Field Insights & Resources

Management by exception is the operating model behind every credible field-efficiency number published this decade, and most implementations of it still stall. What the model actually requires (a data foundation you already own, exceptions that carry dollars, a ranked plan the field will run, closed-loop feedback), why the stall happens (alarm floods, exceptions that all look equal, a field that stopped trusting the plan), and the four-week productized path.

The most common reason smaller operators never start management by exception is a prerequisite list somebody sold them. The honest starting guide for the operator running on SCADA and spreadsheets, or gauge sheets and a phone tree: what you actually need on day one, what the first four weeks look like, and the four things to demand from any vendor before signing anything: read-first integration, a walk-away clause, field adoption evidence, and exceptions defined in dollars.

The loudest argument in oil and gas AI right now is about how much autonomy to give the system. It is the wrong axis. On a lease, no one ever trusted a tool because it had the right permissions. They trusted it because it was right this morning, and right the last three mornings too. Trust is a track record, not a permission setting.

Allocation variance of 5 to 15 percent has been normal for as long as the industry has existed. The trouble is not the gap, it is that the same allocated number drives royalties, severance tax, JIB, and SEC reserves, and almost no one can trace it back to the meter. This is why hiring more accountants does not fix it, and what lineage looks like instead.

Vehicle incidents kill more upstream oil and gas workers than any single on-site hazard. The industry has spent two decades making the drive safer, and those controls work. But they all assume the trip happens. The first-order question hiding under the fatality data is whether the route needed to happen at all.

"We could just build this ourselves" is half right, and the half it gets right is the half that no longer matters. A field-grade look at where the real cost of AI lives, what the evidence says about in-house versus vendor outcomes, and how to decide which pieces to own.

On October 26, 2019, two people died at an Aghorn Operating waterflood station outside Odessa. The U.S. Chemical Safety Board final report names seven independent failures that aligned in a single afternoon. None of them felt critical alone. All of them were lethal together. That is the Swiss cheese model in its purest form, and it is the model every HSE program in upstream oil and gas is supposed to be defending against.

Every operational platform needs a clean, reconciled asset hierarchy to run against. At most operators that hierarchy lives in fourteen disagreeing systems, and the reconciliation falls on the most expensive engineers, costing roughly $150K before the software produces a single decision. A single source of truth is a process, not a workshop artifact.

RBL lenders now pull production-discipline data alongside the reserve report. PE megafunds underwrite operating discipline, not just acreage. Tier-one rock is locked up. The independent CFO who can hand the lender clean per-well cash-flow data on the same rock as the operator next door defends the credit and the seat at the next M&A table. The independent who cannot is renegotiating from behind on a methodology they were not measured against last cycle. The fall redetermination is six months out and the path to book the operating dataset is one week of integration, thirty days to the first ranked daily plan, ninety days to closed-loop deployment.

AI is no longer a slide in the strategic plan. It is a number the CFO is being asked to defend on the quarterly call. Devon, ConocoPhillips, APA, Chevron, and ExxonMobil disclosed AI outcomes at a granularity the investor can underwrite. Gartner, McKinsey, and BCG quantified the gap most operators are still stuck behind. The supermajor proof points all ran against SCADA history the operator already owned, which is the lesson the independent should extract.

The discovery-then-pilot sequence the consulting industry sells is producing decks, not deployments. McKinsey reports 70% of operators are still stuck in pilot phase. Gartner reports 30% of GenAI projects are abandoned after POC. The bar moved while the workshops ran. The 24-hour AI operations diagnostic ingests the operator's SCADA, lease accounting, historian, GIS, and EAM in read-only mode and returns a ranked work list against the operator's own wells by 5:30 AM the next morning. Same vertical-AI substrate that runs the 5,000+ well deployed reference. No license fee, no kill fee, no decks.

