From an Ocean of Words to an Ocean of Wells
We may be living through one of history's great technology shifts. Most professionals have now experienced the remarkable fluency of ChatGPT, Claude and Gemini across language, reasoning and coding. But an equally important story is what the mathematics underlying these systems can do beyond language.
Consider wordsmithing. Change one word, add a qualification or shift the context, and the passage moves. LLMs display extraordinary sensitivity to the tugs and pulls among words, ideas and context that shape meaning.
It demonstrates something profound: relationships among many variables can be learned in extraordinarily high dimensions — including how the influence of one changes in the presence of another.
That requires data volume. LLMs had an ocean of words.
Twenty years into the shale revolution, many operators now have an ocean of wells.
Thousands of enormously expensive experiments combine geology, landing position, spacing, stimulation intensity, flowback, operating conditions and economics, with outcomes now known.
The data volume finally exists to apply high-dimensional pattern recognition to a very different problem.
You Paid for It. Put a Ruler to It.
After two decades of shale development, we don't need AI simply to tell us what matters. We know many of the important variables.
Now we can measure their individual and interacting influence on the outcomes that matter.
How hard does spacing pull? Stimulation intensity? Reservoir quality? How does one influence change in the presence of another?
Put a ruler to the tugs and pulls.
Once those relationships are learned, prediction is only the beginning.
Build an Empirical Simulator
Traditional simulation formalizes our collective understanding of geology, physics and engineering to calculate what should happen.
An Empirical Simulator learns the multidimensional response of the actual development system from thousands of completed experiments — what actually happened.
The two are complementary.
Theoretical simulation: Given what we know, what should happen?
Empirical simulation: Given what actually happened, what does the evidence say will happen?
Where they disagree may itself contain valuable signal.
Now Ask: “What If?”
Given the geology we cannot control, what happens when we change the things we can — spacing, stimulation intensity, landing strategy, flowback and other development decisions?
Once trained, the Empirical Simulator can search that learned development space. But optimization requires first defining what “best” actually means.
Given this geology, what should we do — and what are we trying to maximize?
Recovery per well? Recovery per dollar? Recovery per acre? NPV? Capital efficiency? Payout? Positive FCF lifespan?
Each objective can produce a different optimum. The well that maximizes recovery may not maximize capital efficiency; the design that maximizes individual-well economics may not maximize value across the asset.
The Empirical Simulator can search the learned development space for the combination of controllable parameters best suited to the geology encountered and the economic objective chosen.
Not one generalized recipe.
A custom optimization for the rock in front of us, the capital available, and the value we actually want to create.
The Advantage Is Already in Your Data
The algorithms will increasingly be available to everyone.
Your competitors cannot recreate the thousands of experiments you have already paid to run.
Your development history is not merely a record of past production. It is a proprietary body of empirical knowledge whose value has yet to be fully extracted.
The internet gave AI an ocean of words.
The shale revolution has given us an ocean of wells.
We now have the tools to measure what those wells have been telling us, build an Empirical Simulator from what actually happened, ask “what if?”, and optimize the capital we haven't spent yet.