Operators keep asking for AI without writing down the two numbers that decide whether a build is worth it. The test we use before anything gets built, and the 20 hours a week it gave back at Spearpoint Resources.
How to tell which AI projects actually pay — 5 min watch
An operator and a royalty company have the same temptation. The tools will let you build the app. The app will not tell you whether the month you spent on it came back.
We work with more than a dozen operators, most of them in Houston, Oklahoma, the Midwest, and Midland. The same question shows up in almost every one. Someone wants AI in the business. Nobody has written down the hours the work takes today, or the hours it should take once the thing actually runs.
Knowing you can build it is not the same as knowing what to build.
Two clocks, then one question about the wells. If you cannot fill those in, you do not have a project. You have a demo.
The people who do it, this week. Not a feeling that the process is messy. A count.
The week it has been running. Review stays. The statement the tool could not read stays. Both go in the total.
More wells that should have been drilled. Less time with a well down. A picture of the company you can ask a question of.
The first number gets skipped because it feels obvious. It is not. “Revenue is a mess” is not a number. Thirty hours a week, on revenue reconciliation, is a number.
The second number is the one people invent. They picture the afternoon the tool is finished and call that the new week. Count the week it has been running. Someone still checks the output. Someone still fixes the statement the tool could not read. That time does not leave the total because the demo looked clean.
The return is the gap between the two. A two-hour gap and a quarter of building is already an answer. You do not need a model to see it.
We just finished this with Spearpoint Resources. Revenue reconciliation was thirty hours a week. It is ten. Twenty hours a week went back to the company.
A back office pays you back in time. The company pays you back somewhere else. A build earns its place in the business when it does one of these. If it does none of them, and the hours do not come back, do not build it.
More wells that should have been drilled, or fewer that should not have been. The build has to change a decision, not just a spreadsheet.
A well sitting down in the field is the expensive version of a late report. If the build does not shorten that, it is not a field project.
One source of truth the company can ask. Not a folder per person, and not a chatbot on top of five versions of the same number.
A joint interest billing does not drill a well. A purchaser statement does not either. Both still have to get done. They do not have to get done by the person whose job is the lease. Necessary work and the right use of an operator’s week are different things. That is the distinction that gets lost once building feels free.
Alex Hormozi, How to Use AI in Your Business in 2026
Hormozi makes the point we keep making in a different industry, and he makes it well. You do not need to become an AI company. You integrate it the way you integrated the internet. Nobody advertises that they use email. They just stopped mailing things.
Three of his points matter more in oil and gas than in the businesses he is usually talking to.
A role is not automatable. A task is. “Revenue accountant” is a title. Open the purchaser statement, find the well, match it to the deck, check how severance and transportation were netted, post it by well, tie it to the deposit — that is six tasks. Four of them are reading. Two of them are judgment. You cannot see that split until you write the list.
His framing, and it holds up at a coding desk. When someone who has closed two hundred months looks at an invoice and knows the account, the cost center, and the well, that is not instinct. That is a few thousand prior invoices. Pattern recognition transfers. The parts of the close that run on taste are usually the parts that run on volume.
He warns against handing the build to a technologist who does not know the work, because you get generic automation. In this industry that is the whole ballgame. An automation shop will deliver a statement-to-spreadsheet tool in a week. It will not know that a check stub which does not foot is usually a netting difference and not an error, that one line on the stub is not a well at all, or which of your decks is in eighths. Someone has to carry it from the strategy down to the sheet. Usually that is the owner, and it is not delegable.
His point about training is the one most operators get wrong. You train it the way you train a new hire: name the outcome, name the observable behavior, give examples. “Code it correctly” is not an instruction. “This vendor’s chemical tickets go to that account, on the well named in the ticket header, and if there is no header you stop and ask” is an instruction. The second one can be checked. The first one cannot, which means you will never know whether it worked.
He also says the window is short — that in a year and a half this is table stakes rather than an edge. In oil and gas it is wider than that, and that is not a reason to wait. It is the reason the hour count matters. An edge you cannot measure is not an edge. It is a demo with a longer runway.
Only with both clocks already written down, and with someone who has seen the output before. The hard part is not the model. It is knowing which build returns the hours, and which one only proves that you could.
They sit in the company and build the automations with the people who run the process. What that costs moves with the size of the company. The same seat at the labs that train the models is scarce enough that waiting on one is its own plan. What the good ones leave behind is not a pile of one-off scripts. It is one place the whole staff pulls the same answer from. That is the practical version of an AI-native oil and gas company.
Two shapes, on our side. The software, and your team runs it. Or we come in as the bookkeeping: we automate the month and we do the lifting. JIBs out to partners. Revenue in, by well, from your own production or from another operator or a purchaser. You drill.
The same test applies on our side of the table. We can build something new any week. The hours that bring the next customer are the hours spent writing and calling. A build feels like progress. It is the wrong work when the needle is somewhere else.
For an operator, the somewhere else is the courthouse and the well. Pull the leases. Drill. The back office still has to close. Jibbing out partners, and finding what a well paid you on someone else’s statement, is real work. It is not the work that grows the company. Hand that off.
We will put a number on the time it takes now and the time it should take after. If the gap is real, you can run the software, or we can take the month.
Get quick answers to common queries in our FAQs.
You only pay for what you use — no seat fees and no modules you don't need. Pricing is a monthly platform fee plus usage on documents processed, ACH payments, mailed checks, and the workflows you turn on. We size it to your actual monthly close so it scales with the work, not your headcount.
A smaller operator running 25 documents, 10 ACH payments, and 2 mailed checks a month would be priced on that exact volume. A larger operator at 100 documents, 50 ACH payments, and 10 checks pays predictably more. You always know what you're spending because it tracks the actual close.
Joltly connects directly to QuickBooks and Quorum On-Demand Accounting, and supports file-based export workflows for systems like PakEnergy and Integra. It manages accounts, items, vendors, partner mappings, JIB clearing, revenue liabilities, and netting accounts inside your existing setup.
Both sides of settlement. On expenses: invoice review, GL coding, approvals, JIB creation, ACH and check payments. On revenue: statement OCR, partner distributions, remittance emails, and netting between JIB receivables and revenue payouts — so your team replaces spreadsheet work and email follow-up during close.
Yes. Your wells, partners, revenue interests, GL mappings, approval flow, export formats, and partner-facing statements are configured per operator. Most customers go live on their existing chart of accounts and ERP setup — no rebuild required.
Faster than most teams expect. Your AP workflow can be live in less than a week — and you don't need to change your current accounting system. Revenue typically takes about two weeks. The only thing we need from you is a sample revenue statement so we can train the AI on your format.
We set up a Microsoft Teams channel or email channel for your team — whichever you prefer. You'll also get the founders' phone numbers for text messaging when you need a fast answer.