Three years. Fifteen percent. That is the new operating bar in upstream oil and gas. Devon, ExxonMobil, ConocoPhillips, Chevron, and APA cleared it on public, peer-reviewed deployments. The Lower-48 winners closed the gap in twenty-four months by buying a result instead of building infrastructure. The independents that adopt this year close the gap to the supermajors. The independents that wait become the acquisition.

Every independent has heard the same three sentences from a consultant or a hyperscaler partner. We aren't ready for AI. Our data isn't ready. We are still working on governance. That sequence is a 2017 idea. The supermajors stopped paying for it four years ago, and the operating proof points they now cite (ExxonMobil/SLB 2.2% on 1,300+ wells, ConocoPhillips PLOT 30% on 4,500+ wells) ran against the SCADA history the operators already had. The independent that builds a lake in 2026 is paying for a workaround to a problem that no longer exists.

The lift in tank gauging is not in the gauging. It is in the cadence. Gauging cadence built around a 3-day route is a holdover from an era when no one could predict tank level overnight. The forecast has been on most operator stacks for a decade. Small-to-mid operators have not adopted because they think the build needs new sensors. It does not.

One-third of upstream production downtime is structurally preventable on the same wells, with the same crew, on the equipment already in the field. The signal is already in the historian. Three supermajor case studies confirm the lift. Most independents have not adopted because they think the build is multi-year. It is not.

The average hydraulic study at a North American gas distribution utility costs $15K-plus in loaded engineering time and takes more than two weeks to build. A five-person planning team running the standard loop burns $750K to $1M a year on model maintenance. The math has not changed in 30 years. Closing the loop has.

Most operators expect a multi-quarter build. The shape of an actual pump-by-priority pilot is one week to integrate read-only onto the existing stack, two weeks to put a ranked plan in every truck cab, and one week to measure against a metric the controller signed on Day Zero.

A&M, ExxonMobil, ConocoPhillips, and Chevron all documented the same operating model. Three peer-reviewed citations, three operating levers, one durable conclusion: the approach is validated, the gap for small-to-mid operators is adoption speed.

Nine workers died during manual tank gauging or sampling between 2010 and 2014, every one of them working alone. NIOSH measured benzene at open thief hatches over 2,000x the recommended exposure limit. The API 12F tank that sits behind nearly every onshore wellsite in North America is not a container, it is a vapor generator with a thin steel shell around it.

Most oil and gas AI projects fail for the same reason: the AI has nowhere to live. You need an operational foundation BEFORE agents can do anything useful. Here are the four layers that matter.

Your refinery runs APC. Your field runs on Excel. Why? WorkSync applies the same constraint-and-objective math that keeps a refinery optimal to the daily field work plan. Cash flow as objective, safety as hard constraint.

Lone-worker hardware devices are world-class at what they do, but hardware alone sees a worker at GPS coordinates, not what they're walking into. The platform layer (asset hazards + crew qualifications + weather + JSAs) is what turns that location into a decision before the call goes out.

The average time between a SCADA alert and a dispatch decision is 23 minutes. In that window, superintendents flip between 8-15 software systems, make capital allocation decisions worth thousands of dollars, and hope they got the priority right.

The superintendent manages 30 people, a dozen contractors, and hundreds of wells. But their most demanding boss is the collection of fragile systems that must be fed, tuned, and maintained every day.

At $60 oil, margin protection requires operational discipline. Top private operators are cutting LOE 10-15% and increasing production 3-5% without drilling a single new well.

The story from the field team's perspective: how daily work changed when operators went from spreadsheets to optimized, prioritized routes.

Walk through what a day looks like for a lease operator before and after Work Engine, from legal pads to optimized routes by 6 AM.

Industry analysis suggests that upstream operators lose 15% of operational cash flow annually to reactive decision-making and fragmented data systems.

The energy industry has long optimized for barrels of oil equivalent. But BOE doesn't account for lifting costs, transportation economics, or well-level profitability.

Operators have spent millions on SCADA, CMMS, ERP, and production accounting systems. Despite this investment, most field teams still start their day building spreadsheets.