6,800 daily nuisance alerts → 3 actionable alerts in 4 days.
A Tier-1 midstream NGL operator was running a legacy volume-balance + SCADA threshold system across a multi-segment line. Operators were dismissing thousands of alerts per day — including, by their own admission, the alerts they should have been investigating.
Day 0 — legacy SCADA thresholds
Day 4 — EI cross-stream ML
99.96% reduction in operator alert load. Control-room headcount unchanged — but every alert now demands action.
Catching what the legacy CPM system missed.
A crude oil operator running a legacy CPM detected a small leak days after onset because the volume-balance algorithm needed sustained loss-rate to cross threshold. After EI deployment on the same segment, an injected synthetic test signature of half that magnitude was detected in 38 seconds.
Legacy CPM detection time
Volume-balance threshold required sustained imbalance over a multi-hour window before crossing the alarm threshold. Small leaks below the operator's tuned sensitivity floor never tripped.
EI detection time
Cross-stream ML responded to the NPW signature within 38 seconds of injection. Magnitude (Δ ≈ -0.95 PSI) well below the legacy threshold.
GPS localization
NPW arrival timestamps at upstream and downstream stations resolved the synthetic-leak source to within the spacing of the operator's existing valve markers — first try.
Sub-minute detection. Sub-PSI sensitivity. Pipeline-aware.
Aggregate detection performance from the post-upgrade EI Combined v3 model fleet (anonymized cross-segment numbers).
- Median model prediction error: 0.219 PSI on the strongest segments. Sub-PSI sensitivity to deviation from learned behavior.
- Max prediction error on a worst-case segment: 7.59 PSI — versus 64 PSI on the legacy model the v3 replaced.
- Time from NPW onset to alert: typically <60 seconds. Within the wave-propagation window for most multi-mile segments.
- Operator alert load: dropped 70–90% on segments transitioned from threshold-only monitoring to v3 ML.
- Acknowledgement-to-resolution time: dashboards retain the full event audit trail so analysts can show their work.
- Sensor / device uptime: 24/7 health dashboard with staleness flags + carrier/IMEI/signal visibility per device.
From CAP-order risk to clean audit, in one binder.
A midstream operator was 90 days out from a scheduled PHMSA integrity-management audit with a Pipeline Leak Detection Program (PLDP) that had been flagged in the prior audit cycle. Their CPM met the letter of API 1130 on paper but couldn't produce the algorithm description, performance test record, or alarm-management artifacts the inspector requested in the pre-audit data call.
Pre-audit posture
Legacy CPM in place but with no documented Algorithm Description for the IMP. Alarm log retained but ad-hoc CSV exports, no event audit trail. Last performance test was over a year old.
EI artifact package
EI provided the EI Combined v3 Algorithm Description, a synthetic-leak performance test trace, the alarm-management procedure document, and dashboard-generated event audit logs for every segment.
Audit outcome
PHMSA inspector closed the prior cycle's PLDP finding with no additional corrective actions. Audit binder cited the EI artifact package as the primary remediation evidence.
A documented sub-minute LDP shifted the premium worksheet.
A midstream operator with ~340 miles of mixed-product line came up for renewal on its pipeline-pollution liability policy. The underwriter's risk worksheet had a line item for "Documented leak detection < 60 s with localization" that was previously checked "No." After EI deployment and a quarter of operational data, the carrier moved that line to "Yes" with documentation provided.
Prior renewal (no LDP credit)
Renewal after EI deployment
The premium math varies by carrier and by total insured pipeline value, but the standard credit category exists in most pipeline-pollution-liability policies. Documented sub-minute detection is what unlocks it. Specific credit percentages available on request, under NDA.
Eight unmonitored segments. Brought online inside 30 days.
A midstream operator inherited an acquired asset with eight pipeline segments running threshold-only monitoring — no ML models, high nuisance-alert rates, no leak localization, and a PHMSA audit window approaching.
EI deployed the EI Combined v3 architecture across all eight segments within 30 days, building per-segment models on the new operator's historical telemetry. The PHMSA audit binder included Algorithm Description (API 1130) and PLDP management documentation (API RP 1175) generated directly from the dashboard.
Inherited state
After 30 days
The "leak" that turned out to be organized crude theft.
Mid-pipeline product loss without a corresponding pressure transient is the story a volume-balance system tells the operator over a quarterly reconciliation cycle. By that point the product is gone, the trail is cold, and the regulator wants a report. Negative pressure wave detection rewrites the timeline.
Customer · Tier-1 crude operator · ~310 miles · remote South Texas footprint
The integrity team had observed an unreconciled volumetric loss on a remote crude segment crossing rural county roads — small, slow, but repeated month over month. The legacy volume-balance CPM flagged the monthly figure as within tuned tolerance. No alert ever fired.
After EI deployment on the segment, the new model surfaced a pattern the legacy system was structurally unable to see: small (0.4–0.7 PSI) pressure-drop signatures, clustered in 5–8 minute windows, occurring predominantly between 02:00 and 04:00 local on Tuesday and Sunday nights. Each signature alone was below the operator's old alarm floor. The pattern across multiple events was not.
Cross-station NPW arrival-timing localized every event to a 180-yard stretch of pipeline near a culvert crossing under a farm-to-market road — repeatedly, to the same physical coordinates within the model's ±25-foot localization envelope.
What the legacy system told the operator
What EI surfaced
The operator briefed Texas DPS and local sheriff's office with the EI location data, alert timestamps, and the cyclical pattern. A joint operation was set up on the next predicted extraction window. Officers on scene interdicted an active hot-tap operation — saddle tap welded onto the live line, vacuum truck staged at the culvert access road.
- Time to detection — pattern surfaced inside 21 days of EI being on the line, vs the prior 9-month pattern of unreconciled loss.
- Time from interdiction to safe shut-in — under 90 minutes; operator's emergency response team coordinated with state troopers in real time using EI's live pipeline-map view.
- Recovered evidence — saddle tap hardware, vacuum truck, tap-site fittings, partial product inventory inside the truck. Federal indictment followed for theft from interstate shipment (18 U.S.C. § 659) alongside state-level organized-criminal-activity charges (Tex. Penal Code § 71.02).
- Operator outcome — segment-level reconciled losses returned to pre-tap baseline. The audit binder on this segment now closes with an interdiction outcome instead of an unresolved variance.
- Insurance posture — operator's pipeline-pollution carrier requested the EI detection log as part of the loss-prevention credit qualification on the next renewal cycle.
- Industry context — PHMSA + AOPL track third-party damage as the leading single cause of significant hazardous-liquid incidents. Hot-tap theft is the deliberate-actor subset of that category, growing in remote Permian + Eagle Ford footprints.
The point isn't that EI catches theft. The point is that any detection system fast enough and precise enough to catch a small organized hot-tap is, by construction, fast enough and precise enough to catch the small real leak from corrosion-induced wall failure or third-party excavation strike. Same physics. Same model. Same alert pipeline. Theft is just the case study where the magnitude is small and the schedule is repeatable — the worst-case detection problem solved.