Of all the levers available to an operator, perhaps none has a greater impact on unit economics than wells per section. Over the past few years, operators have stopped thinking of developments on the well-level and have shifted to “cube-level” or unit-level economics. However, deciding on the right unit design for cube development in oil and gas is tough — the existing data is difficult to interpret, with the confounding variables of completions design and rock quality obscuring the results of dense tests.
Machine learning provides a powerful tool to cut through multivariate relationships to understand the impact of cube development design on unit performance and economics. We’ll be doing a case study of ConocoPhillips’s Midland Basin acreage, recently acquired via the Concho deal.
In this post:
The oil and gas cube development analytics workflow
We’ll use machine learning to help us with our cube development optimization. Here’s the workflow we have taken with this public-data model (& this is very similar to how our customers go about addressing any of their business problems with our software):
- Build a unified dataset from production, completions, subsurface, surveys and well header sources.
- Calculate spacing using surveys where available and SHL/BHL if not.
- Train a machine learning model using your dataset and spacing features as inputs
- The model will learn how geology, completions, and spacing/stacking impact production
- Use the machine learning to generate forecasts for hypothetical cube development designs.
We’ll pick up in this post at BULLET 4: using the machine learning model.
But first: where can you cube?
The first question on the minds of Midland Basin operators (or investors) is which rock actually supports multi-zone development? Going from single-zone to multi-zone is a huge delta in well inventory — but clearly, this can’t be done everywhere.
We digitized a shapefile from their Investor-relations report and used that as the basis for laying out pads on their undeveloped acreage:

Next, we’ll use our model to generate forecasts for single-zone developments and multi-zone developments across COP’s acreage:

So, how do forecasts at those locations actually look? In some cases, there’s a huge degradation between well-level performance in a single-zone and well-level performance in a multi-zone development, sometimes over 30%:

But which rock actually can support multi-zone development (especially at lower prices)? Let’s take a look at (half cycle, unburdened) NPV10/acre for multizone developments at $45:

The clear standout is the Mabee area in the northwestern part of the basin, with ten of the top eleven NPVs at a $45 price deck. So, we’ve found an area that supports cube development even at low prices. Now, let’s take a much deeper dive.
Asset Optimization through Scenario Analysis & Oil and Gas Cube Development
Next, we will broader on our analysis to include an 18 WPS development and a 36 WPS development. Here are the unit development plans I built in Forecast Engine:

Next up I just had to lay out these developments across the Mabee acreage. This was very easy with Forecast Engine’s Well Inventory Lasso & Replace workflow:

In addition to these different spacing and stacking designs, we also looked at three completions designs and ran economics on three different price decks with two different completions costs:

This will give us a pretty granular view!
Spacing degradation
Let’s get into it. How do the well spacing and completions design impact the performance?

What are the key takeaways from the above chart?
- Depending on the zone and the completions design, we see that 2-year cum degradation can be in the 20-30% range going from a wide single-zone development to a tight multi-zone development.
- Completions uplift is lower at tighter designs than it is at wider designs.
- Wolfcamp A shows the highest responsiveness to intense completions in this area.
But what does this mean for NPVs? Of course, the answer depends on price deck and completions cost. At $45/bbl flat prices and +30% completions cost, we find NPV maximization at 18 WPS and 2250 #/ft completions (though this barely beats a Lower Spraberry single-zone development). At $65/bbl with base costs, the NPV maximization is found at 36 WPS, 2250 #/ft.

Layering in risk
It’s not all about P50 forecasts. Proper investment decisions require an understanding of risk and uncertainty. We output confidence intervals for all of our forecasts. Our customers will look at a range of metrics when considering uncertainty, including P90/P10 ratio, (P90-P10)/P50 production, and a variety of custom ones built for their internal processes. But for asset optimization, I want to look at the P10s, as it’s a good measure of “what could go wrong??”

The answer is.. a lot! With 1-year P10 cums of 80,000 bbl (for the Wolfcamp B).. that super-dense development might be looking kind of scary. And of course, this doesn’t include the operational risk of bringing on that 36-well megacube, and our NPVs are half-cycle, unburdened with G&A or other corporate expense. So, tread lightly when pushing the envelope….!
Conclusions
- Machine learning and optimization software provide a powerful tool to research and analyze potential oil and gas cube development designs.
- Spacing, completions, and geology all impact each other, and any development analysis must consider all of them in concert.
- The western part of Conoco’s acreage better supports multizone developments.
- Optimal 3-zone development density on Conoco’s Mabee acreage depends on oil price, completions cost, and operator risk tolerance.
Want to hear more? Check out the replay of a webinar we recently hosted on Oil and Gas Cube Development Strategy with Machine Learning.