Why Most AI in Refining Fails to Scale and What Works in Real Refinery Operations

Apr 20, 2026   Written by Duncan Manuel

Why Most AI Projects in Refining Never Move Beyond Proof of Concept

Most AI projects in refining stall at Proof of Concept. Not because the algorithms are wrong but because AI in refining is often disconnected from how refineries actually operate.

In many cases, AI models sit on top of fragmented data, lack awareness of operating constraints, and cannot reconcile their recommendations with the physics of the plant. As a result, operators do not trust the output, and the insights never make it into the control room.

This is the critical reason why AI fails to scale in refinery operations: insight is generated, but not embedded into decision‑making.

AI Disconnected from Refinery Physics and Operations

Successful industrial AI must reflect the reality of refinery operations. That means understanding unit interactions, constraints, energy balances, and the first‑principles relationships that govern plant behavior.

Generic AI tools and optimization models struggle in this environment. Without first‑principles, physics‑based models, AI recommendations often conflict with safe operating limits or established operating strategies creating friction instead of value.

This gap between advanced analytics and operational reality is where most AI initiatives break down.

How Physics‑Based Models and AI Create Decisions Operators Trust

KBC works with clients to close this gap, by connecting AI directly to first‑principles models, real‑time plant data, and operational decision‑making so insights can be trusted and acted on in the control room.

By combining over 45 years of refinery expertise with proven core technologies — Petro‑SIM and Visual MESA — KBC embeds AI directly into the operational fabric of the refinery.

This is not AI bolted onto dashboards. It is AI grounded in first‑principles refinery models, enhanced with large language models and connected to real‑time plant data. The result is a closed‑loop operating model that continuously compares actual plant performance against a physics‑based digital representation of how the refinery should be running. Deviations are analysed, root causes identified, and optimization opportunities translated into clear, actionable guidance.

What AI Looks Like in the Control Room

When AI is embedded into refinery models and workflows, it delivers value across the organization:

  • For operators, AI provides real‑time optimization recommendations that respect plant constraints and operating limits, building trust in the control room.
  • For process engineers, it highlights performance gaps, diagnoses root causes, and surfaces improvement opportunities across units and the full refinery.
  • For management, it directly links operational decisions to business outcomes such as margin improvement, energy efficiency, and emissions reduction.

All of this is delivered through natural‑language interaction, making advanced refinery optimization accessible without adding operational complexity.

Why AI in Refining Will Be Won by Integrated, Physics‑Based Solutions

Refineries today face tighter margins, rising energy costs, increasing emissions pressure, and a shrinking pool of experienced operators. In this environment, the ability to make better decisions, faster is not incremental — it is structural.

AI in refining will not be won by generic tools or standalone analytics platforms. It will be won by solutions that combine:

  • Deep process and refinery expertise
  • First‑principles, physics‑based models
  • Real‑time optimization and decision support
  • AI systems that operators trust and adopt

This is what enables digital transformation to move beyond Proof of Concept and deliver sustained value in the plant.

That is where KBC is different and why KBC is becoming the partner of choice for refiners who want AI to deliver in live operations, not just in pilots.