Reference · Last updated August 26, 2026
Where upstream AI value actually sits
Someone with no product to sell in this category ranked the upstream AI use cases by value. Six of the top ten are field operations. Schedule optimization, the thing most operators file under administrative overhead, ranks inside that top ten. This page is the map: the full ranking, our arithmetic shown in the open, and what a concentrated bet looks like when you are the one who has to run the field on Tuesday.
Source: Can upstream oil and gas produce value from AI's $230 billion pay zone? McKinsey & Company, August 25, 2026, by Bill Ambrose, Giorgio Bresciani and Spandan, with Priyank Singh. All dollar figures below are McKinsey's unless labeled otherwise. The categories, the arithmetic and the argument are ours.
Three findings worth an operator's morning
The value is concentrated, it is concentrated in the field, and almost nobody is measuring whether they captured any of it.
Top-ten use cases that are field operations
Six of the ten highest-value upstream AI use cases on the McKinsey ranking happen on a lease, on a shift, with a truck and a crew and a pump. (McKinsey, August 2026; the field-operations classification is ours and is argued below.)
Carry close to half the identified value
McKinsey describes upstream AI as a concentration play rather than a spread bet, with the top ten use cases driving nearly half of the value they identify, and describes spreading investment across the long tail as value dilution. (McKinsey, August 2026)
Of deployments measure what they produced
Fewer than 20 percent of AI deployments track the value the AI generated and share that KPI with the organization. Four in five programs running in this industry cannot answer the question. (McKinsey, August 2026)
The value map
Ten use cases, ranked. Six of them happen on a lease.
Values are McKinsey's, at full potential, in billions of dollars (McKinsey, August 2026). Gold marks the six we classify as field operations, meaning work that is decided and executed in the daily operation of wells you already own. Our presentation, our categories, their figures.
- ExplorationSubsurface (balance sheet, not P&L)$35B
- ESP optimizationField operations$20B
- Rate of penetrationDrilling$16B
- Gas lift optimizationField operations$14B
- Waterflood optimizationReservoir$11B
- Predictive maintenanceField operations$11B
- Rod pump optimizationField operations$8B
- Schedule optimizationField operations$7B
- Supply chainField operations$5B
- GeosteeringDrilling$5B
The six gold bars, summed. Our arithmetic on McKinsey's full-potential figures, checkable in fifteen seconds.
All ten bars, summed. McKinsey annotates the exhibit to say the top ten drive nearly half of the identified value, and puts the full-potential total at $230B.
McKinsey's full-potential figure for AI across upstream, against roughly $65B they judge reachable in the near term with today's technology.
Read this before you quote the number
Two different $65 billion numbers are now in circulation. They are not the same number.
McKinsey's $65 billion is a near-term total for all of upstream: the annual recurring value they judge reachable with today's technology, against approximately $230 billion at full potential.
Our $65 billion is the sum of six full-potential use-case bars on their exhibit: ESP optimization, gas lift, predictive maintenance, rod pump, schedule optimization and supply chain. It is arithmetic on their chart, not a total for anything.
The collision is a coincidence and we would rather flag it than benefit from it. The exhibit values cannot be near-term figures: six bars alone would consume the entire near-term total. Summing all ten gives about $132 billion, which is the order of magnitude McKinsey's own annotation implies (the top ten drive nearly half of the identified value, against a $230 billion full-potential total) and nowhere near a near-term reading.
So if you carry the six-bar figure into a meeting, carry the label with it: $65 billion of full-potential value, concentrated in daily field operations, summed by us from McKinsey's published bars.
What we excluded, and why
Every judgment call we made shrinks our own number.
