PTQ Q4 2026 Issue

it the most valuable ear- ly-warning signal an operator owns, and the least monitored one in most facilities today. The quieter driver: Fragmentation Underneath both failure modes sits a structural habit that rarely makes the inci- dent report. Most oil and gas organisations still separate engineering, operations, main- tenance, and data into different functions, different systems,

The cost of execution drift One incident – Port Arthur, March 2026

The return on integration Measured outcomes when engineering connects to operations

Engineering study time cycle

Barrels/day of capacity removed from the market 415,000 Separate emission points across a single site 28

157,000 lb

Chemicals released across a 10-day emissions event Of total unit downtime, control room destroyed > 1 month

6 - 8 months (siloed)

3 months (integrated) – 55%

Safety-system capital expenditure

Baseline design

Optimised design – 42%

Figure 3 What execution drift costs when it surfaces and what integration measurably returns when engineering connects to operations

and often different buildings. Then they are surprised when decisions move slowly, and risk pictures arrive incomplete. Across a range of engagements spanning refinery risk- based inspection, front-end engineering design (FEED), liq- uefied petroleum gas (LPG) handling, dynamic quantitative risk assessment, and digital HAZOP, the limiting factor has rarely been a shortage of data. It has been fragmentation: the right information existing somewhere in the organisation but not connected to the people making real-time decisions. This is also why HAZOPs go stale, and control rooms fail at the wrong moment. A study that lives in a document management system, disconnected from the live operating data it was meant to govern, cannot update itself as condi- tions drift. A maintenance history that lives apart from the control system cannot warn an operator that today’s devia- tion matches a pattern from 18 months ago. Fragmentation is not a data problem dressed up as an engineering prob- lem. It is an engineering problem, and it predates every pre- viously described fire. From engineering to operational intelligence The fix is not adding more tools, more reports, or more dashboards sitting on top of the existing system. Most con- versations about ‘AI in oil and gas’ go wrong exactly here. The technology observes the system from above without changing how decisions inside it actually get made. What changes outcomes is connecting first-principles engineer - ing with AI, applied directly to operations, not next to them. The results, where this has been done properly, are con- crete rather than aspirational. In one engagement, connect- ing engineering work to real-time decision-making cut a study timeline from six to eight months down to three. In another, optimising the safety system design delivered a 42% reduction in capital expenditure, not from added tools, but because engineering and operations finally worked from one connected picture instead of two. Production has to continue regardless. What defines performance is how fast an organisation understands its own system engineer- ing becoming operational intelligence in real time, not filed away until the next scheduled review. This is also the honest test to put to any technology ven- dor, including the author’s. Show where the tools actually compress decision time and improve outcomes. Intelligent workflows earn their place when they route a live anomaly

to the engineer who can act on it, with the engineering context attached. Not when they generate another report. Decision automation earns its place when real-time status replaces the morning meeting as the point where opera- tional truth is established. Predictive maintenance earns its place when it demon - strably reduces operational loss and downtime while strengthening compliance, because the model sees the drift a shift log cannot. In uncertain times, what companies actually want is simple to state and hard to build: faster, better-governed decisions. AI that compresses engineer- ing cycle time, prevents process-safety events, and helps operators execute under simultaneous cost, security, and decarbonisation pressure is a resilience-tech strategy. Everything else is software (see Figure 3 ). Discipline before the technology There is a real opportunity in front of refining and petro - chemical leadership right now. However, the sequence matters, and getting it backwards is its own risk. Digitise everything, completely. Every inspection, permit, anomaly, and maintenance action captured in real time, with no gaps and no silos. Partial data produces partial confidence (more dangerous than acknowledged uncertainty). Standardise the operational workflow, not just the doc - ument. Process safety is execution, not paperwork. If two crews handle the same hazard two different ways, the incon- sistency itself is a latent failure waiting for the wrong shift. Build a single source of truth that integrates field data, control systems, and maintenance records into one opera- tional layer. Visibility is the precondition for control, not an accessory to it. Apply AI only once the foundation is clean, typically three to six months of structured, trustworthy data. At that point, predictive models earn their keep, flagging abnormal patterns a shift log would miss and triggering preventive action before a deviation becomes an event. AI does not fix a broken system; it amplifies whatever system it is given, chaos included. Shift the culture from reactive to predictive. An investiga- tion after the fact explains what already happened. The real leverage, the kind that prevents the next Port Arthur, the next Pasadena, the next Dos Bocas, sits upstream of the incident, not downstream of it (see Figure 4 ).

PTQ Q4 2026