step. A faster, unsafe approval is not an improvement. A closed case without proof is not closure. A parallel review is useful only when the reviews are genuinely independent. Where one decision depends on another, the sequence must remain. This guardrail is important because speed can become the wrong target. If teams are measured only on case age, they will learn to close cases administratively. The better measures pair time with quality: recurrence, reopened cases, overdue high-consequence actions, temporary changes past expiry, and the percentage of closures supported by the required verification. American Petroleum Institute Recommended Practice 754 (3rd Edition) provides a framework for process safety perfor- mance indicators in refining and petrochemical operations. Exception measures sit underneath those indicators. They show how quickly and reliably the organisation responds before an abnormal condition becomes a more serious event. A structural lesson from Deer Park Incidents deserve more than a passing reference in an article about technology. They involve people, families, and com- munities. The right reason to study them is to understand the structure that allowed the event to occur. On 10 October 2024, contract workers at the PEMEX Deer Park Refinery in Texas opened a piping flange containing pressurised hydrogen sulphide (H 2 S) instead of the intended nearby isolated segment. The US Chemical Safety and Hazard Investigation Board (CSB) reported that more than 27,000 pounds of H 2 S were released. Two contract workers died, and 13 others were transported to medical facilities. The CSB did not describe a sensor problem. Its final investi - gation identified failures in positive equipment identification, work permitting and hazard control, turnaround contractor management, and conduct of operations. The structural lesson is clear. The work took several forms: drawings, flange lists, tags, permits, field activities, contrac - tor supervision, and operating responsibility. The safety of the task depended on those elements agreeing at the point of execution. An exception management layer would not by itself guar- antee that agreement. No software can substitute for correct equipment identification, field verification, competent super - vision, or stop-work authority. However, a well-designed governance layer can make dependencies explicit. It can require proof of positive identification before authorisation, ensure that deviations from the work package are returned for review, connect contractor assignments to the correct unit state, and preserve a single record of who verified what and when. The lesson is not that a structured workflow would have prevented a specific incident. The right way to look at it is that safety depends on the integrity of the whole chain, especially at the handoffs. Start with one exception that matters Many technology programs become too large before they become useful. Exception management should begin differ- ently. Choose one exception class that is frequent enough to measure and important enough for operations to care about.
An unexpected turnaround discovery is a good example. A temporary operating change, an overdue mechanical integ- rity recommendation, or a recurring equipment anomaly can also work. Map what happens today, including the unofficial path. Where is the data re-entered? Which decision waits for a meeting? Who chases the approval? What happens when the normal approver is unavailable? Where does the next shift get its context? What evidence is required before the case can be closed? Then define the operating path: trigger, accountable owner, consequence-based response time, required technical context, decision authority, and the point at which MOC applies. The edge cases will take most of the discussion, which is normal. Connect only what the first exception class needs. Do not turn the pilot into a replacement program for every underly- ing system. Run it across day and night shifts. Test the absent approver, the false signal, the incomplete work package, the integration failure, and the need for emergency action under existing procedures. Measure the full exposure window before and after, and separate technical work from waiting. Check whether the closure evidence is stronger, whether there is recurrence, and ask the operators whether the case helped them act or simply gave them another screen. If the pilot works, scale the pattern, not the form. The common pattern is qualification, ownership, context, decision, execution, verification, and learning. The technical rules will remain distinct for product quality, mechanical integrity, emissions, work control, and equipment reliability. Where should the next tech investment go? Chief information officers should not jump straight into mak - ing big investments in better AI and data gathering. Do not judge the next investment solely by the model’s intelligence. Ask how the operating system handles its output. Look deeper at how staff are engaging with existing systems. Do they fight with it? Do they ignore information? How can key human roles be better equipped so that they can handle the essential work of exception management? The next technology dollar should go where it shortens the distance between a valid signal and a verified response, without weakening a single operating control. Detection and response are not competing investments. AI and exception management are not competing ideas. One makes the signal sharper. The other makes the organisation act on it with dis- cipline. The governance layer is what makes the technology pay off. Reference 1 US Department of Commerce National Institute of Standards & Technology, Artificial Intelligence Risk Framework (AI RMF 1.0) report: https://doi/org/10.6028/NIST.AI.100-1 Dinesh Varadharajan is the Chief Product Officer at Kissflow, oversee - ing product management, engineering, design, and DevOps. He is driv- ing Kissflow’s evolution into an AI-first, low-code platform, redefining how enterprises build and operate internal systems. Varadharajan was recognised as one of the Top 100 Great People Managers (2020) by the Great Manager Institute in association with Forbes India.
91
PTQ Q4 2026
www.digitalrefining.com
Powered by FlippingBook