The dollar figures are McKinsey's. Deciding which of them count as field operations is a judgment, and it is ours. Here is every use case we left out of the $65 billion, with the reason. Four bars worth $67 billion, more than the number we kept.
| Use case | Value | Our call | Reasoning |
|---|---|---|---|
| Exploration | $35B | Not field operations | Subsurface work, and McKinsey frames its value as balance sheet rather than P&L. It is the single largest bar on the chart and we excluded all of it. |
| Rate of penetration | $16B | Not field operations | Drilling performance. Real value, different organization, different budget cycle, pulled once per well. |
| Waterflood optimization | $11B | Not field operations | A flood pattern is a reservoir decision made by reservoir engineers on a timescale of quarters, not a daily field decision. This is the one an operator could reasonably argue with us about. |
| Geosteering | $5B | Not field operations | Drilling, in real time, by a different discipline. Excluded on the same grounds as rate of penetration. |
If you think waterflood optimization belongs in operations because your production engineers run it week to week, then the field-operations total goes up, not down. That is the direction every available disagreement runs, and it is the reason the $65 billion is worth defending.
The line nobody expects to find on that chart
Schedule optimization is a $7 billion line.
The objection we hear most often, and we hear it from people whose judgment is worth respecting, is that routing crews is administrative overhead. Acreage wins the budget fight. Get the land wrong and nothing downstream saves you. Subsurface sets the ceiling. Drilling and completions decides whether you reach it.
All of that is true, and it is half the picture. The missing half is frequency. Land gets pulled once per position. Subsurface, once per program. Drilling and completions, once per well. Production operations gets pulled every day, on every well you own, for the life of the asset. A decision made once has one chance to be right. A decision made three thousand times a year has three thousand chances, and every structural improvement to it compounds against a denominator no other discipline in the company is working on.
Routing has no reservoir glamour. It does not appear in a reserve report. In most organizations it is a whiteboard, a phone, and nobody's KPI. A third party with nothing to sell in this category put it in the top ten of where AI creates value in upstream (McKinsey, August 2026), within a billion dollars of rod pump optimization and ahead of geosteering. Operators fund rod pump programs without argument. Almost nobody funds the routing decision at all.
What that has measured out to on our side
Fewer field miles driven at the same production coverage, measured across 5,000+ wells in live deployments.
Free cash flow uplift on the same crew, measured across 5,000+ wells in live deployments.
TRIR over the same deployment. The most dangerous task on most leases is the drive to it, and fewer miles is fewer exposures.
WorkSync deployment figures, measured in the field rather than modeled. The mechanics are on the route optimizer module, which ranks stops by value density rather than distance.
Why this lever compounds
A well-run field pays you twice on the same rock.
This is the part that gets underweighted when field operations are treated as a cost center. The monthly cash flow effect is the part everyone sees. The enterprise-value effect arrives twice, and neither instance requires a new acreage position or a new completion design.
More barrels, same rock
Deviations get found and worked in the order of what they cost you, so less deferred production accumulates while a crew is somewhere lower-value. It shows up in the monthly cash flow, on assets you already own.
Your PDP is worth more
Proved developed producing volumes rise and the cost per barrel to operate them falls at the same time. The existing asset base reprices on both sides of the equation, not one.
Your inventory is worth more
Undeveloped locations land into a field that operates better, so every future well inherits the higher-performing operating environment from the day it is turned to sales. You get paid on the wells you have and on the wells you have not drilled yet.
None of this requires new acreage or a new completion design. It requires that the daily decision about where competent people go be made against value rather than against habit, geography, or whoever called the office first. The full argument for pricing the exception before you rank the work is on the pump by priority page, which shows how a deviation gets priced in dollars per day.
How to act on a concentrated map
Concentrate. Do not sprinkle.
McKinsey's guidance on capturing any of this is blunt, and it is the guidance most operator AI programs are currently violating. They describe AI in upstream as “a concentration play, not a ‘thousand flowers bloom’ opportunity”, and describe spreading investment across the long tail as a form of value dilution rather than diversification (McKinsey, August 2026).
That is a direct challenge to the default operator posture of the last three years: a portfolio of small pilots, one per department, each with its own sponsor, each proving something modest, none of them changing how the field runs on a Tuesday. A portfolio of pilots feels like risk management. On a map where the value is this concentrated, it is the risk.
The concentrated bet available to an operator right now is the daily field decision. It is where six of the top ten use cases live, it runs on data the operator already owns, and it is the only lever on the chart that gets pulled again tomorrow morning whether or not anybody optimized it.
Build or partner
On commoditized applications, fleet scale beats internal control.
