September 25, 2026, (Inside AI) — Phillips 66, the Houston-based refining giant, has deployed artificial intelligence to forecast equipment failures and squeeze more crude throughput from its plants, according to Tandra Perkins, an executive vice president at the company.
Perkins disclosed the initiative during a conference session in Austin, Texas, on Thursday, marking one of the most specific public disclosures yet of AI moving from pilot purgatory into the core operations of a major U.S. refiner. The company operates 12 refineries and reported $115 billion in revenue last year.
The AI system analyzes sensor data from pumps, compressors, and heat exchangers to predict outages before they happen. That allows Phillips 66 to schedule maintenance during planned downtime rather than suffer unplanned shutdowns, which can cost a refinery millions of dollars per day in lost production.
"We are using AI to predict outages within our refining systems, save on maintenance costs and increase crude processing," said Tandra Perkins, executive vice president at Phillips 66.
Read: AI Tools Challenge Opaque Refined Fuels Trading
Perkins did not specify which AI vendor Phillips 66 uses, how many refineries have the system installed, or the measured financial impact. Inside AI could not independently verify those details.
From Pilot To Profit Center
The disclosure lands as U.S. refiners face mounting pressure from investors to lift margins without building new plants. Environmental permits and capital costs make greenfield refineries nearly impossible to finance. That leaves operators with two levers: run existing units harder or cut unplanned downtime.
AI-based predictive maintenance directly attacks both. A typical 200,000-barrel-per-day refinery loses roughly $2 million for every day of unplanned outage, according to industry estimates. Even a 1% improvement in uptime translates into tens of millions of dollars annually for a company of Phillips 66's scale.
Phillips 66 is not alone. Marathon Petroleum and Valero Energy have both discussed digital monitoring tools in earnings calls over the past two years. But most mentions remain vague. Perkins's comments put Phillips 66 ahead of peers in publicly tying AI to crude processing gains, not just cost avoidance.
The distinction matters. Predicting a pump failure saves maintenance dollars. Increasing crude processing means the AI is helping operators optimize feed rates, temperatures, and pressures in real time. That is a harder technical problem and a more direct revenue driver.
Refining is a low-margin, high-volume business. Small operational improvements compound quickly. A 0.5% increase in throughput at Phillips 66's refining segment, which processed 1.9 million barrels per day last year, would add roughly 9,500 barrels per day of output. At current crack spreads, that is worth more than $300 million annually.
Read: Employees Override AI Systems Even When They Work, Harvard Professor Says
Perkins did not quantify the actual gain. But her decision to highlight crude processing alongside outage prediction signals the company views AI as a margin expansion tool, not a science project.
Why The Silence On Vendors Matters
The absence of a named technology partner is notable. Industrial AI deployments typically involve vendors like C3.ai, Aspen Technology, or Honeywell. Phillips 66 may be building in-house or using multiple suppliers. Either way, the lack of disclosure makes it harder for investors to assess whether the company has a durable advantage or is simply buying the same tools as everyone else.
History offers a cautionary note. Refiners have invested in advanced process control and real-time optimization for three decades. Those systems delivered incremental gains but rarely transformed competitive positioning. AI may follow the same path: useful, expensive, and eventually standard.
What could break that pattern is scale. Phillips 66 operates a diverse slate of refineries, from the Gulf Coast to the West Coast. If the AI system learns across all of them, the company could accumulate operational knowledge faster than smaller competitors. That would create a genuine data moat.
Perkins did not address whether the system shares learning across sites. That detail will determine whether Phillips 66's AI push is a temporary efficiency play or a structural advantage.
For now, the company is keeping specifics close. Investors will get their next look when Phillips 66 reports third-quarter earnings in late October. Any mention of improved utilization rates or lower maintenance spending will be scrutinized for signs that the AI bet is paying off.
The broader signal is clear. AI has moved past chatbots and code generation into the unglamorous machinery of industrial America. Refineries, chemical plants, and pipelines are becoming test beds for prediction engines that never sleep. The companies that learn to trust those engines, and to act on their warnings, may find themselves running circles around competitors who still rely on human intuition and scheduled maintenance.
