The Wells Have Eyes: AI Sees What Crews Miss

SPE-validated AI well management cut operating costs 5% and lifted output 6% across Canadian unconventional assets

10 Apr 2026

A peer-reviewed case study published in April 2026 by the Society of Petroleum Engineers has confirmed that AI-powered well management delivered a 5% cut in operating costs and a 6% production uplift across Canadian unconventional assets. The findings, published in the Journal of Petroleum Technology, represent one of the most detailed field validations of AI-driven artificial lift optimisation in Canada to date.

The technology at the centre of the study is an Integrated Operations Centre as a Service platform developed by OPX Ai. Rather than replacing legacy equipment or building new control rooms, the system connects to existing field data infrastructure and uses machine learning to monitor wells continuously, flagging problems before they escalate into production losses.

Chevron deployed the platform across its Kaybob Duvernay asset in Alberta. The system detected early warning signs of equipment failures, automatically trimmed unnecessary gas injection in plunger lift operations, and prevented an estimated 71,000 barrels of oil equivalent of deferred production over a single year.

ConocoPhillips applied the same system to its Montney asset in British Columbia, where it identified hidden pipeline freeze-up risks during a harsh winter before human operators had spotted any surface indicators. Within four months, the Montney asset recorded a 3 to 4% production increase above forecast, with operating costs falling by roughly 5%.

The scale of those results is drawing attention across Canada's upstream sector, where producers face sustained pressure to extend the productive life of maturing wells without expanding capital budgets. By shifting teams from constant manual surveillance to exception-based monitoring, the platform allowed individual engineers to oversee significantly larger well portfolios. Separate pilot programmes recorded surveillance efficiency gains of up to 30%.

Whether these results translate consistently across a wider range of asset types and operating conditions remains an open question. The study covers two producers in two provinces over a limited timeframe, and independent replication across different field environments has yet to be published.