Back to Oil & Gas Hub
Downstream

Refinery optimization

One of many downstream use cases. This example focuses on crude blend optimization and product yield, maximizing refining margins by reasoning across feed quality, unit constraints, and market pricing.

The margin squeeze

Refinery margins are determined by the spread between crude cost and product value. Optimizing that spread requires reasoning across crude assay data, unit operating limits, product specifications, and real-time market pricing, simultaneously.

LP models capture some of this, but they're updated monthly and don't adapt to daily feed quality variations or market shifts. The optimization opportunity is measured in dollars per barrel across millions of barrels.

$1+/bbl
Cost savings from AI-driven downstream optimization (McKinsey)
2-4%
Yield improvement from real-time crude blend optimization
$50M+
Annual margin impact for a 200K bbl/d refinery from 1% yield improvement

Why current tools fall short

Traditional approaches were not built for the scale and complexity of modern operations.

01

Monthly LP updates

Linear programming models are updated monthly. Feed quality changes daily. The gap between model and reality costs margin.

02

Disconnected quality systems

Lab data, online analyzers, and tank gauging systems don't feed into a unified reasoning engine.

03

Reactive quality management

Off-spec product is detected after it's made. By then, it's too late to adjust cut points or blend ratios.

04

Experience-dependent optimization

The best crude selection and cut-point decisions depend on experienced planners. Their knowledge doesn't transfer.

The Lumina Approach

How Lumina solves it

Three layers of intelligence working together: reasoning agents, proactive detection, and multi-agent deliberation.

Layer 1

Crude-to-product intelligence

Agents analyze crude assays, predict product yields at various cut points, and optimize blend ratios against current market pricing. Every calculation runs as SQL on your operational data.

Every calculation is SQL you can verify. No black box.

Agent Reasoning
SQL-Backed
Agent Reasoning:

Evaluating Crude Slate Options for March: Option A: 60% WCS + 40% MSW → Diesel yield: 38.2%, Gasoline: 24.1% Option B: 70% WCS + 30% Syncrude → Diesel yield: 36.8%, Gasoline: 26.3% Market spread: Diesel premium $8/bbl over gasoline → Option A margin: +$1.20/bbl vs Option B → Constraint check: Sulfur unit capacity sufficient for Option A blend

All outputs backed by verifiable SQL you can inspect
The Radar
Scanning
Anomaly Detected:

ALERT: CDU overhead temperature trending 8°F above normal for current feed Correlation: Crude TBP curve shows lighter feed than assay predicted Impact: Naphtha in kerosene cut: jet fuel spec risk → Recommend: Adjust kerosene cut point +5°F to maintain flash point specification

Proactive hypothesis testing, like anomaly clustering
Layer 2

Refinery anomaly detection

The Radar monitors process variables across all units: heat exchanger fouling rates, catalyst activity decline, product quality drift, and energy efficiency degradation.

The Radar surfaces issues the operator didn't know to look for. Before they become incidents.

Layer 3

Refinery planning deliberation

Planning, operations, and commercial agents reason together on crude selection, unit scheduling, and product slate optimization.

The output is grounded in facts (SQL results), not hallucination. Every recommendation carries a full audit trail.

The Boardroom
Deliberation
Multi-Agent Debate:

Planning Agent: WCS discount widened to -$18/bbl: increase heavy crude processing Operations Agent: Coker throughput at 95%, limited heavy crude capacity without turnaround Commercial Agent: Diesel crack spread narrowing. Shift yield toward gasoline and petrochemicals → Consensus: Increase WCS to 65% (within coker capacity). Adjust FCC severity to maximize gasoline yield given spread shift.

Agents vote, challenge, and produce a synthesized recommendation

The result: intelligence that scales

Lumina addresses the four strategic problems that hold operators back.

Scale without headcount

Optimize crude selection, blend ratios, and cut points daily instead of monthly, across every operating unit.

Reduce reactive costs

Capture $1+/bbl in margin improvement through real-time crude-to-product optimization.

Knowledge retention

Crude assay relationships, unit operating sweet spots, and planning heuristics are encoded, not dependent on individual planners.

Auditable trust

Every yield prediction, blend calculation, and margin estimate is backed by SQL against your actual lab and process data.

Agents for this use case

Specialized AI agents that power this workflow.

Lumi Refine

Process Optimizer

Specializes in crude blend analysis, cut-point optimization, product yield prediction, and unit performance monitoring.

Lumi Market

Commercial Analyst

Monitors crack spreads, product pricing, crude differentials, and market conditions to inform planning decisions.

Lumi Core

Data Scientist

Handles lab data integration, assay analysis, SPC calculations, and cross-unit correlation.

This is just one of many use cases

Explore what Lumina can do for your operation

Refinery optimization is one example of how Lumina reasons on operational data. Across Oil & Gas, every domain has use cases where AI agents can add value.

Not ready to commit? Stay up to date as we release new capabilities and industry-specific agents.

View all Oil & Gas use cases