Management by Exception in Oil & Gas Operations. Alarm only on what matters. Rank by what pays.

Management by exception is an operating model where attention goes only to the assets that have moved meaningfully away from expected performance. Routine, in-spec activity is left to run. In oil and gas it is adopted to improve performance on wells an operator already owns: less deferred production, fewer avoidable failures, lower unit operating cost from the same crew. The idea originates with Frederick W. Taylor's Shop Management (1903), was carried into the modern management canon by Peter Drucker, and reached upstream through SCADA-driven exception alarms in the 2000s.

They are three rungs of the same ladder. Management by exception is the operating model: act only where the data says action is warranted. Pump by exception is the industry's first-generation implementation of it: SCADA thresholds flag wells that deviate, and crews visit what is flagged. Pump by priority is the productized next step: the flagged exceptions are scored in dollars, sequenced into a constraint-aware route for the crews and qualifications actually available, and every field outcome is fed back so next week's ranking is better. An operator can be doing pump by exception and still not be managing by exception, because an unranked alarm list does not allocate the day.

Three things, and most operators have only the first. (1) A definition of normal per asset: a continuously updated decline forecast, an expected pressure band, an expected runtime envelope. Fixed thresholds set at commissioning are not a definition of normal, because wells decline underneath them. (2) A scoring layer that converts each deviation into dollars at risk, so a separator nuisance alarm and a high-rate well drifting off forecast stop arriving with equal urgency. (3) A delivery layer that puts the ranked result in the field team's hands before the shift starts. WorkSync reads the SCADA or historian, production accounting, EAM or CMMS, and GIS systems already in place, read-only, and the ranked plan is live in 4 weeks.

Yes, with a limit worth stating plainly. Exception detection is only as good as the signal, so a well with no telemetry cannot generate a real-time exception. Those wells are still ranked, using what does exist: production accounting volumes, run tickets, gauge and haul history, work-order and failure history, and the decline forecast. They stay on a time-based cadence until better signal exists, and they are visibly labelled that way rather than quietly treated as healthy. Partial coverage is the normal starting condition, not a disqualifier, and the ranking improves as instrumentation grows.

The work does not get heavier, the order changes. Instead of a standing route driven Monday, Wednesday, and Friday, the crew opens a plan that was rebuilt overnight: the wells whose signal moved are at the top with the dollars at risk attached, the wells whose signal did not move are not on the list. Escalation carries the same economic context, so an exception routes to the right person with a number beside it. In the reference deployment at a top 25 private producer, the same crew drove 35% fewer miles (deployment figure) and the operation measured a 15% free cash flow uplift on the same crew (deployment figure).

The advantage is leverage. Attention is finite and wells are not, so pointing the attention at the small set that has moved buys capacity that headcount would otherwise have to buy. It also makes the operation legible: what got worked, and why, becomes a record instead of a habit. The limits are real and worth naming before you build on it. The model is only as good as the definition of normal behind it, so a stale baseline hides problems as effectively as it raises false ones. It says nothing about magnitude on its own, so a deviation worth $90 a day and a deviation worth $12,500 a day arrive with equal urgency until something prices them (illustrative pair). It can starve the routine work that keeps deviations rare, which is why preventive and regulatory work has to compete in the same ranking rather than sit outside it. And it needs signal: an asset with no telemetry generates no exception, so partial coverage has to be visible rather than read as health. Each of those limits is a missing layer above the detection, which is what the ranking, scoring, and feedback layers supply.

Pick the operating metric first and baseline it before anything is connected. Four measures separate a working program from a busy one. Deferred production, because the point of ranking is that the expensive deviations get worked first. Operating cost per well on the same crew, because the model claims capacity rather than headcount. Miles or hours driven per unit of economic coverage, because a plan that removes low-value stops shows up here before it shows up anywhere else. And the share of exceptions that turned out to be worth the visit, which is the only measure that tells you whether the queue itself is improving. At the deployed reference, a top 25 private producer running 5,000+ wells, the measured deployment figures are a 15% free cash flow uplift on the same crew and 35% fewer miles driven for the same coverage.

Four failure modes account for most of it. Alarm fatigue: hundreds of exceptions a shift with no ranking, so crews triage by recency and the highest-value item gets buried. Stale normal: baselines that do not move as wells decline, producing both false alarms and missed real ones. Missing economic ranking: exceptions worked in the order they arrived rather than the order they pay, which is the difference between a busy field and a productive one. No closed-loop learning: when a flagged exception turns out to be nothing, the next identical pattern is flagged at exactly the same priority. All four are ranking and feedback problems rather than sensor problems, which is why buying more instrumentation rarely fixes them on its own.