Extracting oil and gas is rife with complexity and the potential for mishaps, not unlike driving on a Los Angeles freeway during rush hour. Now a solution is emerging to the enormous driving toll that is mainly caused by operator error—nearly 33,000 deaths and another 2.3 million injuries in the U.S. each year. A type of car in California and Texas has traveled 1 million miles without a death or injury, all without a human driver. The newest model doesn’t even have a steering wheel and brakes!

Now consider the economics tormenting the U.S. oil and gas industry. In the extraction of shale oil—which has made the U.S. the world’s largest oil producer—90% of the oil that could be recovered gets left behind. That inefficiency, brought on by the limits of what humans can accomplish using today’s technology, begs for a similar revolutionary response to the one Google found for cars.

Enter the world of artificial intelligence.

Just as Google vehicles are designed to utilize AI to make cars safer and more efficient, so can the oil and gas industry tap into AI to blunt the inefficiencies that have been its Achilles’ heel.

Oil companies compete based on their ability to identify, produce and replace reserves, with markets rewarding efficiency, speed, consistency and safety. Today’s volatile pricing environment ensures that only the most efficient companies will survive.

Over the last century, the industry overcame many technical challenges. To find and develop more oil and gas, the industry leveraged seismic data, electronic logging, horizontal drilling and hydraulic fracturing. Those are just a few of the innovations that changed the face of the industry forever.

That’s not enough, however. Demand for oil and gas is expected to rise in the coming decades. ExxonMobil’s recently published “The Outlook for Energy: A View to 2040” anticipates a 25% increase in global demand over the next 25 years. Consequently, oil output will need to rise to 112 million barrels per day (MMbbl/d), up from about 92 MMbbl now. The International Energy Agency estimates that closing this gap will require an annual investment of $750 billion in exploration and production.

As output from conventional fields declines over time, the majority of that growth will need to come from technology-driven supplies—shale oil, NGLs, oil sands and deepwater production. How can the oil and gas industry rethink what it does to better exploit conventional and unconventional energy sources?

Prescriptive analytics can improve all parts of the upstream value chain, from acquisition of acreage to production.

What would Google do?

Google wrestled with a similar concern when it reconsidered the way cars are driven. Instead of seeing a car as the sum of its mechanical parts, Google envisioned a software-driven device, powered by AI.

Google arrived at the autonomous vehicle solution by framing the challenge as a computer science problem rather than an engineering problem.

Designed as a fully integrated AI system, the Google vehicle uses sensors, data and algorithms to see, hear, read, understand, decide and act—just as human drivers do. The autonomous car must be able to navigate complex scenarios and obstacles, including unpredictable pedestrians and cyclists, inclement weather and, of course, other drivers.

AI and related disciplines strive to make machines do what comes naturally to humans (e.g., recognize someone is smiling) and combine them with what computers do easily (e.g., multiply a nine-digit number with an 11-digit number and divide the result by a 13-digit number). Today, AI is often used as an umbrella phrase to describe several scientific disciplines that make technology feats such as an autonomous car possible.

Unstructured datasets—images, videos, sounds and text—are growing at a much faster pace than structured datasets (i.e., numbers). According to IBM Research, there will be eight times more unstructured data than structured data in 2020. AI technologies shown above are, and will continue to be, the ones making sense of—and enabling decisions from—these hybrid datasets.

Take shale oil as an example. Like the distracted drivers Google’s autonomous cars must share the road with, unconventional reservoirs aren’t always well-behaved. Shale plays are notoriously complex and inconsistent, with a large number of interrelated, widely variable engineering parameters and geologic properties. Overwhelmed operators in search of better performance often resort to trial and error, with mixed results.

To take out some of the guesswork, upstream operators can turn to a Google-style AI technology that the industry is already starting to use. It’s called prescriptive analytics, and it relies on software to analyze data, in all of its various forms, to help craft a custom solution for each well.

Like Google’s self-driving car, prescriptive analytics software incorporates all available data, regardless of source, structure, size or format, and continually analyzes that information. That means videos, images, sounds, text and numbers can be examined together to prescribe specific actions to help operators drill more commercially viable wells.

Just as algorithms govern the behavior and performance of self-driving cars so that they can learn and adapt, the more data that prescriptive analytics software takes in, the smarter it gets.

Prescriptive analytics improves all parts of the upstream value chain, from acquisition of acreage to production.

The more data that prescriptive analytics software takes in, the smarter it gets.

Close enough isn’t good enough

Drilling a horizontal well into shale rock is a lot like drilling a hole in your wall to hang a picture, except that the wall is three miles underground and two miles away from where you’re standing. Where should you drill? Exactly how deep? What sort of rock will you encounter along the way, and how should you adjust to ensure maximum efficiency and safety?

With drilling costs often more than $3 million per shale well, one would hope that the answers to these questions would not surprise anyone. Unfortunately, they often do.

It’s not uncommon to see designs calling for drillers to stay within a 100-foot horizontal window. The idea is that precision is less important than speed and that a big fracture will capture all that’s recoverable. Sound familiar?

