Pump by Exception vs Pump by Priority.Two operating models, one question apart.

Pump by exception asks which wells have a problem: SCADA thresholds trip alarms, alarms generate callouts, and the field responds to the exception list. Pump by priority asks which problem is worth solving first: every exception and every open task is scored in dollars, the scored list is sequenced into a constraint-aware route, and every field outcome retrains tomorrow’s ranking. The exception is not the decision. It is the input to the decision.

Management by exception is the management philosophy: leadership and field attention go to the deviations that matter rather than to routine confirmation that things are normal. Pump by exception is the oilfield’s first-generation implementation of that philosophy, built on SCADA alarms and callouts. The philosophy is sound; the first implementation struggled because it stopped at detection.

Three failure modes repeat almost universally. Alarm fatigue: static, conservatively set thresholds produce hundreds of exceptions a day with no ranking, teaching the field that most alarms are noise. No economic context: all exceptions look equally urgent, so crews work them in windshield order. No routing and no learning: an exception list is not a drivable day, and the thresholds never improve. The failure was the implementation stopping at detection, not the principle.

No. Some SCADA packages ship a prioritization feature that ranks a down-well list by production impact against target. That is useful, and it is not the operating model: it ranks volumes rather than dollars, it hands the field a sorted list rather than a constraint-sequenced route, and the ranking does not learn from what the crew found. Three test questions: is the priority a dollar figure a controller would defend; does the output arrive as a routed day in the truck cab; and does yesterday’s field outcome change tomorrow’s ranking?

When the binding problem is that nobody knows a well is down until the next calendar visit, detection alone is real progress: an operator with no alarm discipline should fix that first. But detection should be priced as detection. There is no published case we know of where detection alone moved LOE or deferred production at portfolio scale; the published full-loop results, including the 15% FCF uplift on the same crew at the 5,000+ well deployed reference, all come from systems that price and route the exceptions, not just flag them.

Keep your SCADA. Keep your historian. Keep your exception system. The migration adds the missing layers on top: economic scoring on the exceptions you already surface, constraint-aware routing that turns the scored list into a drivable day, and closed-loop feedback from the field. The productized path runs the full loop thin on one field in four weeks against one signed metric, then widens field by field.

Field operations has moved in four generations. Generation one, fixed routes: visit every well on a cadence. Generation two, pump by exception: visit wells when something changes. Generation three, pump by priority: visit wells when the economics justify it. Generation four, intelligent operations: the system continuously decides what the operation should do next, running the full loop of flag, price, rank, route, execute, learn. Each generation answered the previous one’s binding constraint, and pump by priority is the step that makes the fourth reachable.