidance makes the implementation consequence clear. For DirectQuery and Dual tables, Power Query queries must fold; for large Import models, folding gives the refresh engine less work to do. Incremental refresh also depends on a well-defined time window and a query that can apply it efficiently. In practice, prepare and aggregate PI data in the source-side pipeline, then let Power BI model and present the governed result.

Publish trust, not just access

A promoted semantic model tells colleagues that its owner recommends it for reuse. Certification is a stronger organisational decision: Microsoft reserves it for reviewers authorised by the administrator. Use that distinction deliberately. Start with a clearly described, access-controlled model; reconcile it against a known operational report; then promote or request certification only after the owner signs off on the contract.

The dashboard failure modes are usually ordinary

• Two reports calculate uptime differently because one includes planned downtime and the other does not.
• A successful refresh looks current even though the upstream PI extract stopped several hours ago.
• A new tag or asset-name convention reaches one report but not its executive counterpart.
• An analyst fixes a measure in a personal file, leaving no owner, test or change record for the next shift.

A seven-day first pilot

Choose one metric that already causes disagreement. On day one, have operations, maintenance and finance agree its definition and exceptions. By day three, trace the exact PI inputs and create a small prepared dataset with a visible freshness check. By day five, build one shared semantic model and compare it with the current manual report. On day seven, document the owner, refresh failure response and change rule. Only then publish the first report for wider use. This is a modest piece of engineering, but it is how a dashboard becomes an operating instrument instead of a monthly debate.

If your PI data reaches Power BI but the numbers still prompt arguments, start with one metric contract and one data path. Basira can scope the pipeline, semantic model and acceptance checks around the operational decision that matters first.

1. 01AVEVA PI System — AVEVA
2. 02Endorse your content — Microsoft Learn
3. 03Query folding guidance in Power BI Desktop — Microsoft Learn
4. 04Configure incremental refresh for Power BI semantic models — Microsoft Learn