McKinsey names build-versus-buy as a barrier to scaling, observes that these decisions are often driven by a desire for control rather than by outcome, and warns that excessive internal build tends to be expensive and slow.
Their substantive point is sharper than the general advice. Where an application is commoditized, they argue, a provider with “fleet-scale data across thousands of installations” will likely outperform any single operator's in-house effort, and they name predictive maintenance of rotating equipment and standard production surveillance as exactly that class of application (McKinsey, August 2026).
This is worth sitting with, because it is a third party saying the thing a vendor cannot credibly say about itself. The reason a partner wins on this class of problem is not that the operator's engineers are less capable. It is that a pump failure signature learned across thousands of installations is a better model than the same signature learned across one fleet, and no amount of internal talent closes a data-volume gap of that shape. The operator's durable advantage is the rock, the acreage, the completion design: the things that are genuinely proprietary. Rebuilding a commoditized production surveillance stack in house spends the scarcest resource you have, engineering attention, on the one category where scale beats you by construction.
Two of the six field-operations use cases on the map, predictive maintenance at $11B and the production surveillance that sits underneath the artificial-lift lines, are precisely the zone McKinsey flags as partner territory.
We wrote the long version of this before McKinsey published theirs.
Our open letter to operators makes the same case from the primary filings: eleven of the twelve largest US operators report no research and development expense line at all, and zero of them earn a dollar of software revenue (SEC FY2025 10-K filings). Read what building field software in house actually costs, in the filings and after the demo.
The target state, described by someone else
Full-field review, with engineers on the wells carrying the highest expected value of intervention.
McKinsey's description of what good looks like in production operations is a shift from periodic review to full-field review with exception-based optimization, where engineers focus and act on “the wells and facilities with the highest expected value of intervention” while an AI system manages the lower-producing tail (McKinsey, August 2026).
That is the category WorkSync builds in, written by someone with no product in it. Rank every well and every exception by what acting on it is worth, work the top of that list, stop spending your best people on the tail. The gap between that sentence and a running field is entirely mechanical, and the mechanics have a name.
State-based logistics, the six steps
Know the state before a truck rolls
Every site carries a current state assembled from the systems you already own, not from a phone call at 6:40 AM. The dispatcher stops guessing and starts reading.
Price every deviation in dollars per day
A stuck plunger and a low-consequence alarm stop looking alike on a screen. If a live exception does not carry an economic value a controller would defend, it is an alarm, and you already own an alarm system.
Rank everything on one ruler
One list, priced, for the field and the office. Two lists is how the argument about priority starts, and the argument costs more than either list.
Route the day around the highest-value work
Not the shortest route. The most profitable one. Value density decides the sequence, not geography and not the order the calls came in.
Send the right competency the first time
The second trip costs the miles and the deferred production both. Matching the skill to the exception at dispatch is where most of the recoverable waste actually lives.
Capture by voice, so tomorrow learns from today
What actually happened gets captured at the wellsite by voice, not typed into a form on Friday. The ranking is better tomorrow because the field told it the truth today.
Twenty minutes of screen time a day. The rest is wells. That loop is why field operations compound: every one of those six steps runs again tomorrow, and the ranking is sharper tomorrow because the field told it what actually happened today. To see the output before you see the software, read a real ranked 6 AM plan, priced stop by stop.
The finding that is not on the chart
Fewer than one in five deployments measures what it produced.
The most useful number in the McKinsey piece is not on the ranking. It is their finding that fewer than 20 percent of AI deployments track the value the AI generated and share that KPI with the organization (McKinsey, August 2026).
Read that as an operator rather than as a statistic. More than four in five AI programs running in this industry cannot tell you what they produced. Not because the value was necessarily zero, but because nobody instrumented the question. That is why the boardroom conversation about AI in oil and gas keeps circling: the proponents cannot prove the win, the skeptics cannot prove the loss, and the program renews or dies on narrative.
It is also the strongest external case anyone has made for structuring the commercial arrangement around a measured number. If fewer than one in five deployments measures its own output, then a vendor's willingness to be paid against a metric the operator picked is not a pricing gimmick. It is the mechanism that forces the measurement to exist at all.
