/** Google Analytics start **/ /** Google Analytics end **/

The Three Ways AI Will Impact Oil and Gas

A journalist contacted me this week to ask my perspectives on how artificial intelligence would impact oil and gas. Here’s broadly what I sent back.

Defining Artificial Intelligence

A discussion of artificial intelligence must begin with a definition of AI. My definition is that AI is a computerized capability to execute work requiring cognitive skills that are normally associated with humans. Examples include natural language processing, translation of languages, visual perception, auditory interpretation, and tool creation. Basically, AI’s superpower is the ability to ingest massive amounts of data in whatever form or source it originates, interpret that data, and take some action. Other senses, such as touch and taste, and interpretation of things like emotional state, are also being worked on by the AI industry. The common ingredient is data, mountains and mountains of it.

What we declare to be AI will be a work in progress for some time.  But trend lines suggest that we will continue to advance computing to process those other cognitive skills and we will continue to be surprised at the level of cognitive processing in species other than humans.

Meanwhile, AI is already having three pretty dramatic impacts on oil and gas.   

  • First, in the upstream, AI is being applied to subsurface data and, through better interpretation of that enormous trove of data, petroleum engineers are confidently expanding the amount of oil and gas that is recoverable. 
  • Second, AI is helping reduce demand for fossil fuels by transforming transportation markets. 
  • Finally, AI is reducing costs and improving productivity throughout production, midstream and downstream. 

When managers combine AI with other digital technologies, opportunities for the industry are even greater, and I can even see the potential for new business models to emerge. 

Upstream Impacts

The biggest game for AI in the upstream is through interpreting and analyzing available subsurface data, modeling that data, and improving recovery rates from resources. Shale wells enter into swift decline quite quickly in the life of the well (perhaps 3 years after initial production). Conventional wells generally have a much shallower decline curve, and the amount of recovered product is much greater. 

By shifting the decline curve for shale to more closely match that of conventional wells (basically, lengthen the curve, and increase the volume of output), petroleum engineers will add 5% to global reserves (or 500b barrels of oil equivalent). This is primarily a math problem to understand and model out permeability and porosity, and to use that insight to frack shale oil and gas wells differently, a job for which artificial intelligence is well suited.  

The prize is substantial. About 90% of the resource in a conventional gas wells is recovered, about 40% in a typical conventional oil, but for shale, recoveries are 20% or so. A 5% increase in total reserves is ~$20T in value. No small beans.

Shale is primarily a North American phenomenon so this wealth will accrue principally to the US, who have the shale resource, a hotly competitive and well developed  services industry, cloud computing companies, data scientists, and so forth. Many countries have exceptional and similar shale resources, but lag  the US in all these additional attributes. 

In a competitive world, this incremental resource, if its cost is lower, will displace other higher cost resources from the market. And of course, if supply expands, then price will decline, all things being equal.

AI needs a lot of data to work properly, and there’s lots of data accumulated over the years from conventional oil and gas production.  That data could be fed into AI engines to enhance existing production or to extend the life of existing wells. Older wells, with low levels of productivity, could be brought back to life via AI. There are now data service companies popping up that offer this as a service. 

Downstream Impacts

The second big impact from AI is looming, but not yet manifest. As transportation technologies evolve, the demand for fuel may be dramatically impacted. I say “may” because no one can predict with certainty what will happen as society adopts more connected, autonomous, shared and electric cars and trucks (CASE for short). Car and truck manufacturers around the world are all rapidly converting their manufacturing supply lines to produce these new transportation technologies. CASE vehicles, particularly the autonomous varieties, are dependent on AI engines to interpret all that mobility data generated by on board cameras and sensors. 

Until transportation shifts to electric, demand for fuel could go up if everyone abandons buses in favour of Uber, but for services companies in logistics and freight transportation, carbon is a big concern. Companies in these sectors will be highly motivated to remove carbon from their business models (driven by tax policies now in force in Canada, or by outright bans that are coming to big cities in Europe), in which case demand should decline. So far, companies who have implemented AI-enabled trucks and haulers report fuel savings as the trucks run optimally all the time. 

