Energy companies are being asked to generate more cash, with less capital, under more long-term demand uncertainty than at any point this century. That combination has been the defining condition of the sector for several years now, and it has already reshaped how operators drill wells, allocate capital, and talk to investors.
What it has not yet reshaped, at least not to the same degree, is the surrounding commercial and administrative work: the proposals, the tenders, the contracts, and all the internal approvals. This work ensures strong operational execution translates to cash, and it has remained stubbornly manual while many other elements of the oilfield got automated.
This piece argues that the commercial process is where the next round of energy productivity gains has to come from, and that AI is critical to realizing this promise. But it also argues that AI investment will be held to exactly the same standard as every other capital allocation decision in this industry right now. In other words, the returns threshold applies. Enthusiasm alone is not sufficient.
Start with what the industry has already accomplished, because it explains where we find the remaining opportunity.
Cost cutting creates a higher floor under an operator's financial performance. It does not raise the financial performance ceiling. In asset-heavy businesses like energy, the ceiling is raised through gains in labor productivity: how much output you can extract from the same workforce. And on that measure, the US oilfield has been remarkable. US crude oil production per extraction worker has more than doubled over the past decade, even as the total workforce has remained relatively flat.
That gain was real, and it was hard won. But look closely at where it came from:
Longer laterals
Pad drilling
Faster rig moves
Automated drilling systems
Remote monitoring
Nearly all of it was applied to the physical work of finding and producing hydrocarbons. The field got dramatically more productive.
The office did not, or at least not to the same degree. The people writing proposals, assembling tender responses, routing contracts through legal review, and chasing internal approvals are working in substantially the same way they were a decade ago. They have faster laptops, better connectivity, and a document management system that may or may not be used consistently. But the underlying process is still a human being staring at a blank page, then a series of other human beings marking up what that person produced.
The corporate cost data across a cohort of 15 publicly traded US independent E&Ps shows real corporate costs per produced barrel of oil equivalent falling from around $2.50 in early 2019 to around $1.20 today. The direction is right. But the pace of that decline has slowed considerably in recent years, which is what you would expect from a category where the easy reductions have been taken and the remaining costs are largely people doing knowledge work.
That is the productivity ceiling. And it is the kind of work that has been hardest to automate, right up until recently.
In a business where the commercial cycle runs from concept to contract to cash, the time it takes to produce a credible first draft is a direct input into working capital, bid capacity, and win rate.
Consider the mechanics. A proposal that takes three weeks to reach a reviewable first draft consumes senior technical people during the window when their input is most valuable and least available. It compresses the time left for review, which is where we catch and correct quality and compliance problems. It limits how many opportunities an organization can credibly pursue in a given quarter, because capacity is bounded by the slowest step in the process. And it pushes the eventual contract, and the eventual cash, further into the future.
Now compress that first draft from three weeks to three days. The senior people re-enter the process at review rather than at creation, which is a better use of expertise. The review window expands, which improves quality. Bid capacity increases without adding headcount, which speaks directly to increased labor productivity. And the cash arrives sooner.
None of those outcomes require believing anything speculative about artificial intelligence. They follow from arithmetic. Cycle time is a cost. Compressing it produces measurable financial results in categories critically important to the CFO.
This is why the commercial process is a more interesting target for AI in energy than many of the applications that get more attention. Subsurface interpretation and predictive maintenance are genuinely valuable, but they are also areas where the industry has been investing for years and where incremental gains are getting harder. The blank page, by contrast, has barely been touched.
Halliburton's CEO, explaining why the company was idling frac fleets on its July 2025 earnings call, described stacking equipment that did not meet "our returns threshold." A decade ago that would have been unusual language on an oilfield service earnings call. Today it captures how thoroughly capital discipline has permeated operational decision-making throughout the oilfield. That same discipline is going to be applied to AI, and the industry will perform better as a result.
What that means practically is that AI investments in energy will be evaluated the way every other investment is evaluated right now: against a specific, quantifiable return, on a defined timeline, with a clear owner.
It also means the useful question is where AI produces returns that endure through a robust procurement process and a CFO's scrutiny. On current evidence, the strongest candidates share a profile: the task is high-volume, the current process is manual, the output is reviewable by a human before it goes outside the organization, and the time saved is measurable in hours that already translate to dollars in an existing cost structure.