Let’s see what prescriptive analytics found. Its combined analysis of 3-D seismic data, well data, geological and geophysical data and production history produced a specific recommendation about exactly where to place the drill for each well. Slight variations in placement may drive as much as a 20% difference in production after 180 days. Close enough isn’t good enough.

Completions: a fuller picture

Now let’s get the oil. The many variables that influence production are often unclear or unknown, making it difficult to confidently determine the recipe for the most productive well. Because each well is unique, what works best for one well might not work best for another.

Prescriptive analytics makes sense of all of these seemingly disparate pieces of information to recommend a specific completion recipe for a well.

Consider the works of the French neo-impressionist artist Georges Seurat, who pioneered a painting technique called Divisionism. Rather than physically mix pigments, Seurat painted with tiny dots of individual colors.

Seurat then relied on the eyes of the viewer—or more specifically, the viewer’s brain—to mix the colors and transform millions of individual dots into a recognizable image.

Imagine each unique color that Seurat used as a variable that may influence a well’s productivity. These variables may be controllable, such as the well’s cluster density, or uncontrollable, such as geological attributes.

Like a Seurat masterpiece, prescriptive analytics connects the dots, or various data points, to create a richer, more vivid picture. It uses an adaptive combination of AI techniques to predict what a well will produce and to prescribe the optimal combination of controllable variables to achieve the desired performance in terms of production and/or economics. The software arranges all available data in terms of spatial (x, y, z) and temporal (t) coordinates and elicits stable patterns that are actionable. Prescriptive analytics-generated completion recipes have already been shown to improve best-practice performance by more than 10% after the first 180 days.

Another advantage of prescriptive analytics is its ability to measure known phenomena that current practices can’t. These factors may profoundly impact the economics of a well, a pad or a section of acreage. Using data that most operators routinely collect, the software creates new variables that identify and quantify these phenomena and then factors them into completion designs. We call these creations “synthetic variables.”

Examples of synthetic variables include those that:

• Account for resource competition among nearby wells;

• measure the impact on production from wells sharing a pad, stemming from the timing and sequencing of completion and flowback activities;

• capture and quantify the effects of completing infill wells in the vicinity of producing wells; and

• distinguish the impact of economic-based policy decisions from engineering decisions.

Synthetic variables are enabling software to reflect reality in ways never before possible, as if Seurat were given an entirely new palette of colors to work with. Gone are the days of trial and error at several million dollars a pop.

Production: maximizing potential

Prescriptive analytics also answers questions that are relevant during the well’s production phase.

For example, what type of artificial lift is most effective for a specific set of reservoir conditions? If there is an artificial lift, when could it fail and is there a way to extend its lifetime? Is it possible to improve return on investment using enhanced oil recovery techniques? Where is the point of diminishing returns? What makes a well a favorable candidate for refracking?

When the goal is to maximize an asset’s economic value at every point in its serviceable lifetime, answering these questions correctly is paramount.

Prescriptive analytics makes sense of seemingly disparate pieces of information to recommend a specific completion recipe for a well.

A&D: knowledge is power

Whenever there is a downturn in commodity prices, assets change hands. These A&D transactions often determine winners and losers for decades to come.

Uncertainty in assessing the value of oil and gas properties has always been the challenge. If the value assessment is too high and you bit, you paid too much. If the value assessment is too low and you passed, you missed out on the deal. What can you do to ensure you’re on the right side of transactions?

Many variables affect value. Some are difficult to measure, and others carry a high level of uncertainty. Data often come in multiple formats, and the time to prepare a bid can be short. Because data play such a key role in estimating value, the company that can process that information quickly and accurately holds the advantage.

The essence of deal-making is knowing more than your competitors do and having the confidence to act on that knowledge.

Prescriptive analytics can forecast the production for a property with unprecedented accuracy. It can also provide detailed, actionable prescriptions on how to achieve production targets while staying within financial and operational guidelines. It brings hard science to an A&D space that has historically been predicated on art, faith and hope.

Think differently

Instead of engineering a better, safer car, Google took a step back and looked at the source of what causes most crashes: human drivers. The oil and gas industry can look beyond making incremental changes by tapping into the same AI technologies.

Three forces have converged over the past decade to make things possible that even a few years ago were in the realm of science fiction: Data volume, variety and velocity keep increasing, storage keeps getting cheaper and computing power keeps improving.

Advances in AI capitalize on this convergence to generate insights, predictions and pre­scriptions, using all available data, with accuracy and speed that haven’t been possible to date. Every industry, from automotive to defense to agriculture, could benefit.

How can oil and gas—an $8 trillion global industry that makes some of the most complex decisions conceivable—sit on the sidelines? It can’t.

Daniel Mohan is the senior vice president of business development of Ayata, and Atanu Basu is the CEO and president. The company’s customers include several Fortune 500 integrated oil companies and E&P companies. The company invented prescriptive analytics technology. Ayata wishes to acknowledge Gabe Santos, Homestead Capital managing partner, for the “How Would Google Drill?” concept.

Comments