I raise this issue of demand destruction because the industry doesn’t often recognize that AI is both expanding and destroying the business at the same time.

The issue for oil and gas is what to do with all their infrastructure dedicated to the challenge of distributing petroleum products in a market facing decline. Some oil companies now acknowledge that their retail businesses are effectively obsolete as a business model.  

Production and Midstream Impacts

In the rest of the oil and gas value chain, AI is being deployed primarily to augment human decision making, rather than to displace humans, and in ways that help optimize asset execution, and predict asset performance. There are some counter examples, but that’s how I see it. 

For example, Woodside uses IBM Watson (a suite of tools that comprise IBM’s AI offering), in a number of areas (there are several YouTube videos on their progress). My favourite is how they use Watson to work with the engineering team to catalogue all the previous engineering studies and documents about their enormous gas project off the coast of Western Australia. The engineers can then ask Watson in natural language anything they want, Watson instantly and correctly interprets the question, and then presents the findings. Woodside estimated that their engineers used to spend up to 40% of their time just finding previous studies. 

Here in Calgary, Plains Midstream also uses IBM Watson to optimize their midstream business (ie, raise productivity of assets and reduce costs).

CRUX OCM , an AI start up, uses a 10 minute look ahead prediction based on analytics to implement automated pipeline commands on behalf of overworked control room operators to increase the efficiency of pipeline assets. Stream Systems uses AI to model complex network assets like pipelines and tank farms in the cloud (creating a digital twin of the asset), to help owners optimize the network. NAL uses AI to interpret land contacts to pull out the contractual terms that are the basis of royalty calculations. A large oil sands company uses AI to help it understand its oil reservoirs and plan out better extraction techniques related to steam injection and the interface between different producing well. I see these examples (except NAL, see below), as first order uses – deploying AI on its own to help with complex analytic tasks and to support humans in that regard. 


As I’ve sketched out, AI by itself delivers clear benefits (ie, helping interpret massive data sets), but in combination with other technologies, AI unlocks much more value. For example, a pump armed with sensors generates lots of data, but with AI, that pump could interpret the data and take action, independently of a human operator. This is really important. Air conditioners respond to heat measurement, but not price signals. But an AI engine added to an aircon unit could switch on and off based on temperature and the price of power, or to switch between power suppliers.  

The NAL case is particularly illustrative. An AI engine reads the complex JV contracts, pulls out the terms that defines the royalty calculation, feeds the results into a robotic process automation tool (RPA) that converts the terms into code, and that code becomes a smart contract on block chain. The best human performance at this task (read the contract, calculate the royalty, produce the payment), is about 800 seconds per royalty contract, end to end, a pace that is largely unsustainable. AI, RPA and blockchain does this in seconds, fault free. Together, they reduce accounts payable costs, eliminate disputes between royalty partners, and reduce the stress  of some 50 specialised resources.

It’s estimated that the world-wide installed base of sensors today is 8b units, rising to 20 by 2020 or so, in the rush to add internet-enabled capability to things (the so-called internet of things revolution). These devices will generate so much data that the only way to handle and interpret it all is via AI. There should be significant demand for AI, data science and other related skills across many industries, not just oil and gas. 

New Business Models


Most worryingly for oil industry executives, AI is unlocking new business models which could be very disruptive. The sheer analytic horsepower from cloud computing now rivals the best in-house compute data centers in the biggest oil and gas companies, and is available to anyone on a variety of economic models. Cloud computing can be rented and AI algorithms are available on a per use basis, whereas in-house data centers are largely fixed cost assets. Smaller firms may be able to create the massive kinds of data sets that benefit from AI by aggregating data from multiple companies using cloud computing. Sophisticated artificial intelligence interpretation capabilities, which would be otherwise inaccessible, are now in reach.

AI is one of the hottest digital technologies, and will be hot for many years to come.

* corrected the description of CRUX OCM and how it works.

Share this ... Facebooktwittergoogle_plusredditlinkedinmail

Leave a Reply

%d bloggers like this: