The well-by-well workflow remotely changes gas-lift injection-rate setpoints, performs multirate tests, automatically acquires well data, and iteratively uses the test response to select a production-optimizing setting. The abstract says a machine-learning model supplies recommendations on wells without downhole pressure gauges, while production data are quality-checked and uplift is model-tracked using productivity and steady-state tests.
Company
ExxonMobil
Department
Production & field operations
Record type
company_adoption
Disclosed stage
Scaled. JPT/SPE highlight dated 2025-10-01 reports closed-loop ML gas-lift optimization deployed across more than 1,300 ExxonMobil/XTO Permian wells.
Source date
2025-10-01
Source-date precision
day
Source provenance
B · Primary publication
Review disposition
Source-attributed; not an independent validation
Reported result
Operator-authored SPE abstract reports about 2% average oil uplift. JPT’s October 2024 interview reports 1.87% in a three-month human-approval stage and about 2.2% in the next three-month, mostly automatic stage. These summaries are not additive; periods and cohorts are incompletely disclosed.
What this does not establish
The primary paper's accessible abstract states the full paper is unavailable, so full technical validation is not possible. The operator-reported uplift lacks a disclosed counterfactual or confidence interval. Gas-lift performance can drift when water cut or productivity changes; JPT says an existing setting may remain suboptimal for years without frequent monitoring. The control-boundary details and automatic-versus-human exception rates are undisclosed.
This is a distinct ExxonMobil operational workflow, not merely a company-level AI mention. Exact dedup review found AI-022 (ExxonMobil Permian/operating-site AI) to be broad executive commentary without gas-lift workflow, named SPE paper, staged human approvals, setpoint control, 1,300-well deployment, or these results. Related existing record: AI-022. The common company and upstream setting do not establish that AI-022 and SPE 219553 refer to the same initiative; retain as a new, specific case unless a later source crosswalk proves identity.
Closed-loop machine-learning gas-lift optimization across 1,300+ Permian wells
The well-by-well workflow remotely changes gas-lift injection-rate setpoints, performs multirate tests, automatically acquires well data, and iteratively uses the test response to select a production-optimizing setting. The abstract says a machine-learning model supplies recommendations on wells without downhole pressure gauges, while production data are quality-checked and uplift is model-tracked using productivity and steady-state tests.
A source grade describes provenance, not independent verification. This is a research profile, not a measure of net return. Report a correction.