Commercial document production fits that profile almost perfectly. It is high-volume. It is manual. Nothing goes out the door without human review. And the hours are already being counted, because they belong to people whose time is tracked and allocated.
SLB's experience offers a useful data point on how the market is responding to digital more broadly. The company reports digital as its own segment, and its CEO recently described it as "our fastest-growing business in recent years." Digital revenue grew +9% year-over-year in a period when the rest of the company's revenue declined 2%. Investors noticed that reporting choice, and they rewarded the transparency. The same logic will apply to AI. The companies that can report on it in financial terms will get credit for it.
Xait has spent two decades on the specific document described above: the large, multi-author, deadline-bound submission. Deliverables such as tender responses, bids, and Plans for Development and Operation. Its platform, XaitPorter, replaces the emailed Word file with a single live document, in which contributors write in assigned sections under role-based permissions and reviewers approve in place. The AI layer does two narrow things. It reads an incoming RFP, extracts the requirements, and flags compliance risks buried inside it. And as a contributor writes, it surfaces the organization's own approved and previously winning language rather than generating text from scratch. It runs on the customer's content, behind the customer's chosen model, with nothing feeding a public system. We know firsthand the importance our customers place on protecting the sensitive technical and pricing details that feed these documents.
Measured against the three variables a CFO already tracks, the evidence looks like this.
Senior hours moved from drafting to reviewing. A tender manager at DeepOcean described the shift as allowing her to concentrate on her assigned section rather than a 100-page document. She no longer had to spend the final two days before a submission working past midnight. This is a person in a senior role, deploying her time in a direction that aligns with the highest value she brings.
Pursuits run in parallel at the same headcount. Honeywell UOP, coordinating proposals across three lines of business and multiple global sites, reports close to a year of cumulative time saved in its first year on the platform and, speaking directly to productivity uplift, an increase in the number of proposals it can issue annually with the same team. This is a win across both efficiency and output volume, an improvement that yields immediate, tangible financial results.
Calendar days from receipt to submission. Aker BP's Plans for Development and Operation draw on geology, reservoir, drilling, installation, and HSE teams, then have to survive regulatory review. On one recent PDO, 34 writers and 41 reviewers worked inside the same document at the same time, with no manual reassembly at the end. The company has since submitted a record number of PDOs in a single window. Xait's aggregate figure across its customer base is documents produced up to 70% faster, delivered already formatted and compliant.
Two caveats belong here. First, most of what is measured above is a coordination gain from the shared-document model, which predates the AI layer. The AI contribution to cycle time is newer, and its effect is only beginning to separate out in customer data. Second, the link to win rate is inferred rather than demonstrated. There is no reason faster, more compliant submissions should not win more often. Xait tracks the inputs and leaves the win-rate conclusion to each customer's own numbers.
We want to leave you with three ideas.
First, the energy industry's productivity gains have been concentrated in the field, and the field is running out of easy gains. The commercial and administrative process is the largest remaining pool of manual knowledge work in the sector, and it has been substantially untouched by the existing automation wave.
Second, time to first draft is a financial variable rather than a convenience. It determines how many opportunities an organization can pursue, how much senior expertise gets consumed by production rather than judgment, and how quickly a commercial cycle converts into cash. Compressing it produces results in categories that matter most to the CFO.
Third, AI in energy will be held to the same returns threshold as everything else, and that is a good thing. It will filter out the applications that demonstrate well and measure poorly. What survives will be the uses where the task is high-volume, the process is manual, human review is preserved, and the time saved is already counted somewhere in the cost structure.
The industry spent the last decade proving it could produce more oil with the same number of people. The next decade will be about proving it can run the back office with the same discipline. That is a smaller story than the one usually told about AI. It is also a considerably more believable one, and it is where the returns are most likely to show up first.
"We hold our own AI to the same standard this piece argues for," says Eirik Gudmundsen, CEO of Xait. "It earns its place in the commercial process only where it demonstrably shortens the path from RFP to submission, and only with a qualified reviewer standing behind every word that goes out the door."