A pressure deviation began appearing in a DCS historian trace at 14:32. The operator acknowledged the alarm at 14:34, but the first corrective action came at 15:01. By then, the deviation had spread through two downstream control loops and recovery required a production rate reduction. The event data came from a site in our pilot program. Its 27-minute gap from first signal to effective action is longer than usual, but the pattern is not unusual in process operations.
What the 90-second threshold means
Industrial incident research repeatedly places a threshold at about 90 seconds between the first anomaly signal and operator action. Passing it does not mean an event will always become an incident, nor that a response within 90 seconds will always resolve it cleanly. Rather, containment becomes materially harder as the condition develops. The first 90 seconds can allow a small correction to contain a deviation that would demand major intervention if it continues.
The 90-second figure comes from reviews of process safety incidents where data showed the initial anomaly before operators acted, allowing investigators to calculate when containment shifted from a minor intervention to a major one. It describes a change in intervention cost, not a physical limit on process behavior. Each process has its own effective threshold, but intervention costs generally increase as a deviation persists.
Four reasons the gap develops
The interval from the first anomaly signal to effective action has multiple causes, and no single intervention addresses all of them.
First comes alarm delay. SCADA and DCS systems commonly use deliberate delays to limit chattering, so an alarm waits until a condition persists, usually between 10 and 60 seconds. This engineering tradeoff lowers alarm volume at the expense of speed. Where alarm rates are high, settings tend to be cautious, causing genuinely diagnostic events to appear later than necessary.
Second is prioritization. Depending on the process and site alarm practices, operators commonly handle between 20 and 200 alarms per shift. A developing deviation must compete with standing alarms, less urgent process changes, and equipment problems already in progress. Someone accustomed to dismissing frequent low-consequence alarms may fail to give a new one the urgency it deserves.
Third is interpretation. Promptly attending an alarm does not instantly explain its significance. Which equipment produced the reading? What is normal for this batch stage? Does it match a familiar pattern or signal something new? On a complex process unit, especially at a measurement point the operator rarely uses, interpretation may take several minutes before action can begin.
Fourth is the decision delay. After identifying the situation, the operator must choose a response. A familiar event covered by a trained procedure is quick; an ambiguous or novel event, or one between procedures, takes longer. Experienced operators generally do this faster, making their departure a real process safety concern during workforce transitions.
From 90 seconds to 27 minutes
The 27-minute interval above was not a period of inaction. The operator acknowledged the alarm at 14:34, a normal response for a non-critical classification, then investigated, checked related readings, reviewed the unit procedure, consulted the shift supervisor, and made a cautious first correction at 15:01. That step did not contain the deviation. The operator took the next action 12 minutes later, after determining that the expected response had not occurred.
In practice, the gap is often a chain of sensible decisions rather than a refusal to act. The operator recognized the problem within two minutes of the first alarm, but diagnosis and choice consumed the window in which less costly interventions were available. Meaningful action arrived 27 minutes after the initial signal because the process could not afford the time those steps required.
How decision support changes calculation
Decision support is most useful in the interpretation and decision portions of the gap. Alarm acknowledgement depends on operator attention and cannot be compelled by a recommendation system. Once attention is on the anomaly, moving from "I see this alarm" to "I know what action to take" is where a well-designed recommendation can produce measurable benefit.
A recommendation such as "pressure deviation on the overhead condenser outlet: check coolant flow rate, likely cause is cooling duty reduction, suggested initial action is to verify coolant pump running status and increase setpoint by 5%" does more than the acknowledge button. It shortens the reasoning that follows, letting the operator check the suggested cause against their own view and act without rebuilding the causal chain.
The recommendation does not replace operator judgment. An operator with contrary evidence or relevant context, such as knowing the coolant pump just returned from maintenance, can override it and record the reason. It provides a starting point for assessment. Its benefit is compressing the path from alarm to considered action, not carrying out the action automatically.
Decision latency is a system design issue
Decision latency is often framed as an operator metric: how quickly does each person respond to alarms? The measurement can be made at the individual level, but that is not where the main intervention lies. Response times vary by person, while systematic causes sit in alarm design, procedure quality, and the information shown at the decision point.
The sites we have worked with that move from decision to action fastest do not simply employ faster operators. They have rationalized alarm environments, clear procedures for common deviations, and orderly shift handovers. Individual skill matters at the margin; system design establishes the baseline.
The 90-second threshold applies to the operational system, not as an individual performance score. Lower latency comes from improving the conditions around operators, rather than asking them to move faster inside a poorly designed system. These are separate problems, and treating them as one leads to ineffective remedies.