Multi-Well Pad Interference Rate (MWPIR)

GMI-OG-011Oil and GasIntegration-Required

Definition

Multi-Well Pad Interference Rate, or MWPIR, is the percentage of pad-level AI production-optimization adjustments where a beneficial change on one well is followed by a measurable decline on at least one offset well on the same pad inside a defined observation window. The metric exists to answer one decision: is the optimizer''s local rationality producing systematic neighborhood damage that aggregate field metrics are hiding?

The Measurement Gap

Trent runs production operations for a Permian operator with twelve four-well pads spread across Lea and Eddy counties. Last year his company deployed an AI production optimizer that adjusts ESP frequency, gas-lift injection rate, and choke positions in real time against a per-well decline model. The platform produces hundreds of adjustments per day across the pad inventory, each one calculated to improve barrels-per-day on the specific well it acts on. The first six months looked good: aggregate field production was up 4.1% against the pre-AI baseline, and the optimizer''s ledger of individual-well improvements added up to 220 incremental barrels per day across the fleet.

Then Trent''s reservoir engineer flagged something. The middle wells on three of the pads were trending down faster than their decline curves predicted, and the falloff started about ninety days after the optimizer went into full production mode. Pulling the data well by well, he found a pattern. Every time the AI raised production on a pad''s flank wells (boosted the ESP frequency on Well 1 or Well 4) the interior wells, 2 and 3, would show a 6–9% rate decline over the following five to seven days that the optimizer had no model for. The optimizer''s ledger credited the flank-well gains. Nothing in the system debited the interior-well losses. The 220 incremental barrels per day on the credit side were partially offset by 140 barrels per day in cross-well damage that lived in the production decline of wells the optimizer had never been told to think about.

The damage mechanism is not exotic. On a tight pad with wells producing from the same reservoir interval, the interior wells share pressure support with their neighbors. When the optimizer pulls harder on a flank well, the pressure boundary for the interior well changes. The interior well''s productivity index doesn''t fail; the reservoir voidage geometry just shifts in a way that the per-well optimization model can''t see because it has no neighbor in its objective function. The AI is making locally rational decisions that are globally suboptimal, and the structure of the optimizer guarantees the pattern will repeat.

Legacy production metrics do not catch this. Field-level production is up: the optimizer is delivering on aggregate. Per-well economics looks fine on the wells the optimizer is touching. The standard production dashboards report what the optimizer was asked to optimize. They do not report what the optimizer was not asked to think about. Decline curve analysis flags the interior wells eventually, but only after the cumulative interference has produced a measurable departure from forecast, which can be many months downstream. Choke optimization software reports the change it made to Well 1; it does not report what happened to Well 2 in the week after.

What is missing is a metric that watches the neighborhood, not the individual well. When the AI raises production on one well, what happens to the wells next to it inside the same pad? That is the question MWPIR is built to answer.

Formula

Multi-Well Pad Interference Rate, or MWPIR, is the percentage of pad-level AI production-optimization adjustments where a beneficial change on one well is followed by a measurable decline on at least one offset well on the same pad inside a defined observation window. The metric exists to answer one decision: is the optimizer''s local rationality producing systematic neighborhood damage that aggregate field metrics are hiding?

The structural form is straightforward. For each AI-initiated production adjustment (choke change, ESP frequency change, gas-lift rate change, or any other rate-affecting action) the operator monitors offset wells on the same pad for seven days following the change. If any offset well shows a rate decline of more than five percent or a flowing pressure change of more than ten percent inside that window, and the change cannot be attributed to the offset well''s own operations or to a documented external cause, the original adjustment is flagged as an interference event. MWPIR is the count of interference events divided by total AI adjustments across the measurement period, expressed as a percentage.

The five-percent rate threshold and the seven-day window are starting parameters. The threshold has to be large enough to clear normal production noise and small enough to catch real interference. The window has to be long enough to capture the pressure-transient response between offset wells and short enough that other production drivers do not contaminate the signal. Operators should plan to recalibrate both parameters after the first six months based on the natural variability of their pad behavior.

Attribution is the part of the metric that determines whether it is honest. An interference event requires that the offset well''s decline be unexplained by its own operations. If the offset well had its own optimizer adjustment, a workover, an artificial lift change, or a documented operational issue inside the window, the event is excluded or marked ambiguous. If two adjacent wells are adjusted simultaneously, the resulting offset-well behavior cannot be attributed to either change cleanly and should be flagged for engineering review rather than counted. The metric''s integrity depends on disciplined attribution; a sloppy attribution practice will produce numbers that swing with operational coincidence rather than physical interference.

Proposed formula, pending validation pilots.

At a Glance

Category
Subsurface Decision Quality
← Field-Level Recovery Interference Index (FLRII)Proxy Model Fidelity Decay Rate (PMFDR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment