The situation
A mid-size upstream and midstream operator manages a producing asset base spanning wells, gathering systems, compressor stations, and processing facilities across multiple basins. Asset data lives across SCADA systems, a production accounting platform, a maintenance management system, an EHS database, and an ERP — each with its own entity identifiers, none reconciled. The same compressor station appears under different names in operations, maintenance, and regulatory records.
The organization had deployed AI for production optimization, equipment health monitoring, and certain commercial decisions. The board had asked a specific governance question: are these AI-influenced decisions appropriate, traceable, and producing the outcomes they were designed to produce? The organization needed to answer that question with documented evidence, particularly in jurisdictions requiring documented support for operational decisions affecting environmental and safety outcomes.
The solution
The Knowledge Fabric™ connected the SCADA systems, production accounting platform, SAP PM, EHS database, and regulatory records. Entity resolution matched every physical asset — well, compressor, separator, pipeline segment — across all system identifiers into one canonical Trust-Scored record. Every production reading, condition signal, and compliance status carries a PolyPhaze Trust Score™ and full lineage. The same governed foundation serves both human engineers and AI agents simultaneously.
The Asset Twin capability built the live digital model of the producing asset base. Predictive maintenance runs against condition-based signals and OEM life limits. Compliance status is tracked continuously rather than reconstructed before inspections.
AI Decision Trace established a governed baseline before any AI changes were made, capturing normal variance bands, historical decision patterns, and the pre-AI human oversight baseline including approval steps and override rates. Post-deployment monitoring detected that the production optimization AI had begun adjusting compressor set points at a velocity 31 times the human operational baseline, consistent with algorithmic control. The governance function investigated, confirmed the AI was operating correctly but outside the process safety management system’s human-in-the-loop oversight requirement, and redesigned the oversight protocol before any safety or regulatory event occurred.