Chevron began deploying an AI-driven Integrated Operations Center as a Service (IOCaaS), developed by OPX Ai, at its Kaybob Duvernay operation in Alberta in 2020. Over the first year, the operator recorded a 5% reduction in lease operating expenses (LOE) in the Kaybob asset, and after normalizing for new-drill wells, BOE output was about 6% higher than originally projected. The operator's internal analysis indicated an avoided estimated 71,000 BOE of deferred production and several million dollars in cost savings over a 12-month period, though the article notes that commodity prices and facility debottlenecking also influenced economics.
The article describes Kaybob Duvernay as a large, mature asset with legacy infrastructure. IOCaaS provides a remote, cloud- or edge-hosted operations center that uses existing field data streams, with microservices that interface with SCADA systems, historians and ERP databases, so legacy systems are not replaced. Chevron partnered with OPX Ai to deploy it across dozens of gas wells, associated compressors and a central processing facility. The deployment was phased over about 12 months, and by early 2022 the IOCaaS was fully live, effectively functioning as a remote operations center for the asset.
In the first phase, data pipelines were established from field SCADA and historians into the cloud-based platform, including real-time wellhead pressures, compressor statuses, plunger lift cycles and maintenance work orders from Chevron's ERP. OPX Ai's domain experts worked with Chevron's production engineers to configure the AI models to site-specific conditions, for example tailoring the hydrate model to the field's gas composition and pipeline network. Integration challenges included mapping the myriad SCADA tags and cleaning years of historical data for training, and because IOCaaS overlays running systems, much of the setup occurred in parallel with normal production.
Daily production calls and surveillance meetings were augmented with IOCaaS dashboards that highlight only the wells and facilities that deviate from expected behavior. In one example, the AI flagged an anomalous drop in plunger travel velocity on a single well, a subtle sign of liquid loading, and an operator intervened hours before the well would have accumulated enough fluid to trigger an automatic shutdown. The AI artificial lift optimizer also trimmed plunger gas injection on many wells, saving fuel and compressor runtime.
During implementation, engineers had to manually review and approve each AI-recommended action. As the models proved their accuracy, more decisions were automated, and by the end of the deployment year the operator allowed the IOCaaS to autonomously adjust certain gas lift valve set points and regenerate compressor restarts within safety limits without direct human approval. The article adds that IOCaaS improved surveillance efficiency by up to 30% in separate pilots.