AI-103 · Source-linked project profile. Read the scope, reported result and what remains unknown.
Open the interactive projectHow to read this evidenceA narrowly described human-versus-model test in physical gas marketing illustrates a measurable decision-support experiment. Its practical value is possible improvement in how pipeline/storage capacity and local gas positions are marketed, while the disclosure does not say this test became a live execution workflow.
No quantified results disclosed
The test is described as occurring in “last year” relative to August 2025; it may fall inside or outside the 2024-09-26 start of the review window. Exact test date is not disclosed. This is secondary? The transcript is issuer-hosted and primary, so the claim is company-reported; score benchmark is not independently validated. No evidence that trading decisions became automated or that result generalized to other markets.
Concrete test with a crisp comparison, but small scope and hindsight benchmark limit inference; meaningful signal of commercial AI exploration, not proof of a scaled trading system.
Locator not independently checked
CEO Chad Zamarin said Williams hired a new chief information officer with 'a lot of passion around AI' to drive cost efficiency and business transformation, including AI in project execution and analytics in market intelligence. In a month-long internal test marketing gas in Dallas-Fort Worth, young data-analytics staff building AI models beat a 10-year Sequent trader. Williams is also applying data to maintenance strategy.
A source grade describes provenance, not independent verification. This is a research profile, not a measure of net return. Report a correction.