The Impact Guarantee, in one paragraph
Pick one operating metric in Week 0: deferred production found per week, days from anomaly to first field response, truck rolls per lease per month, free cash flow on the same crew. Write down the baseline and agree what counts as moved. The four-week pilot runs on one field against that metric, with the walk-away clause in writing, and license fees start when the metric moves. You end up in the minority of deployments that can answer the board's question, which is worth something on its own.
Where upstream AI value sits, common questions
Where does AI create the most value in upstream oil and gas?
On McKinsey’s August 2026 ranking of upstream AI use cases by value, the ten highest-value use cases are exploration ($35B), ESP optimization ($20B), rate of penetration ($16B), gas lift optimization ($14B), waterflood optimization ($11B), predictive maintenance ($11B), rod pump optimization ($8B), schedule optimization ($7B), supply chain ($5B) and geosteering ($5B), all at full potential. Six of those ten (ESP, gas lift, predictive maintenance, rod pump, schedule and supply chain) are daily field operations. Summing those six gives $65B of full-potential value, which is our arithmetic on McKinsey’s published figures rather than a McKinsey total.
Is the $65 billion figure McKinsey’s number or WorkSync’s?
Both numbers exist and they are not the same number, which is exactly why we spell it out. McKinsey states that AI can unlock approximately $65 billion in annual recurring value across all of upstream in the near term with today’s technology, against roughly $230 billion at full potential. Separately, summing the six field-operations bars on their use-case exhibit also gives $65 billion, but those are full-potential figures for six specific use cases, not a near-term total for the industry. The collision is a coincidence. Sum all ten ranked bars and you get about $132 billion, which is consistent with the top ten carrying close to half of the identified full-potential value and inconsistent with any near-term reading.
Why is schedule optimization ranked so highly?
Because it is a decision that gets made again every single day. Schedule optimization sits at $7B on the McKinsey ranking, inside the top ten, within a billion dollars of rod pump optimization ($8B) and ahead of geosteering ($5B), despite being the item most operators file under administrative overhead rather than value creation. Land is pulled once per position, subsurface once per program, drilling and completions once per well. The daily routing decision is pulled every crew, every day, every well, for the life of the asset, so a fixed unit of improvement compounds against a denominator no other discipline is working on.
Should an operator build production surveillance in house or partner?
McKinsey’s position is that where an application is commoditized, including predictive maintenance of rotating equipment and standard production surveillance, a provider with fleet-scale data across thousands of installations will likely outperform any single operator’s in-house effort. They also name build-versus-buy decisions driven by a desire for control as a barrier to scaling, and observe that excessive internal build tends to be expensive and slow. The operator’s durable advantage is the rock, the acreage and the completion design. Two of the six field-operations use cases on the ranking sit squarely in the zone they flag as partner territory.
What is exception-based optimization?
It is the target state McKinsey describes for production operations: moving surveillance from periodic review to full-field review, with engineers focusing on the wells and facilities carrying the highest expected value of intervention while an AI system manages the lower-producing tail. In plain terms, rank every well and every exception by what acting on it is worth, work the top of that list, and stop spending your best people on the tail. That is the category WorkSync builds in, described by a third party with no product to sell in it.
How does an operator prove an AI deployment actually produced value?
By instrumenting the question before the deployment starts, which fewer than 20 percent of deployments currently do according to McKinsey. Pick one operating metric in Week 0 (deferred production found per week, days from anomaly to first field response, truck rolls per lease per month, free cash flow on the same crew), write down the baseline, and agree what counts as moved. WorkSync structures the commercial arrangement around that number under the Impact Guarantee: the four-week pilot runs on one field against the metric you picked, and license fees start when the metric moves. If fewer than one in five deployments measures its own output, a vendor willing to be paid against your number is the mechanism that forces the measurement to exist.
Are your best people working on your most valuable work today?
Your land team won the acreage. Your ops team decides what it is worth.
Six of the top ten places AI creates value in upstream sit inside the decision your operation makes every morning about where competent people go. See that decision made against your own SCADA and production data, ranked in dollars, well by well.
Go deeper: the build vs buy open letter · how an exception gets priced and ranked · routing by value density, not distance · the field research library